TerraMosaic Daily Digest: Feb 11, 2026
Daily Summary
Across 198 selected papers, hazard forecasting is being recalibrated around uncertainty-aware evidence. A 6000-year paleoseismic record supports Poisson-like behavior for major events (M >= 6.5), while complementary studies show that inequality metrics in seismic release and high-precision aftershock relocation still expose meaningful structure in rupture preparation and fault geometry.
Operationally, the strongest advances come from robust, transferable workflows: automated lahar detection from seismic signatures, orbit-specific tropospheric correction for Sentinel-1 deformation analysis, and 3D InSAR inversion for groundwater-driven subsidence. At the decision level, risk optimization is widening its objective set by treating equity (not only efficiency) as a formal design target in multi-hazard retrofit planning.
Key Trends
- Seismic hazard baselines are shifting from periodic recurrence to stochastic framing: Long records indicate Poisson-like timing for major earthquakes, while event inequality and dense relocation analyses refine short-term interpretation without reverting to rigid cycles.
- Monitoring pipelines are becoming automated and deployment-ready: Lahar classifiers, orbit-aware InSAR error handling, and 3D deformation inversion show a move from bespoke studies to transferable, near-operational workflows.
- Compound hydro-coastal risk is treated as a 3D and transition problem: New work quantifies three-dimensional storm-erosion amplification and tests how well models capture drought-to-flood transitions, exposing limits of conventional 2D assumptions.
- Infrastructure design is coupling mechanics with lifecycle resilience: Tunnel joints, dampers, scour foundations, and permafrost transport sections are evaluated under realistic loading paths and degradation trajectories.
- Risk governance is becoming explicitly distribution-aware: Multi-objective retrofit planning now places equity metrics (e.g., Theil index) alongside loss, downtime, and displacement to make tradeoffs transparent.
Selected Papers
This digest features 198 selected papers from 1024 papers analyzed (out of 2649 raw papers scanned; 1024 new papers after deduplication). Each paper has been evaluated for its relevance to landslide and broader geohazard research and includes links to the original publications.
1. Large earthquakes follow highly unequal ones
Core Problem: Understanding the dynamics of large earthquake events and identifying reliable indicators for their imminent occurrence, based on the theory of self-organized criticality in tectonic plates, remains a challenge for hazard estimation.
Key Innovation: The study shows, through numerical simulations and seismic data analysis, that large earthquake events tend to follow periods of highly unequal seismic activity (quantified by Gini and Kolkata indices), suggesting these inequality indices can serve as useful indicators for hazard estimates and proximity to criticality.
2. Occurrence of major earthquakes is as stochastic as smaller ones
Core Problem: Seismic hazard estimates often rely on cyclic or quasiperiodic recurrence models for major earthquakes, which may be inaccurate and lead to underestimation of risk.
Key Innovation: Demonstrated, using a 6000-year lake-sediment record and statistical analyses, that major earthquake (M ≥ 6.5) interevent times robustly follow a Poisson distribution, implying stochasticity rather than periodicity, thereby challenging existing recurrence models and increasing seismic hazard estimates.
3. A decision-support framework for fair building retrofit allocation under multi-hazard risk
Core Problem: Hazard mitigation planning often overlooks distributional effects, leading to unequal outcomes for vulnerable groups in building retrofit allocation under multi-hazard risk.
Key Innovation: Developed a multi-objective optimization framework that treats equity (measured by the Theil index) as a first-class objective alongside efficiency (economic loss, repair time, displacement) for building retrofit allocation under coupled earthquake-tsunami hazards, providing transparent, spatially explicit portfolios for stakeholder deliberation.
4. Seismic Characterization of Lahars on Volcán de Fuego Toward the Development of a Machine Learning‐Based Detection Algorithm
Core Problem: The challenge of early detection and automated alerts for frequent lahars on Volcán de Fuego, which often rely on manual monitoring and sparse visual confirmation.
Key Innovation: Characterized varied short-term lahar seismic behavior (increasing activity, shift to lower frequencies downstream) and implemented computationally efficient K-nearest neighbor (KNN) based detectors using seismic signal attributes, achieving high accuracy for moderate-to-large flows without additional instrumentation.
5. High‐Precision Aftershock Distribution Highlights the Complex Fault Geometry of the 2024 Mw 7.5 Noto Peninsula Earthquake
Core Problem: Understanding the complex rupture behavior and underlying fault geometry of the 2024 Mw 7.5 Noto Peninsula earthquake, particularly near its hypocenter.
Key Innovation: Deploying 30 temporary seismic stations and using machine learning to detect and precisely locate 46,252 aftershocks, revealing several fault planes and suggesting complex rupture episodes from successive ruptures on adjacent fault planes.
6. R2RAG-Flood: A reasoning-reinforced training-free retrieval augmentation generation framework for flood damage nowcasting
Core Problem: Rapid and accurate post-storm property damage nowcasting is crucial, but existing methods may lack interpretability or require extensive task-specific training, especially for large language models.
Key Innovation: Developed R2RAG-Flood, a reasoning-reinforced, training-free retrieval-augmented generation framework that leverages a knowledge base of reasoning trajectories to enable LLMs to emulate and adapt prior reasoning for flood damage nowcasting, achieving competitive accuracy and high efficiency without fine-tuning.
7. Hyperspectral Smoke Segmentation via Mixture of Prototypes
Core Problem: Traditional visible-light methods for smoke segmentation in wildfire management face limitations due to insufficient spectral information, struggling with cloud interference and semi-transparent smoke regions, and different spectral bands exhibit varying discriminative capabilities.
Key Innovation: Introduces hyperspectral imaging for smoke segmentation, presents the first hyperspectral smoke segmentation dataset (HSSDataset), and proposes a Mixture of Prototypes (MoP) network with band split, prototype-based spectral representation, and a dual-level router for adaptive spatial-aware band weighting, demonstrating superior performance.
8. Experimental study of clear-water local scour around a large-diameter monopile in silty sand
Core Problem: Conventional predictors for clear-water local scour around large-diameter monopiles are inadequate for silty sand environments, failing to account for distinct scour behaviors such as prolonged equilibrium times, coarse-grained armor layer formation, and shallower equilibrium depths, which are influenced by high pile-to-sediment size ratios and non-uniform silty sand.
Key Innovation: Conducts an experimental study of clear-water scour around a large-diameter monopile in silty sand, systematically examining the effects of approach velocity and flow shallowness, and revealing the inadequacy of conventional predictors. It highlights the critical need to incorporate partial-depth boundary-layer development, pile-to-sediment size ratio, and armoring effects for accurate scour design in estuarine and nearshore foundations.
9. Three-dimensional amplification of storm-driven beach erosion: Implications for setback lines at Narrabeen-Collaroy Beach, Australia
Core Problem: Existing approaches for estimating short-term coastal erosion (storm demand) often rely on 2D beach profiles, potentially underestimating erosion due to localized three-dimensional (3D) effects, leading to inaccurate setback lines.
Key Innovation: An empirical analysis combining nearly five decades of 2D beach profile data with targeted 3D storm response observations to quantify and account for 3D amplification of storm erosion, leading to improved erosion estimates and the development of empirical design curves for enhanced setback line planning.
10. Unveiling the drivers of rainfall-triggered landslide spatial distribution: Insights from event-based and historical landslide inventories in Lisbon region, Portugal
Core Problem: Understanding the factors influencing rainfall-triggered landslide spatial distribution and evaluating the impact of using different types of inventories (event-based vs. historical) for landslide susceptibility assessment remains a challenge.
Key Innovation: Demonstrated that predisposing factors are key for landslide-prone regions, highlighted the individual strengths and spatial differences of event-based and historical inventories due to localized rainfall, and emphasized the importance of integrating both datasets for comprehensive analysis.
11. Orbit-specific tropospheric effects on Sentinel-1A/B interferometric synthetic aperture radar observations: insights for deformation analysis and future mission design
Core Problem: Tropospheric delays in InSAR observations complicate accurate interpretation of surface deformation, and orbit-specific tropospheric features remain underexplored.
Key Innovation: Investigation and quantification of spatiotemporal characteristics of orbit-specific tropospheric effects on Sentinel-1A/B InSAR observations using nine years of global data, providing insights and recommendations for improving deformation inversion accuracy and future SAR mission design.
12. Spatiotemporal prediction and mapping of fractures from successive tunnel faces: an integrated SMF and ST-LSTM framework
Core Problem: Accurately predicting the spatiotemporal distribution of fractures ahead of a tunnel face is crucial for construction safety and mitigating geological risks, but existing methods struggle with capturing long-term development and short-term dynamic disturbances.
Key Innovation: Proposed an integrated spatiotemporal sequence-based deep learning framework (enhanced PredRNN with zigzag-shaped SMF and dual-channel ST-LSTM) for predicting fracture maps ahead of tunnel faces, demonstrating superior stability and accuracy in forecasting fracture distribution.
13. Experimental investigation of frictional healing in granite: temperature-mineral composition interplay
Core Problem: Understanding the influence of temperature and mineral composition on frictional healing of rock fractures is crucial for assessing the long-term stability of geological repositories and predicting unstable slip in fractured rock masses.
Key Innovation: Experimental investigation using Slide-Hold-Slide (SHS) direct shear experiments on different granites at varying temperatures, revealing lithology-dependent thermo-mechanical-chemical effects on frictional healing rates and providing mechanistic insights into the complex interplay of these processes governing fractured rock stability and potential for unstable slip.
14. Forecasting step-like reservoir landslide using a physical-mechanical-numerical framework
Core Problem: The complex creep behavior and dynamic evolution of step-like reservoir landslides pose significant challenges for accurate and interpretable displacement forecasting.
Key Innovation: Proposed a novel Physical-Mechanical-Numerical (PMN) framework that integrates a Seepage-mechanical-deformation (SMD) block model with Bayesian updating and MCMC methods for improved interpretability and accurate long-term forecasting of step-like reservoir landslide displacements.
15. Toward Systematic Modeling of Volcano Deformation Sources Using Automatically‐Generated InSAR Products
Core Problem: The need for a robust (semi-)automated approach to systematically pre-process and model volcano deformation signals from routinely acquired InSAR data globally, to effectively catalog, model, and compare deformation.
Key Innovation: Developed GBIS-BULK, a framework combining filtering, clustering, ICA noise reduction, and Otsu thresholding to semi-automatically locate and delimit volcano deformation signals in InSAR data. Validated the approach using synthetic interferograms and Sentinel-1 data from the East African Rift System, showing consistency with previous bespoke modeling studies.
16. Why Firn Quakes
Core Problem: Firn quakes, audible propagating collapse events in Antarctic and Arctic snowfields, are poorly understood, lacking a stable theory for their conditioning, triggering, and propagation.
Key Innovation: A theory combining granular and continuum mechanics is proposed, suggesting that unconsolidated firn at depth, supported by solid-like internal structures formed by pressure sintering, is a central condition. Dynamic amplification triggers brittle failure of these supports, leading to a cascade of collapse propagation. The flexural wavespeed matches recorded firn quake velocities (~100 m/s), providing mechanical and structural reasoning for these loud collapse events and a framework for similar mechanics in compacting granular systems with hazardous consequences like landslides or avalanches.
17. Retrieving Three‐Dimensional Deformation in Groundwater Pumping Areas Based on InSAR Data
Core Problem: Monitoring three-dimensional (3D) surface deformation in groundwater pumping areas remains a significant challenge due to conventional InSAR's limited sensitivity to north-south displacement.
Key Innovation: Developing an inversion approach, the VerticalGradient-Constrained Strain Model (VG-SM3D), that combines multi-track InSAR data with a strain model and a physical prior to successfully reconstruct 3D deformation fields, validated in the Tianjin region.
18. Investigation on the inclination angle of undrained shear slip surface in saturated soils based on mixture theory
Core Problem: Understanding the precise inclination angle of undrained shear slip surfaces in saturated soils, considering hydro-mechanical coupling and permeability variations, is crucial for predicting soil instability.
Key Innovation: The paper derives the energy conservation equation and reveals the effective stress principle's mechanical mechanism based on mixture theory and non-equilibrium thermodynamics. It provides specific recommendations for uncoupled versus fully coupled hydro-mechanical analysis based on permeability extremes and explores the slip surface inclination angle using Mohr-Coulomb theory.
19. OrbitChain: Orchestrating In-orbit Real-time Analytics of Earth Observation Data
Core Problem: Current Earth observation analytics suffer from significant delays (hours to days) in data download and analysis due to limited ground connections, diminishing the value of data for time-sensitive applications like disaster monitoring or search-and-rescue.
Key Innovation: Proposes OrbitChain, an in-orbit multi-satellite Earth analytics framework that uses a pipelined design to decompose workflows and orchestrate constellation-wide resources, enabling real-time analytics (results in minutes) for Earth sensing applications and advanced workflows like in-orbit tip-and-cue.
20. Oblique wave interaction with (M+1) U-shaped thick porous structures interconnected with (M) thin porous boxes in the presence of dual trenches near a partially reflected wall
Core Problem: There is a need for effective multifunctional coastal systems to mitigate wave impacts, reduce wave run-up, and manage hydrodynamic forces for applications like harbor tranquility, sediment transport, and breakwaters.
Key Innovation: Presents a comprehensive investigation of a hybrid coastal structure (interconnected U-shaped thick porous structures and thin porous boxes with dual trenches near a partially reflected wall) under oblique wave incidence, revealing that increasing interconnected structures, specific radii, thick porosity, and U-shaped structures significantly attenuate wave reflection, run-up, forces, and moments, offering valuable guidance for coastal engineering.
21. How well do hydrological models simulate streamflow extremes and drought-to-flood transitions?
Core Problem: It is unclear how well hydrological models capture compound extreme events like drought-to-flood transitions and which modeling decisions are most important for performance, despite their utility in understanding underlying processes and identifying prone regions.
Key Innovation: Evaluated four conceptual hydrological models with different calibration strategies across 63 catchments, demonstrating that model performance (KGE) does not guarantee good detection of streamflow extremes and transitions, and highlighting the importance of streamflow timing, model structure, catchment characteristics, and meteorological forcings for improved simulation of drought-to-flood transitions.
22. Bridging disaster resilience gaps: exploring adaptation preferences and time commitment willingness in isolated island communities
Core Problem: Strengthening disaster resilience in geographically isolated communities requires understanding how residents prioritize and engage in adaptation activities under conditions of limited institutional support.
Key Innovation: Explores adaptation preferences and time commitment willingness in isolated island communities to bridge disaster resilience gaps, providing insights into community engagement for disaster preparedness.
23. Calculation of safe roof thickness for irregular goaf underlying a subgrade based on the backward elimination method
Core Problem: Accurately determining the safe roof thickness for irregular goafs is critical for ensuring the stability of overlying infrastructure like expressway subgrades, but existing methods may not adequately account for irregular geometries and multiple influencing factors.
Key Innovation: Established a multi-factor calculation model for goaf safe roof thickness using the backward elimination method, validated by 3D numerical simulations and field monitoring, providing a novel and reliable approach for engineering design.
24. Experimental apparatus and mechanical properties of shield tunnel segment joint under cyclic bending moment
Core Problem: Lack of understanding and experimental devices to investigate the mechanical behavior of shield tunnel segment joints under cyclic (seismic) bending moments.
Key Innovation: Development of a novel cyclic loading device capable of applying both positive and negative bending moments to segment joints, and its use to characterize the non-ductile failure modes, hysteretic behavior, stiffness degradation, and energy dissipation capacity of joints under cyclic loading, providing a basis for seismic design.
25. Experimental and numerical research on the tensile performance of a novel three-layer ring spring flexible joint for shield tunnels
Core Problem: Shield tunnel joints are prone to tensile failure under longitudinal ground deformation in geologically active zones, and conventional flexible joints may suffer from waterproofing failure due to excessive deformation.
Key Innovation: Proposal and validation of a novel three-layer ring spring (TRS) flexible joint that reduces joint stiffness, provides energy-dissipation and self-centering characteristics, and incorporates a displacement-limiting mechanism to prevent waterproofing failure, thereby enhancing tunnel protection against tensile damage.
26. Behaviour of tuned hysteretic inerter viscous damper assisted structure subjected to soil dependent stochastic seismic excitation and recorded ground motions
Core Problem: Enhancing the seismic performance of structures, particularly under soil-dependent stochastic seismic excitation, by developing more effective passive control strategies than existing devices.
Key Innovation: Proposed a novel Tuned Hysteretic Inerter Viscous Damper (THIVD) configuration, demonstrating superior performance in reducing structural displacement and acceleration responses (53-62% and 37-45% respectively) compared to uncontrolled structures and linear TVMDs, under various soil conditions and ground motions.
27. Distribution and evolution of distress in embankment-bridge transition sections of the Gonghe–Yushu Expressway in degrading permafrost regions
Core Problem: Systematic investigations into the distribution characteristics and evolutionary mechanisms of distress in embankment-bridge transition sections (EBTS) on permafrost expressways, particularly under climate warming and permafrost degradation, are limited.
Key Innovation: Conducted the first systematic field survey and analysis of EBTS distresses on a permafrost expressway, classifying five primary types of distress, elucidating their distribution and progressive evolutionary mechanisms, and proposing a novel EBTS structure for mitigation.
28. Effects of structural plane on shearing behavior of anchored rocks under true triaxial disturbance: Morphology, roughness, and lithology
Core Problem: The safety and stability of deep underground engineering structures are critically dependent on rock mass structural planes, but experimental studies on anchored structural planes under true 3D stress conditions, considering varying foliation morphologies, roughness, and lithologies, are limited, leading to an incomplete understanding of their mechanical behaviors and failure mechanisms.
Key Innovation: Employed a newly developed true triaxial dynamic–static combined shear testing system to systematically examine the effects of foliation morphology, initial roughness, and wall plate lithology on the mechanical behavior and fracture evolution of anchored structural planes in various rocks, revealing how these factors influence disturbance shear strength, deformation, and microscopic fracturing mechanisms.
29. Pleistocene Smoothing and Resurfacing of Appalachian Ridgelines by Periglaciation
Core Problem: Isolating the geomorphic impact of Pleistocene periglacial conditions on mid-latitude landscapes from modern climate, tectonics, and rock strength, and understanding how these conditions shaped Appalachian ridgelines.
Key Innovation: Demonstrating that hilltop curvature and hillslope length in the Appalachians vary with paleotemperature (periglacial conditions), not modern climate or uplift, and showing that frost cracking and solifluction enhanced hilltop lowering and valley infilling, leaving a long-lived geomorphic signature of permafrost-driven processes.
30. Tide‐Modulated Ocean‐to‐Earth Energy Conversion Quantified With Coastal Fiber Sensing
Core Problem: Quantifying the time-resolved, in-situ transfer of energy from incident ocean waves into seismic surface waves along the nearshore, which is essential for understanding coastal hazards and Earth-ocean coupling.
Key Innovation: Using a beach-deployed distributed acoustic sensing (DAS) array co-located with an ocean-bottom node to directly measure and quantify the energy-conversion efficiency from wave impacts to Rayleigh-type ground motion (on the order of 10^-6), demonstrating its strong modulation by tide, and establishing a framework for coastal monitoring using existing fiber infrastructure.
31. Physics‐Informed, Differentiable Hydrologic Models for Capturing Unseen Extreme Events
Core Problem: Uncertainty in how well hybrid machine learning/process-based hydrologic models generalize to extreme floods outside of training data, and whether optimizing for extreme events compromises spatial generalizability or physical significance.
Key Innovation: Evaluating differentiable hydrologic models (δHBV) against LSTM for predicting unseen extreme events, demonstrating that δHBV models outperform LSTM for events with longer return periods due to their mass balance and first-order exchange terms, improving reliability for stakeholder preparedness.
32. FGAA-FPN: Foreground-Guided Angle-Aware Feature Pyramid Network for Oriented Object Detection
Core Problem: Oriented object detection in high-resolution remote sensing and aerial imagery is challenging due to cluttered backgrounds, severe scale variation, and large orientation changes, with existing approaches often lacking explicit foreground modeling and geometric orientation priors.
Key Innovation: Proposes FGAA-FPN, a Foreground-Guided Angle-Aware Feature Pyramid Network that uses a Foreground-Guided Feature Modulation module to enhance object regions and suppress background interference, and an Angle-Aware Multi-Head Attention module to encode relative orientation relationships, achieving state-of-the-art results for oriented object detection in disaster response and geographic information updating.
33. GMG: A Video Prediction Method Based on Global Focus and Motion Guided
Core Problem: Accurately predicting weather is challenging due to rapid meteorological data variability, teleconnections, and the inability of current spatiotemporal models to capture global features or adapt to non-rigid body deformations (e.g., cloud growth/dissipation).
Key Innovation: Proposes the GMG model, a video prediction method that addresses challenges in weather forecasting by incorporating a Global Focus Module to enhance the global receptive field and a Motion Guided Module to adapt to the growth or dissipation processes of non-rigid bodies, demonstrating improved predictive accuracy for complex spatiotemporal data.
34. Spatiotemporal Field Generation Based on Hybrid Mamba-Transformer with Physics-informed Fine-tuning
Core Problem: Data-driven models for spatiotemporal physical field generation often suffer from substantial discrepancies with underlying physical equations.
Key Innovation: Development of HMT-PF, a hybrid Mamba-Transformer model with a physics-informed fine-tuning block that uses a point query mechanism and self-supervised learning to effectively reduce physical equation discrepancies while generating spatiotemporal fields.
35. Noise is All You Need: rethinking the value of noise on seismic denoising via diffusion models
Core Problem: Conventional seismic denoising methods depend on synthetic datasets or clean signal labels, which are often unavailable or inaccurate for field data, limiting their effectiveness in challenging field environments.
Key Innovation: Proposes SeisDiff-denoNIA, a diffusion-based framework that trains directly on extracted field noise, explicitly learning the true noise distribution. This eliminates reliance on synthetic data and significantly improves denoising performance on field DAS-VSP surveys, enhancing seismic event continuity and reflection visibility.
36. A UAV-Based VNIR Hyperspectral Benchmark Dataset for Landmine and UXO Detection
Core Problem: A critical gap exists in open-access UAV-based hyperspectral data specifically designed for landmine and unexploded ordnance (UXO) detection research, hindering reproducible and comparative studies.
Key Innovation: Introducing a novel, open-access UAV-based VNIR hyperspectral benchmark dataset for landmine and UXO detection, collected over a controlled test field with realistic targets, complete with raw data, calibration details, and reference spectra, enabling reproducible research and multi-sensor benchmarking.
37. THE INFLUENCE OF STRUCTURE PERMEABILITY ON WAVE OVERTOPPING AT ROCK ARMOURED SLOPES ALONG EUROTOP
Core Problem: The EurOtop Manual's roughness factor (γf) for structure permeability in rock-armoured slopes, particularly for impermeable cores, has not been consistently validated, potentially leading to inaccuracies in wave overtopping predictions.
Key Innovation: A comprehensive study with 591 new small-scale physical model tests investigating the influence of structural permeability on wave overtopping, leading to a revision of the roughness factor for impermeable cores and improved wave steepness influence, significantly enhancing the prediction accuracy of the EurOtop Manual formulation.
38. AGILE v0.1: The Open Global Glacier Data Assimilation Framework
Core Problem: Integrating heterogeneous glacier observations into dynamically consistent models and reducing uncertainties in current glacier volume and area estimates remains a challenge for global glacier modeling.
Key Innovation: Developed AGILE v0.1, an open global glacier data assimilation framework using a time-dependent variational method and PyTorch for full differentiability, efficiently optimizing glacier bed topography and ice volume reconstruction through transient calibration.
39. AI image-based method for a robust automatic real-time water level monitoring: a long-term application case
Core Problem: Traditional water level monitoring methods often lack robustness, real-time capabilities, and cost-effectiveness for long-term application in remote or ungauged areas, especially under adverse weather and lighting conditions.
Key Innovation: A robust, AI-driven, image-based camera gauge system for real-time, long-term water level monitoring was developed, demonstrating high accuracy (MAE 0.96-2.66 cm) and resilience to environmental conditions, enabling 24/7 operation for flood detection and mitigation.
40. Community Flood Resilience Factors; A Community’s Perspective
Core Problem: Existing community flood resilience models and factors often overlook the experiences and 'lay knowledge' of at-risk community members, leading to an incomplete understanding and measurement of true community resilience.
Key Innovation: Designed and distributed a community flood resilience survey to incorporate lay knowledge, highlighting the importance of community members in framework design and providing a community-specific flood resilience framework with both established and novel factors for stakeholders and community use.
41. An Integrated Assessment of Heat Hazard, Vulnerability, and Accessibility to Climate Shelter Networks for Identifying Urban Adaptation Priority Areas in Andalusia (Spain)
Core Problem: The deployment of urban climate shelters as an adaptation measure often lacks systematic consideration of spatial patterns of heat hazard and social vulnerability, leading to mismatches between need and provision.
Key Innovation: Developed an integrated framework combining hazard and vulnerability indices with network-based accessibility analysis to identify urban adaptation priority areas for climate shelter deployment, providing a transferable, equity-oriented methodology for disaster risk reduction and climate adaptation.
42. Resilience assessment of interdependent urban infrastructure systems under intensifying typhoon scenarios: Comparing recovery strategies for coupled networks
Core Problem: There is a lack of comprehensive understanding and systematic comparison of recovery strategies for interdependent urban infrastructure systems (traffic and power) under intensifying typhoon scenarios, particularly regarding cascading failure mechanisms and their propagation dynamics.
Key Innovation: Developed a comprehensive simulation framework incorporating Monte Carlo analysis to evaluate system performance under intensifying typhoons, explicitly modeling functional, spatial, and restoration interdependencies, and comparing four distinct recovery strategies to optimize infrastructure investments and integrated recovery.
43. Influence of subgrade frost heave on ballasted track dynamics via a hybrid DEM-vehicle model
Core Problem: Subgrade frost heave disrupts the stability and structural integrity of ballasted railway tracks, leading to unstable dynamic responses and posing significant threats to operational safety, especially under high-speed conditions.
Key Innovation: Developed a hybrid DEM-vehicle model to non-iteratively couple the ballasted track-subgrade structure with vehicle dynamics, systematically analyzing the microscopic mechanical evolution and dynamic response characteristics of the system under frost heave.
44. Time-dependent mechanical response of buried water pipelines to cold waves under coupled multiphysics loading
Core Problem: The mechanical response and failure risk of buried urban water pipelines under time-varying multiphysics coupling, especially during extreme cold waves, are poorly understood, hindering resilience and safety enhancements.
Key Innovation: Developed and validated a 3D Finite Element model to simulate the transient, non-linear mechanical response of buried pipelines to coupled multiphysics loading (earth pressure, traffic, groundwater, cold waves), identifying key influencing factors and localized synergistic amplification of stress.
45. Adaptive infrastructure intelligence: integrated urban stormwater management framework for climate-resilient cities
Core Problem: The need for an integrated urban stormwater management framework to create climate-resilient cities, addressing challenges like peak flow, flood occurrence, and pollutant removal.
Key Innovation: Proposing the Adaptive Infrastructure Intelligence (AII) framework, a socio-technical system integrating urban ecology and adaptive governance, which uses multimodal sensing networks to predictively optimize stormwater system performance, achieving significant reductions in peak flow (67%), flood occurrence (83%), and pollutant removal (72%).
46. Integrating distributed hydrologic simulation with low-flow resilience: a spatiotemporal perspective
Core Problem: Limited understanding of prolonged low-flow conditions and their spatiotemporal dynamics, hindering effective drought preparedness and water resource management in poorly gauged basins.
Key Innovation: Leveraging a fully distributed watershed hydrologic model to investigate low-flow resilience, delineating and characterizing low-flow recessions spatiotemporally, and identifying a 3-day rainfall deficit as a critical window for intervention, providing a transferable framework for drought preparedness and coordinated water management.
47. Modeling non-Newtonian fluid–solid flows containing non-spherical particles by the SPH-DEM coupling model
Core Problem: Accurately simulating the complex interaction dynamics of non-Newtonian fluids and non-spherical particles in fluid-solid systems, which is crucial for geological studies and engineering applications like debris flow modeling.
Key Innovation: Developed a resolved SPH-DEM coupling model for non-Newtonian fluid-non-spherical particle interactions, incorporating super-ellipsoid and polyhedral particle models, a modified boundary repulsive force, and an improved method for generating body-fitted boundary dummy particles, validated through complex dam-break scenarios.
48. Hydro-mechanical responses of irregular twin tunnels with unequal burial depths in anisotropic soil layer
Core Problem: Existing studies on tunnel behavior often simplify geometries and soil conditions, failing to adequately address multi-tunnel interference, unequal burial depths, and noncircular sections in anisotropic soils, which are common in practical engineering and critical for safety.
Key Innovation: Develops a hybrid Physics-Informed Neural Network (PINN)-Finite Element Method (FEM) framework to efficiently and accurately analyze steady-state seepage and stress fields for complex irregular twin tunnels with unequal burial depths in anisotropic soil layers, combining PINN's flexibility with FEM's robustness.
49. Multiphysics modeling of thermo-hydraulic fracturing during CO2 sequestration in multilayered reservoirs at Ordos, China
Core Problem: Unintended thermo-hydraulic fracturing and significant permeability changes in multilayered geological CO2 sequestration reservoirs, leading to reduced storage capacity and potential CO2 leakage risks.
Key Innovation: Development of a numerical model incorporating 21 aquifers and aquitards with multiphase and THM coupling to accurately simulate and explain the observed permeability variations and thermal-hydraulic fracturing phenomena during CO2 injection, revealing the underlying mechanisms.
50. Monitoring the In Situ Nonlinear Elasticity Near the Dalk Glacier Area, Antarctica, Using Dense Seismic Arrays
Core Problem: Monitoring disturbances in the subsurface medium of Antarctic glaciers and their connection to environmental changes (e.g., sea level rise due to melting) to detect low-amplitude precursors to systemic imbalances.
Key Innovation: Utilized ambient noise data from dense seismic arrays and codawave interferometry to measure in situ nonlinear elasticity (dv/v) near Dalk Glacier, observing delayed dv/v responses to environmental forcing, semi-diurnal variations controlled by tidal strain, humidity, and ice melt dynamics, and demonstrating a cost-effective, high-resolution monitoring technique.
51. Antarctic Meltwater Accelerates Southern Ocean Evolution Under Projected Atmospheric Warming
Core Problem: Previous studies on Antarctic ice shelf meltwater impact on the Southern Ocean did not self-consistently evolve melt rates and ocean states, leading to biases in projected ocean-melt-driven ice loss.
Key Innovation: Using an Earth-system model with interactive ice-shelf basal melting, the study finds that increasing meltwater accelerates the evolution of continental shelf warming/cooling patterns. The net feedback is negative over the 21st century in a high-emissions scenario, leading to a ~35% reduction in ice-shelf meltwater input, which significantly reduces projected melt rates.
52. Quantifying Changes in Water Loading in the U.S. Southwest via Comparison of GNSS, GRACE, and SWE Data Sets
Core Problem: Limited GNSS station numbers and GRACE resolution make hydrological partitioning difficult to unravel, especially in the topographically variable Colorado River Basin, hindering understanding of water loading impacts on crustal deformation.
Key Innovation: Comparing GNSS vertical displacement, GRACE surface mass change, and Snow Water Equivalent (SWE) data using elastic surface displacement modeling and signal localization to quantify region-dependent seasonal partitioning of water loading and its effect on crustal movement.
53. RESIdual STability (RESIST) Calibration for Improved Hydrological Model Time Generalizability
Core Problem: Hydrological models calibrated on specific periods often show significant accuracy decrease when extrapolated to periods with different climate conditions, limiting their temporal generalizability.
Key Innovation: Developing a novel calibration objective (RESIdual STability - RESIST) that jointly calibrates model accuracy and time-invariance of residuals, demonstrating improved temporal generalizability of hydrological models for climate-impact studies.
54. Comparison of generative algorithms for conceptual groundwater modeling of coastal volcanic aquifer features with disparate, sparse and extremely imbalanced data
Core Problem: The accuracy of freshwater-seawater exchange simulations in coastal volcanic aquifers is limited by traditional conceptual groundwater models (CGMs) that struggle to integrate disparate, sparse, and imbalanced hydrogeophysical features.
Key Innovation: Proposed an AI-assisted workflow using SOM and generative algorithms to construct stochastic CGMs, successfully representing complex aquifer features and improving simulations of groundwater flow and freshwater-seawater exchange, including seawater intrusion.
55. Signature-Kernel Based Evaluation Metrics for Robust Probabilistic and Tail-Event Forecasting
Core Problem: Current probabilistic forecasting evaluation frameworks lack a consensus metric, assume independence across time steps/variables, and are insensitive to critical tail events, hindering the development of robust forecasting algorithms.
Key Innovation: Proposal of two kernel-based metrics, Sig-MMD and censored Sig-MMD (CSig-MMD), which leverage the signature kernel to capture complex inter-variate and inter-temporal dependencies, remain robust to missing data, and prioritize tail-event prediction while maintaining properness.
56. A Multimodal Conditional Mixture Model with Distribution-Level Physics Priors
Core Problem: Learning the full conditional distribution of admissible outcomes in scientific and engineering systems with intrinsic multimodal behavior, in a physically consistent and interpretable manner, especially when integrating with physics constraints or limited data.
Key Innovation: Develops a physics-informed multimodal conditional modeling framework based on mixture density networks (MDNs), embedding physical knowledge through component-specific regularization terms to accommodate non-uniqueness and stochasticity efficiently.
57. Towards Remote Sensing Change Detection with Neural Memory
Core Problem: Current remote sensing change detection methods struggle to capture long-range dependencies efficiently and fail to capture intricate spatiotemporal relationships, leading to scalability challenges and suboptimal accuracy.
Key Innovation: Proposes ChangeTitans, a Titans-based framework for remote sensing change detection, featuring VTitans (a vision backbone with neural memory and segmented local attention) and TS-CBAM (a two-stream fusion module leveraging cross-temporal attention) to capture long-range dependencies and enhance accuracy while remaining computationally competitive.
58. 1%>100%: High-Efficiency Visual Adapter with Complex Linear Projection Optimization
Core Problem: Deploying vision foundation models typically relies on efficient adaptation strategies, but conventional full fine-tuning suffers from prohibitive costs and low efficiency, and delta-tuning advantages from LLMs do not directly transfer to vision tasks.
Key Innovation: Proposes CoLin, a novel low-rank complex adapter that introduces only about 1% parameters to the backbone, and addresses convergence issues of low-rank composite matrices with a tailored loss, achieving state-of-the-art performance and efficiency across various vision tasks, including remote sensing scenarios.
59. Enhancing Weakly Supervised Multimodal Video Anomaly Detection through Text Guidance
Core Problem: In weakly supervised multimodal video anomaly detection, the potential of the text modality is under-explored due to challenges in extracting anomaly-specific text features (general language models lack nuance, scarcity of descriptions) and issues with redundancy and imbalance in multimodal fusion.
Key Innovation: A novel text-guided framework that includes an in-context learning-based multi-stage text augmentation mechanism to generate high-quality anomaly text samples for fine-tuning text feature extractors, and a multi-scale bottleneck Transformer fusion module using compressed bottleneck tokens to progressively integrate information across modalities, achieving state-of-the-art performance on UCF-Crime and XD-Violence datasets.
60. Field-Deployable Hybrid Gravimetry: Projecting Absolute Accuracy Across a Remote 24km$^2$ Survey via Daily Quantum Calibration
Core Problem: Absolute gravimeters are accurate but bulky, while relative gravimeters are mobile but suffer from drift, limiting high-precision, large-area geophysical surveys in challenging environments.
Key Innovation: Demonstrates a hybrid quantum-enabled gravimetry approach where an on-site atomic gravimeter provides routine, high-precision calibration for mobile spring gravimeters, effectively suppressing instrumental drift and enabling high-fidelity regional gravity gradient capture in remote, dense tropical terrain.
61. TRACE: Theoretical Risk Attribution under Covariate-shift Effects
Core Problem: When a model is deployed or updated with shifted data, its performance on the original domain can change unpredictably, and there's a need for an interpretable framework to understand and diagnose these risk changes.
Key Innovation: Introduces TRACE (Theoretical Risk Attribution under Covariate-shift Effects), a framework that decomposes the absolute risk change of a model under covariate shift into four actionable factors (two generalization gaps, a model change penalty, and a covariate shift penalty), providing a computable diagnostic tool and a deployment gate score for safe, label-efficient model replacement.
62. On the Role of Consistency Between Physics and Data in Physics-Informed Neural Networks
Core Problem: The implications of data-to-PDE inconsistencies (due to noise, errors, assumptions) on the accuracy and convergence of Physics-Informed Neural Networks (PINNs) are insufficiently understood, limiting their reliability as surrogate models.
Key Innovation: Systematically analyzes how data inconsistency fundamentally limits PINN accuracy, introducing the concept of a 'consistency barrier' as an intrinsic lower bound on error, demonstrating that PINN accuracy saturates at an error level dictated by data inconsistency, and providing guidance for constructing and interpreting physics-informed surrogate models.
63. (MGS)$^2$-Net: Unifying Micro-Geometric Scale and Macro-Geometric Structure for Cross-View Geo-Localization
Core Problem: Cross-view geo-localization (CVGL) for GNSS-denied UAV navigation is brittle due to drastic geometric misalignment between oblique aerial views and orthographic satellite references, particularly concerning view-dependent vertical facades and scale variations.
Key Innovation: Proposes (MGS)$^2$-Net, a geometry-grounded framework that unifies micro-geometric scale and macro-geometric structure through a Macro-Geometric Structure Filtering (MGSF) module to filter facade artifacts, a Micro-Geometric Scale Adaptation (MGSA) module for dynamic scale rectification, and a Geometric-Appearance Contrastive Distillation (GACD) loss, achieving state-of-the-art performance in CVGL.
64. Spectral-Spatial Contrastive Learning Framework for Regression on Hyperspectral Data
Core Problem: While contrastive learning has succeeded in classification, there is a shortage of studies applying it to regression tasks, particularly for hyperspectral data, limiting its potential for enhanced representation learning in this domain.
Key Innovation: The authors propose a model-agnostic spectral-spatial contrastive learning framework for regression tasks on hyperspectral data, along with a collection of relevant augmentation transformations, which significantly improves the performance of various backbone models.
65. Enhancing Multivariate Time Series Forecasting with Global Temporal Retrieval
Core Problem: Multivariate time series forecasting (MTSF) models are constrained by their reliance on limited historical context, preventing them from effectively capturing crucial global periodic patterns that span significantly longer cycles.
Key Innovation: Introduces the Global Temporal Retriever (GTR), a lightweight, plug-and-play module that extends any forecasting model's temporal awareness by maintaining an adaptive global temporal embedding, dynamically retrieving and aligning relevant global segments, and jointly modeling local and global dependencies to achieve state-of-the-art performance with minimal overhead.
66. Towards Learning a Generalizable 3D Scene Representation from 2D Observations
Core Problem: Prior 3D scene representation methods often operate in camera-centric coordinates, limiting their direct applicability to robotic manipulation and generalizability to unseen object arrangements without scene-specific finetuning.
Key Innovation: Introduces a Generalizable Neural Radiance Field approach that constructs occupancy representations in a global workspace frame from egocentric robot observations, allowing generalization to unseen object arrangements and achieving accurate 3D reconstruction (26mm error) including occluded regions, beyond traditional stereo vision.
67. TabICLv2: A better, faster, scalable, and open tabular foundation model
Core Problem: Existing tabular foundation models, while powerful, can be limited in terms of performance, speed, and scalability, especially for larger datasets, and often require extensive hyperparameter tuning.
Key Innovation: TabICLv2, a new state-of-the-art tabular foundation model that surpasses previous benchmarks through a novel synthetic data generation engine, architectural innovations (e.g., scalable softmax attention), and optimized pretraining protocols, offering improved performance, speed, and scalability for regression and classification tasks without extensive tuning.
68. Generalized Robust Adaptive-Bandwidth Multi-View Manifold Learning in High Dimensions with Noise
Core Problem: Existing multiview data fusion methods offer limited theoretical guarantees, especially when dealing with heterogeneous and high-dimensional noise across different data sources, hindering robust recovery of shared intrinsic structures.
Key Innovation: Proposing GRAB-MDM, a kernel-based diffusion geometry framework that integrates multiple noisy data sources using a novel view-dependent bandwidth selection strategy, which adapts to each view's geometry and noise, leading to provably robust recovery of shared intrinsic structure and improved embedding quality.
69. Statistical Inference and Learning for Shapley Additive Explanations (SHAP)
Core Problem: The lack of statistical inference approaches for global measures of feature importance derived from SHAP values (e.g., mean absolute or mean squared SHAP), despite their ubiquity in improving model explainability and performing feature selection.
Key Innovation: Developing a semi-parametric approach for calibrating confidence in estimates of the pth powers of SHAP, constructing asymptotically normal de-biased estimators (U-statistics with Neyman orthogonal scores) for p ≥ 2 and smoothed alternatives for 1 ≤ p < 2, thereby enabling reliable statistical inference for SHAP-based feature importance.
70. Highly Adaptive Principal Component Regression
Core Problem: The Highly Adaptive Lasso (HAL) achieves good convergence rates but can be computationally prohibitive in high dimensions, and while HAR is a scalable alternative, further improvements in efficiency are desirable.
Key Innovation: Introducing Principal Component based Highly Adaptive Lasso (PCHAL) and Principal Component based Highly Adaptive Ridge (PCHAR) estimators, which offer substantial gains in computational efficiency through outcome-blind dimension reduction while matching the empirical performance of HAL and HAR.
71. A Unified Experimental Architecture for Informative Path Planning: from Simulation to Deployment with GuadalPlanner
Core Problem: The evaluation and transferability of informative path planning algorithms for autonomous vehicles are often hindered by fragmented execution pipelines and limited ability to move consistently between simulation and real-world deployment.
Key Innovation: Introducing GuadalPlanner, a unified experimental architecture that decouples high-level decision-making from vehicle-specific control, enabling consistent evaluation and deployment of path planning algorithms across fully simulated environments, software-in-the-loop, and physical autonomous vehicles (validated with water quality monitoring using an autonomous surface vehicle).
72. Symmetrization Weighted Binary Cross-Entropy: Modeling Perceptual Asymmetry for Human-Consistent Neural Edge Detection
Core Problem: Deep neural networks for edge detection often achieve high numerical accuracy but fail to produce visually sharp and perceptually consistent edges, limiting their reliability in intelligent vision systems.
Key Innovation: Introduces the Symmetrization Weighted Binary Cross-Entropy (SWBCE) loss, which extends conventional WBCE by incorporating prediction-guided symmetry to model perceptual asymmetry in human edge recognition. This aligns optimization with human perception, enhancing edge recall and suppressing false positives, leading to superior quantitative accuracy and perceptual fidelity in edge detection.
73. OmniDiff: A Comprehensive Benchmark for Fine-grained Image Difference Captioning
Core Problem: Existing Image Difference Captioning (IDC) datasets lack breadth and depth, limiting the development of models that can precisely localize visual changes and generate coherent, fine-grained descriptions in complex, dynamic environments.
Key Innovation: Introduces OmniDiff, a comprehensive dataset for fine-grained image difference captioning with diverse scenarios and detailed human annotations, and proposes M$^3$Diff, a MultiModal LLM enhanced with a Multi-scale Differential Perception (MDP) module, achieving state-of-the-art performance in identifying and describing inter-image differences across various benchmarks.
74. Spectrum from Defocus: Fast Spectral Imaging with Chromatic Focal Stack
Core Problem: Hyperspectral cameras face harsh trade-offs between spatial, spectral, and temporal resolution, especially in low-photon conditions, and existing computational imaging systems often require complex optics or extensive computation.
Key Innovation: Introduces Spectrum from Defocus (SfD), a chromatic focal sweep method that achieves state-of-the-art hyperspectral imaging using simple optics (two off-the-shelf lenses, grayscale sensor) and a fast physics-based iterative algorithm, delivering sharp, accurate, photon-efficient, and interpretable hyperspectral images.
75. Geospatial Representation Learning: A Survey from Deep Learning to The LLM Era
Core Problem: The need to effectively transform location-centric geospatial data into meaningful computational representations for spatial analysis and decision-making, especially with the evolution of deep learning and large language models.
Key Innovation: A comprehensive survey and structured taxonomy of Geospatial Representation Learning (GRL) across deep learning and LLM eras, highlighting advancements, limitations, and future directions for extracting latent structures and semantic patterns from geographic data.
76. From Pixels to Images: A Structural Survey of Deep Learning Paradigms in Remote Sensing Image Semantic Segmentation
Core Problem: Traditional methods struggle with efficiency and accuracy in semantic segmentation of diverse and voluminous Remote Sensing Images (RSIs), which is critical for fine-grained interpretation of surface features.
Key Innovation: A comprehensive, structured survey of deep learning-based semantic segmentation in remote sensing images, organized into a pixel-patch-tile-image hierarchy, providing a holistic understanding of the field's evolution, representative datasets, and open challenges.
77. GeoPurify: A Data-Efficient Geometric Distillation Framework for Open-Vocabulary 3D Segmentation
Core Problem: Transferring 2D Vision-Language Model (VLM) features to 3D semantic segmentation results in noisy/fragmented predictions, while enforcing geometric coherence requires costly training and large 3D annotated datasets, creating a persistent trade-off.
Key Innovation: Proposing GeoPurify, a data-efficient geometric distillation framework that uses a small Student Affinity Network to purify 2D VLM-generated 3D point features with geometric priors from a 3D self-supervised teacher, and a Geometry-Guided Pooling module for inference-time denoising, achieving state-of-the-art performance with significantly less training data.
78. GeoZero: Incentivizing Reasoning from Scratch on Geospatial Scenes
Core Problem: Enhancing reasoning capabilities of remote sensing MLLMs typically relies on expensive, human-biased chain-of-thought (CoT) data, which limits diversity and scalability of model reasoning.
Key Innovation: GeoZero, a framework enabling MLLMs to perform geospatial reasoning without predefined CoT supervision, uses two datasets (GeoZero-Instruct, GeoZero-Hard) and Answer-Anchored Group Relative Policy Optimization (A^2GRPO) to foster diverse and accurate emergent reasoning, surpassing SOTA on remote sensing benchmarks.
79. Localized Graph-Based Neural Dynamics Models for Terrain Manipulation
Core Problem: Predictive models for robot terrain manipulation face challenges with high-dimensional terrain state representations, especially for fine-resolution details and unknown depth, and the computational cost of modeling large terrain graphs when only localized portions are active.
Key Innovation: A learning-based approach for terrain dynamics modeling that leverages Graph-based Neural Dynamics (GBND) by identifying a small active subgraph within a large terrain graph, combined with a learning-based region of interest (RoI) identification and a novel domain boundary feature encoding, achieving faster and more accurate prediction of terrain deformation.
80. LatticeVision: Image to Image Networks for Modeling Non-Stationary Spatial Data
Core Problem: Parameter inference for parametric statistical models, particularly spatially autoregressive (SAR) models applied to large, non-stationary spatially distributed random variables ('fields'), is computationally prohibitive using traditional maximum likelihood estimation (MLE).
Key Innovation: LatticeVision, an approach that leverages image-to-image (I2I) networks by viewing both spatial fields (inputs) and spatially arranged model parameters (outputs) as images, enabling faster and more accurate parameter estimation for a class of non-stationary spatially autoregressive (SAR) models with unprecedented complexity.
81. Long-Term Mapping of the Douro River Plume with Multi-Agent Reinforcement Learning
Core Problem: The challenge of long-term (multiple days), energy- and communication-efficient mapping of dynamic river plumes using multiple autonomous underwater vehicles (AUVs).
Key Innovation: An energy- and communication-efficient multi-agent reinforcement learning approach that integrates spatiotemporal Gaussian process regression (GPR) with a multi-head Q-network controller to regulate AUV direction and speed, demonstrating superior accuracy and operational endurance for long-term plume mapping.
82. Quantifying pore-scale heterogeneity of drying and wetting soil water retention behavior using X-ray computed tomography
Core Problem: Lack of effective methods for determining the local degree of saturation and matric suction and explaining hydraulic hysteresis in individual pores with sufficiently high spatial resolution in unsaturated soils.
Key Innovation: Proposed a novel workflow using X-ray CT images to quantify local Sr and ψm at individual pores and explore hysteretic water retention behavior, revealing how pore geometry, size, and connectivity govern drainage and imbibition, and approximating full-range pore-scale WRC from a single CT scan.
83. Storage effects on tube specimens of a highly plastic marine clay
Core Problem: Evaluating the influence of storage time on the compressibility and undrained shear strength behavior of highly plastic marine clay tube specimens, which is critical for reliable laboratory testing and characterization of natural soft soil deposits.
Key Innovation: Experimental investigation showing that for highly plastic soils of low/moderate sensitivity, the disturbance induced by the sampler is the main factor controlling measured soil response, with storage time having minor influence on mechanical properties when fixed-piston samplers are used correctly, allowing confidence in long-term stored specimens.
84. Numerical analysis of post-cyclic ultimate bearing capacity of helical anchors in clay
Core Problem: Accurately assessing the post-cyclic ultimate bearing capacity of helical anchors in clay, which is significantly impacted by cyclic loading and crucial for the design of floating offshore structures.
Key Innovation: Developed a 3D finite element model incorporating the E-R model to investigate the effects of cyclic loading patterns and soil parameters on bearing capacity, exploring mechanisms of soil strength degradation and failure, and developing a predictive expression for post-cyclic ultimate bearing capacity.
85. Reply to Discussion by A. Khayyer and C.H. Lee on “Comparative study on volume conservation among various SPH models for flows of different levels of violence, coastal engineering, volume 191, August 2024, 104521”
Core Problem: Addressing specific issues raised in a discussion regarding the volume conservation and mechanical energy conservation of various SPH models when simulating violent fluid flows, particularly in coastal engineering applications like wave shoaling and breaking.
Key Innovation: A further investigation providing new insights into volume error indicators, volume and mechanical energy conservation of different SPH models, their performance during wave shoaling, breaking, and violent sloshing, and the effects of boundary conditions on volume conservation, in response to a critical discussion.
86. The effect of digitalization on disaster response in local government: Quasi-experimental evidence from smart city infrastructure in China
Core Problem: The causal effect of digitalization, specifically smart city (SC) infrastructure, on enhancing local governments' disaster response capacity and shortening response times, particularly in the context of increasing extreme natural disasters, remains underexplored with robust evidence.
Key Innovation: Provided quasi-experimental evidence using a multi-period difference-in-differences approach, demonstrating that smart city infrastructure shortens disaster response times by an average of 12.5% through improved information transmission efficiency, optimized resource allocation, and enhanced interdepartmental collaboration.
87. Optimizing post-disaster road restoration with reinforcement learning: A traveler-behavior-aware approach
Core Problem: Optimizing short-term post-disaster road network restoration is complex due to the need to maximize traffic acceleration while accounting for traveler behavior, gradual adaptation to network changes, limited resources, and uncertainties in recovery times.
Key Innovation: Developed the Traveler-Adaptive Restoration Mechanism (TARM), an AI-driven method integrating Reinforcement Learning, Markov Decision Process, and optimization-based day-to-day traffic simulation, to optimize post-disaster road restoration by considering traveler behavior and demonstrating that simply shortening restoration periods may not always enhance traffic efficiency.
88. Vulnerability assessment of storage tanks subject to fires under multi-source uncertainties using a novel ensemble-based non-parametric approach
Core Problem: Vulnerability assessment of storage tanks subject to fires encounters various multi-source uncertainties, which could affect the reliability and credibility of the results.
Key Innovation: Proposes a novel ensemble-based non-parametric approach using tolerance intervals to quantify multi-source uncertainties in vulnerability assessment of storage tanks subject to fires, providing a robust uncertainty analysis tool as a supplement to crisp values.
89. Intelligent and accurate recognition method of lining cracks under complex seam interference and engineering application
Core Problem: Accurate detection of micro-scale lining cracks in tunnels, especially distinguishing them from construction seams, is challenging and crucial for structural integrity.
Key Innovation: An enhanced U-Net with a multi-view fusion module and compound loss function for micro-crack detection, combined with a two-step skeleton processing strategy and Graph Convolutional Network (GCN) to classify and eliminate seam-like structures, significantly improving precision and recall.
90. Calculation and prediction of CO2 concentrations inside a ventilation gallery of Madrid Calle 30 urban tunnels
Core Problem: High CO2 concentrations in urban tunnels lead to concrete carbonation, significantly reducing infrastructure lifespan, and there's a need for reliable methods to calculate and predict these concentrations.
Key Innovation: A methodology to calculate and forecast CO2 concentrations inside urban tunnel ventilation galleries based on traffic intensity and circulating fleet evolution, providing a powerful tool for predicting concrete carbonation and informing design for extended infrastructure lifespan.
91. Simulation of brittle particle breakage using a spherical harmonic-based discrete element method
Core Problem: Conventional discrete element methods are limited in capturing irregular particle morphology and complex brittle breakage mechanisms, which significantly influence material strength and deformation in geotechnical engineering.
Key Innovation: Developed a 3D breakage model using SH-DEM with a dual-center definition for irregular particles, identifying fracture planes via Boussinesq–Cerruti solution and Dijkstra path search, demonstrating morphology-dependent brittle breakage mechanisms.
92. Strength prediction using binders’ water-absorption capacity and a saturation-based compaction quality control framework for treated clays
Core Problem: Conventional compaction control methods for treated clays in road embankments do not adequately capture water-binder interactions, leading to inconsistent and unpredictable field performance, and there's a need for better strength prediction models for these materials.
Key Innovation: Proposes a new strength prediction model integrating binder water absorption capacity (Wab) with mixture design parameters, and a saturation-based compaction quality control framework using degree of saturation (Sr) and air void volume (Va) to ensure stable post-compaction soil properties, offering improved consistency and predictability.
93. Quantifying and Regionalizing Land Use Impacts on Catchment Response Times With High‐Frequency Observations
Core Problem: Robustly quantifying and regionalizing the impact of land use on catchment hydrological response times at scale is difficult, hindering rainfall-runoff model development.
Key Innovation: Provided statistically significant evidence that intensive land use quickens hydrological response times in tropical Andean catchments, developing a robust methodology for regionalization using high-frequency observations and a linear mixed-effects model.
94. Multi-encoder ConvNeXt Network with Smooth Attentional Feature Fusion for Multispectral Semantic Segmentation
Core Problem: Improving the accuracy and efficiency of land cover semantic segmentation in multispectral imagery, especially by effectively fusing information from different spectral channels and scales, remains a challenge.
Key Innovation: The paper proposes MeCSAFNet, a multi-branch encoder-decoder architecture with dual ConvNeXt encoders, a dedicated fusion decoder with CBAM attention, and ASAU activation, demonstrating significant performance gains on FBP and Potsdam datasets for multispectral land cover segmentation.
95. Versor: A Geometric Sequence Architecture
Core Problem: Limitations of traditional sequence architectures in achieving structural generalization, interpretability, and efficiency, especially for tasks requiring native representation of geometric relationships and handling chaotic dynamics.
Key Innovation: Introduction of Versor, a geometric sequence architecture that uses Conformal Geometric Algebra (CGA) to natively represent SE(3)-equivariant relationships, leading to significant performance improvements, orders of magnitude fewer parameters, improved interpretability, and linear complexity, particularly validated on chaotic N-body dynamics.
96. Hardware Co-Design Scaling Laws via Roofline Modelling for On-Device LLMs
Core Problem: Selecting appropriate LLM backbones for resource-constrained on-device settings (e.g., autonomous vehicles) requires balancing accuracy with strict inference latency and hardware efficiency, making hardware-software co-design a critical but challenging requirement.
Key Innovation: Proposing a hardware co-design law that jointly captures model accuracy and inference performance by modeling training loss as a function of architectural hyperparameters and characterizing inference latency via roofline modeling, leading to a principled framework for optimizing on-device LLM deployment.
97. Colorful Talks with Graphs: Human-Interpretable Graph Encodings for Large Language Models
Core Problem: Graph problems are fundamentally challenging for LLMs due to their text-based representations, requiring reasoning over explicit structure, permutation invariance, and complex relationships, which current LLMs struggle with.
Key Innovation: Introduction of a human-interpretable structural encoding strategy for graph-to-text translation that injects graph structure into natural language prompts using Weisfeiler-Lehman similarity classes mapped to 'color tokens,' significantly enhancing LLM performance on graph tasks requiring global structure reasoning.
98. Experimental Demonstration of Online Learning-Based Concept Drift Adaptation for Failure Detection in Optical Networks
Core Problem: Conventional static models for failure detection in dynamic systems (like optical networks) struggle with concept drift, leading to degraded performance over time.
Key Innovation: Presentation of a novel online learning-based approach for concept drift adaptation in optical network failure detection, achieving up to a 70% performance improvement over static models while maintaining low latency.
99. A Dual-Stream Physics-Augmented Unsupervised Architecture for Runtime Embedded Vehicle Health Monitoring
Core Problem: Traditional vehicle health monitoring metrics are insufficient, and unsupervised deep learning models often conflate statistically stable high-load steady states (which cause significant fatigue) with mechanical rest, failing to capture critical mechanical burden.
Key Innovation: A Dual-Stream Architecture that fuses unsupervised learning for surface anomaly detection with macroscopic physics proxies for cumulative load estimation, generating a multi-dimensional health vector to distinguish dynamic hazards from sustained mechanical effort for comprehensive, edge-based vehicle health monitoring.
100. Learning Adaptive Distribution Alignment with Neural Characteristic Function for Graph Domain Adaptation
Core Problem: Graph Domain Adaptation (GDA) is challenged by complex, multi-faceted distributional shifts, and existing methods rely on inflexible, manually selected graph elements or heuristics for alignment.
Key Innovation: Proposes ADAlign, an Adaptive Distribution Alignment framework for GDA that automatically identifies and aligns the most relevant discrepancies in each transfer scenario using Neural Spectral Discrepancy (NSD) and a learnable frequency sampler, making it flexible, scenario-aware, and robust.
101. End-to-End LiDAR optimization for 3D point cloud registration
Core Problem: LiDAR sensors are typically designed independently of downstream tasks like point cloud registration, leading to suboptimal data collection, significant computational overhead for processing, and fixed configurations that limit performance.
Key Innovation: Proposes an adaptive LiDAR sensing framework that dynamically adjusts sensor parameters and jointly optimizes LiDAR acquisition and registration hyperparameters, integrating registration feedback into the sensing loop to optimally balance point density, noise, and sparsity, thereby improving registration accuracy and efficiency.
102. MapVerse: A Benchmark for Geospatial Question Answering on Diverse Real-World Maps
Core Problem: Current large language models (LLMs) and vision-language models (VLMs) struggle with integrating spatial relationships, visual cues, real-world context, and domain-specific expertise for reasoning over maps, and existing map-based reasoning benchmarks are narrow, domain-specific, and rely on artificial content.
Key Innovation: MapVerse, a large-scale benchmark with 11,837 human-authored question-answer pairs across 1,025 real-world maps from ten diverse categories, provides a rich setting to evaluate map reading, interpretation, and multimodal reasoning, revealing significant gaps in current VLM performance on complex spatial reasoning tasks.
103. LLM-Based Scientific Equation Discovery via Physics-Informed Token-Regularized Policy Optimization
Core Problem: Existing LLM-based symbolic regression frameworks treat LLMs as static generators, failing to update internal representations based on search feedback, often yielding physically inconsistent or mathematically redundant scientific equations.
Key Innovation: Proposes PiT-PO (Physics-informed Token-regularized Policy Optimization), a unified reinforcement learning framework that evolves LLMs into adaptive generators for scientific equation discovery, using a dual-constraint mechanism to enforce hierarchical physical validity and apply token-level penalties for structural parsimony, leading to state-of-the-art performance and discovery of novel turbulence models.
104. Eliminating VAE for Fast and High-Resolution Generative Detail Restoration
Core Problem: Diffusion models for super-resolution are slow and memory-intensive, especially for high-resolution images, due to the VAE bottleneck.
Key Innovation: Proposing GenDR-Pix, which eliminates the VAE using pixel-(un)shuffle operations and multi-stage adversarial distillation, achieving significant acceleration and memory saving for high-resolution image restoration, potentially useful for remote sensing imagery.
105. AurigaNet: A Real-Time Multi-Task Network for Enhanced Urban Driving Perception
Core Problem: Developing reliable and efficient AI systems for autonomous vehicles requires robust multi-task perception capabilities for object detection, lane detection, and drivable area segmentation.
Key Innovation: Introduction of AurigaNet, an advanced multi-task network integrating object detection, lane detection, and end-to-end instance segmentation for drivable areas, achieving state-of-the-art accuracy and real-time performance on embedded devices for urban driving perception, with potential adaptability for remote sensing applications.
106. Ecological mapping with geospatial foundation models
Core Problem: The utility of Geospatial Foundation Models (GFMs) for high-value ecological mapping use cases, such as LULC generation, forest functional trait mapping, and peatlands detection, has not been fully explored.
Key Innovation: This study demonstrates that fine-tuned GFMs (TerraMind, Prithvi) significantly outperform baseline ResNet models for various ecological mapping tasks, highlighting their potential and identifying considerations for input data divergence and label quality.
107. RSHallu: Dual-Mode Hallucination Evaluation for Remote-Sensing Multimodal Large Language Models with Domain-Tailored Mitigation
Core Problem: Multimodal large language models (MLLMs) applied in remote sensing suffer from hallucinations (responses inconsistent with input RS images), which severely hinder their deployment in high-stakes scenarios like emergency management, and these hallucinations remain under-explored in the RS domain.
Key Innovation: Presents RSHallu, a systematic study that formalizes RS hallucinations with a domain-oriented taxonomy, builds a dual-mode hallucination benchmark (RSHalluEval), and introduces domain-tailored mitigation strategies (dataset RSHalluShield, decoding-time logit correction, RS-aware prompting) to significantly improve the hallucination-free rate in RS-MLLMs.
108. FPGA Implementation of Sketched LiDAR for a 192 x 128 SPAD Image Sensor
Core Problem: Emerging high-spatial-resolution single-photon avalanche diode (SPAD) arrays in LiDAR systems face a critical challenge of massive data transfer bandwidth, reaching tens of gigabytes per second.
Key Innovation: Presents an efficient FPGA implementation of a polynomial spline function-based statistical compression algorithm for LiDAR data, achieving a 512x compression ratio, enabling histogram-free online depth reconstruction with high fidelity, and effectively alleviating the time-stamp transfer bottleneck of SPAD arrays.
109. Tuning the burn-in phase in training recurrent neural networks improves their performance
Core Problem: Training recurrent neural networks (RNNs) with standard backpropagation through time (BPTT) is challenging for long input sequences, and truncated BPTT, while practical, has performance implications related to the burn-in phase.
Key Innovation: Establishes theoretical bounds on accuracy and performance loss when optimizing over subsequences, revealing the importance of the burn-in phase. Demonstrates that proper tuning of the burn-in phase can significantly reduce prediction error in time series forecasting.
110. PuriLight: A Lightweight Shuffle and Purification Framework for Monocular Depth Estimation
Core Problem: Existing self-supervised monocular depth estimation methods are constrained by either bulky architectures that compromise practicality or lightweight models that sacrifice structural precision, creating a critical need for lightweight yet structurally precise architectures.
Key Innovation: PuriLight, a lightweight and efficient framework for self-supervised monocular depth estimation, addresses these limitations through a three-stage architecture incorporating novel Shuffle-Dilation Convolution (SDC) for local feature extraction, Rotation-Adaptive Kernel Attention (RAKA) for hierarchical feature enhancement, and Deep Frequency Signal Purification (DFSP) for global feature purification, achieving state-of-the-art performance with minimal parameters and exceptional computational efficiency.
111. Direct Learning of Calibration-Aware Uncertainty for Neural PDE Surrogates
Core Problem: Neural PDE surrogates often lack calibrated uncertainty, which is crucial for downstream decisions in data-limited or partially observed regimes, and existing uncertainty quantification methods are often post hoc or require specific noise tuning.
Key Innovation: Proposes 'cross-regularized uncertainty,' a framework that directly learns uncertainty parameters during training using a held-out regularization split, yielding regime-adaptive and better-calibrated predictive distributions for neural PDE surrogates, with uncertainty fields concentrating in high-error regions.
112. Statistical Learning Analysis of Physics-Informed Neural Networks
Core Problem: The training and performance of Physics-Informed Neural Networks (PINNs) for initial and boundary value problems lack a comprehensive statistical learning understanding, particularly regarding the role of the physics penalty and the quantification of predictive uncertainty.
Key Innovation: Reformulates PINN parameter estimation as a statistical learning problem, interpreting the physics penalty as an infinite source of indirect data, and analyzes the learning process as a singular learning problem using Local Learning Coefficient, providing insights into predictive uncertainty and extrapolation capacity.
113. From Circuits to Dynamics: Understanding and Stabilizing Failure in 3D Diffusion Transformers
Core Problem: 3D diffusion transformers for surface completion from sparse point clouds exhibit a catastrophic failure mode ("Meltdown") where small input perturbations can fracture the output into disconnected pieces.
Key Innovation: Localizes the "Meltdown" failure to a specific cross-attention activation, identifies its singular-value spectrum as a proxy for fragmentation, and introduces PowerRemap, a test-time control that effectively stabilizes sparse point-cloud conditioning, improving reliability for 3D reconstruction.
114. Physically Interpretable AlphaEarth Foundation Model Embeddings Enable LLM-Based Land Surface Intelligence
Core Problem: The physical interpretability of satellite foundation model embeddings is poorly understood, limiting their integration into environmental decision systems.
Key Innovation: Presents a comprehensive interpretability analysis of Google AlphaEarth's 64-dimensional embeddings, demonstrating that individual dimensions map to specific land surface properties, and develops an LLM-based Land Surface Intelligence system for natural language environmental queries.
115. Robust Semantic Transmission for Low-Altitude UAVs: Predictive Channel-Aware Scheduling and Generative Reconstruction
Core Problem: UAV downlink transmission for time-sensitive visual applications is constrained by bandwidth scarcity and dynamic channel impairments, leading to global reconstruction failure with conventional methods during deep fading.
Key Innovation: Developing a predictive transmission framework for UAVs that decouples semantic content into structural and texture components, using a split-stream variational codec and a channel-aware scheduler to prioritize structural layout, achieving significant SNR gains and maintaining fidelity under channel prediction mismatch.
116. Solving PDEs in One Shot via Fourier Features with Exact Analytical Derivatives
Core Problem: Recent random feature methods for solving PDEs still rely on iterative optimization or expensive derivative computation, and PINNs are slower and less accurate.
Key Innovation: Proposing FastLSQ, which combines frozen random Fourier features with analytical operator assembly to solve linear PDEs via a single least-squares call, and extending it to nonlinear PDEs via Newton-Raphson, achieving significantly higher accuracy and speed.
117. Are Dense Labels Always Necessary for 3D Object Detection from Point Cloud?
Core Problem: Current state-of-the-art 3D object detection methods from point clouds require a large amount of costly dense 3D bounding box annotations for training, which is a significant burden.
Key Innovation: Proposes SS3D++, a novel sparsely-annotated framework that requires annotating only one 3D object per scene, and progressively generates confident fully-annotated scenes using a missing-annotated instance mining module and a reliable background mining module, achieving competitive performance with significantly reduced annotation costs.
118. Measuring Orthogonality as the Blind-Spot of Uncertainty Disentanglement
Core Problem: Jointly estimating aleatoric and epistemic uncertainties is problematic, and current evaluation methods for their disentanglement are insufficient, often failing to ensure that each uncertainty estimate is not affected by the other (orthogonality).
Key Innovation: Proposes that aleatoric and epistemic uncertainty estimates should be orthogonally disentangled, proving that orthogonality and consistency are necessary and sufficient criteria. Introduces Uncertainty Disentanglement Error (UDE) as a metric to measure these criteria, demonstrating its utility in evaluating and optimizing disentanglement in models.
119. GenDR: Lighten Generative Detail Restoration
Core Problem: Applying text-to-image diffusion models to real-world super-resolution (SR) results in a suboptimal trade-off between inference speed and detail fidelity due to target misalignment, with existing models being either slow/inefficient or overqualified.
Key Innovation: Presents GenDR, a one-step diffusion model for generative detail restoration, distilled from a tailored diffusion model with a larger latent space (SD2.1-VAE16), and introduces consistent score identity distillation (CiD) and its adversarial extension (CiDA) to achieve state-of-the-art SR performance with improved speed and fidelity.
120. Enhancing Vehicle Detection under Adverse Weather Conditions with Contrastive Learning
Core Problem: Vehicle detection from UAV images in Nordic regions faces strong visibility challenges and domain shifts due to diverse snow coverage, and annotated data are expensive.
Key Innovation: Proposes a sideload-CL-adaptation framework that uses unannotated data to train a CNN-based representation extractor via contrastive learning in a pretraining stage, then sideloads it to a frozen YOLO11n backbone in fine-tuning, significantly improving vehicle detection performance under adverse weather.
121. Defect-aware Hybrid Prompt Optimization via Progressive Tuning for Zero-Shot Multi-type Anomaly Detection and Segmentation
Core Problem: Recent VLM-based anomaly detection models often neglect fine-grained defect types, limiting specific insight into anomalies and hindering targeted corrective measures, while handcrafted prompts are time-consuming and susceptible to human bias.
Key Innovation: DAPO, a novel Defect-aware Prompt Optimization approach, uses progressive tuning to learn hybrid defect-aware prompts for zero-shot multi-type anomaly detection and segmentation, significantly improving AUROC and precision on various benchmarks by aligning image features with fine-grained text semantics.
122. CoRe3D: Collaborative Reasoning as a Foundation for 3D Intelligence
Core Problem: Extending explicit reasoning mechanisms from language and vision tasks to 3D understanding and generation remains underdeveloped, limiting model reliability and interpretability in 3D.
Key Innovation: CoRe3D introduces a unified 3D understanding and generation reasoning framework that jointly operates over semantic and spatial abstractions, using a spatially grounded reasoning representation to decompose 3D latent space for compositional and procedural reasoning.
123. Efficient Learning on Large Graphs using a Densifying Regularity Lemma
Core Problem: Learning on large graphs with traditional Message Passing Neural Networks (MPNNs) is hindered by computational and memory costs that scale linearly with the number of edges, making them inefficient for massive datasets.
Key Innovation: The Intersecting Block Graph (IBG), a low-rank factorization of large directed graphs based on intersecting bipartite components, which efficiently approximates any graph and enables a graph neural network architecture with memory and computational complexity linear in the number of nodes rather than edges, demonstrating competitive performance on various graph tasks.
124. Agricultural Land Management Practices in the Conterminous United States from 1980–2023
Core Problem: A lack of comprehensive, long-term, and spatially detailed historical data on agricultural land management practices in the conterminous United States (CONUS) to understand their environmental impacts.
Key Innovation: Compilation of a comprehensive 0.25-degree gridded dataset detailing agricultural land management practices (planting/harvest dates, fertilizer, tillage, cover crops) in the CONUS from 1980–2023, based on the National Resources Inventory.
125. A benchmark dataset for half-hourly evapotranspiration estimation in China from 2000 to 2024
Core Problem: Existing ChinaFlux observations for evapotranspiration (LE) are limited by short observation periods and extensive data gaps, constraining their applicability in long-term change analyses and multi-scale hydrological studies.
Key Innovation: Development of the first continuous half-hourly ground-based LE benchmark dataset covering China for 2000–2024, using an AutoML-H2O framework for accurate gap-filling and temporal prolongation of observations from 50 ChinaFlux sites.
126. The LULUCF Data Hub: translating global land use emissions estimates into the national GHG inventory framework
Core Problem: Significant discrepancies exist between global estimates of anthropogenic land use CO2 fluxes (from models and Earth observations) and national GHG inventories (NGHGI) due to varying conceptual definitions.
Key Innovation: Development of the 'LULUCF Data Hub,' an online platform that compiles and 'translates' global land use CO2 emissions and removals data (from GCB and GFW) to align conceptually with NGHGI definitions, enhancing comparability and fostering consensus.
127. The countrywide historical gravity dataset of Lithuanian territory
Core Problem: A lack of a unified, modernized, and publicly available historical gravity dataset for Lithuania, hindering applications in quasi-geoid modeling, Earth geopotential models, and geological interpretation.
Key Innovation: Compilation and modernization of a countrywide historical gravity dataset for Lithuania (1951–1962 measurements), converting raw data from paper catalogues into modern gravity and coordinate reference systems, suitable for various geophysical and geological applications.
128. HUST-Grace2026s: A high-resolution static gravity field product from GRACE and GRACE-FO observations (2002–2025)
Core Problem: The need for an improved, high-resolution static gravity field model from GRACE/GRACE-FO observations to accurately monitor mass distribution and serve as a benchmark for long-term mass change studies, as merely adding GRACE-FO data offers limited improvement to existing models.
Key Innovation: Development of HUST-Grace2026s, a high-resolution (up to degree/order 180) static gravity field product from GRACE and GRACE-FO observations (2002–2025), achieved through the application of a stochastic model based on postfit residuals, significantly enhancing accuracy and spatial resolution.
129. Ice sheets big and small
Core Problem: The factors controlling the stability of ice sheets and their future response to environmental perturbations and a warming climate are not fully understood.
Key Innovation: Reconstructions of past ice sheet variability provide crucial insights into the controls on their stability and offer predictions for how they may behave in a warming future.
130. Diffusiophoretic transport of colloids in porous media
Core Problem: Classical models of colloid transport in porous media largely ignore diffusiophoretic migration driven by chemical gradients, potentially misrepresenting actual transport dynamics.
Key Innovation: Demonstrated, through experiments and modeling, that even moderate solute gradients can significantly alter colloid transport in porous media by inducing cross-streamline phoretic migration, challenging classical models and showing broad implications for environmental applications.
131. Semantic rule-guided three-dimensional reservoir modeling method using an improved multiple-point geostatistics simulation
Core Problem: Traditional Multiple-Point Geostatistics (MPS) methods for 3D reservoir modeling rely on manual parameter adjustment, lack uncertainty quantification, and generally do not utilize geological semantic rules, limiting model reliability.
Key Innovation: Proposed a 3D reservoir modeling method integrating geological semantic rule guidance (for sequence stratigraphy and fault kinematics) and an improved MPS simulation (with automatic domain kernel selection and rapid parameter optimization), enhancing geological reliability and accuracy of reservoir models.
132. Integrating OBN seismic data and machine learning for enhanced fluid discrimination in pre-salt carbonate reservoirs
Core Problem: Accurately identifying reservoir fluids in ultra-deep pre-salt reservoirs remains challenging due to complex lithology, strong heterogeneity, and limited well data.
Key Innovation: Developed an integrated workflow combining OBN seismic data, pre-stack inversion, AVO analysis, and an LSTM rock physics model to improve fluid prediction in complex pre-salt reservoirs, autonomously learning nonlinear relationships and reducing identification uncertainties.
133. Don’t quit your day job. Part-time firefighters in rural Norway.
Core Problem: Understanding how part-time firefighters in rural areas, often with limited formal training and resources, effectively respond to a wide array of incidents and contribute to community resilience, especially in the context of reorganizing fire and rescue services, is crucial but often overlooked.
Key Innovation: Analyzed the value of diversity in day-time jobs, practical orientation, and the role of social networks and communal values among part-time firefighters, demonstrating how their embeddedness in rural communities is a key factor for effective emergency preparedness and community resilience.
134. Active learning Kriging with functional dimension reduction for reliability analysis of stochastic dynamical systems
Core Problem: Building high-accuracy surrogate models for efficient reliability analysis of complex, computationally intensive stochastic dynamical systems, especially for estimating first-passage failure probabilities, with limited computational resources.
Key Innovation: Proposes a novel active learning Kriging method based on functional dimension reduction (AKFDR) that constructs Kriging models in a latent functional space, uses a new learning function with a weighted correlation criterion and trajectory misclassification probability for sample selection, and an error-based stopping criterion, leading to efficient and accurate first-passage probability evaluation.
135. Frequency-domain approach to automated and efficient multivariate kernel density estimation for probabilistic modeling
Core Problem: Accurate and efficient multivariate Kernel Density Estimation (KDE) for probabilistic modeling is limited by high computational cost, suboptimal bandwidth selection, and density leakage, especially for large-scale, multidimensional datasets.
Key Innovation: Proposes a frequency-domain approach that reformulates bandwidth selection as a gradient-based optimization task using the discrete cosine transform, decoupling computational complexity from dataset size. It constructs a differentiable objective function with fidelity loss and regularization, minimized by Adam optimizer, leading to improved efficiency and accuracy over classical and transformation-based estimators.
136. Experimental evaluation on bearing performance of tire cell-geogrid reinforced subgrades
Core Problem: There is a need for experimental investigation into the bearing performance and deformation behavior of innovative reinforcement systems like tire cells and tire cell-geogrid combinations for subgrade stabilization under static loading.
Key Innovation: Experimentally demonstrated that tire cells substantially reduce surface settlement and redistribute vertical stresses, and that the addition of a geogrid layer further enhances stress spreading and structural stiffness, highlighting the synergistic potential of the tire cell-geogrid system for sustainable subgrade stabilization.
137. Quantifying Vein Network Permeability in Dehydrated Serpentinites Using Thermodynamics and Generative AI
Core Problem: Poor understanding of fluid escape mechanisms in subduction zones, specifically how dehydration vein networks in serpentinites facilitate fluid percolation.
Key Innovation: Combined X-ray tomography, drone imagery, generative machine learning, electron microscopy, and equilibrium thermodynamics to model and analyze fluid pathways, demonstrating that dehydration vein networks act as efficient drainage systems enabling rapid fluid percolation even at low porosities, potentially influencing fluid migration on local to regional scales.
138. A Physically Consistent Particle Size Distribution Modeling of the Microphysics of Precipitation for Weather and Climate Models
Core Problem: The difficulty in modeling the probability density function of precipitation drops in weather and climate models, where current approaches often make problematic assumptions leading to unphysical results.
Key Innovation: Introduction of a new, mathematically and physically consistent model for particle size distribution of precipitation, seamlessly integrable into microphysics parameterizations, which outperforms existing models and has implications for understanding cloud processes and climate sensitivity.
139. Benchmarking the Dynamic Modes of Stratosphere‐Troposphere Coupling in Northern Annular Modes (NAM) in CMIP6 Models
Core Problem: The dynamics and uncertainty in vertical coupling of the Northern Annular Mode (NAM) across atmospheric levels are not fully resolved, particularly how extreme stratospheric events influence surface weather patterns.
Key Innovation: Applying a linear inverse model (LIM) to NAM indices reveals a leading eigenmode with a deep structure extending from the stratosphere to the surface, an e-folding timescale of ~30 days, and captures downward propagation during weak polar vortex events. This approach reduces uncertainty in surface response to stratospheric extreme events by >50%.
140. Early to Mid‐Holocene Cool Temperature‐Induced Drought in Asian Interior
Core Problem: Hydroclimatic variations in mid-latitude Asia during the early to mid-Holocene and their associated mechanisms remain disputed, hindering understanding of atmospheric circulation controls on regional climatic changes.
Key Innovation: Alkenone records from Siberian lakes and synthesized regional data show early to mid-Holocene drought in the Asian interior, associated with an enhanced anticyclonic system over Eurasia induced by cold air masses. This explains spatial heterogeneity of hydrological changes and contrasting temperature-moisture associations.
141. Attributing Long‐Term Trends in Marine Low Cloud Morphologies to Aerosols and Large‐Scale Meteorology With Deep Learning
Core Problem: The response of marine low-cloud mesoscale morphologies to climate change and emission reductions, and the attribution of long-term trends to aerosols and large-scale meteorology, remain poorly understood.
Key Innovation: A deep learning model (UMorNet) was developed to predict instantaneous cloud morphologies from meteorology and cloud droplet number concentration (aerosol proxy). It achieves 0.55 average test accuracy and captures spatial patterns and long-term trends. Sensitivity experiments identify aerosols (Nd), marine cold-air outbreak index, sea surface temperature, and inversion strength as key drivers, highlighting the role of aerosols in shaping cloud morphological changes.
142. Diverging Spring Warming and Growing Season Shifts Across Eurasia and North America Under Future Climate
Core Problem: Large model uncertainty limits climate risk assessment regarding springtime warming over Northern Mid-High-latitude Land and its effects on plant lifecycles and water resources.
Key Innovation: Developing a novel emergent constraint targeting model divergence in surface-albedo feedback, which halves the spread of projected warming and reveals pronounced geographical asymmetry in future warming and growing season shifts.
143. Simulated Hydrologic Impacts of Cloud Seeding in the North Platte and Little Snake River Basins of Wyoming
Core Problem: Evaluating the precise hydrological impacts and efficacy of cloud seeding for water supply enhancement amidst climate change requires detailed simulation.
Key Innovation: Used a 36-member ensemble of WRF-WxMod simulations to force a hydrological model (WRF-Hydro), quantifying that simulated cloud seeding increases snow water equivalent and streamflow, with 78% of precipitation gains enhancing streamflow.
144. XSPLAIN: XAI-enabling Splat-based Prototype Learning for Attribute-aware INterpretability
Core Problem: 3D Gaussian Splatting (3DGS) models, despite their high-fidelity 3D reconstruction capabilities, lack interpretability for classification tasks, hindering their adoption in critical domains, and existing XAI methods for other 3D representations are often ambiguous.
Key Innovation: XSPLAIN, the first ante-hoc, prototype-based interpretability framework specifically for 3DGS classification. It leverages a voxel-aggregated PointNet backbone and a novel invertible orthogonal transformation to disentangle feature channels for interpretability while strictly preserving decision boundaries, providing intuitive, example-grounded explanations.
145. From Classical to Topological Neural Networks Under Uncertainty
Core Problem: Maximizing the potential of artificial intelligence in domains like military applications requires enhancing the robustness, interpretability, and generalization of AI models, especially when dealing with complex data types and uncertainty.
Key Innovation: Explores and integrates classical neural networks, topological data analysis, topological deep learning, and statistical Bayesian methods to process images, time series, and graphs, demonstrating how topology-aware and uncertainty-aware models can improve AI performance.
146. ERGO: Excess-Risk-Guided Optimization for High-Fidelity Monocular 3D Gaussian Splatting
Core Problem: Generating high-fidelity 3D content from a single image is challenging due to missing geometric/textural information and inconsistencies/misalignments in synthesized auxiliary views, which propagate artifacts during 3D reconstruction.
Key Innovation: Proposed ERGO, an adaptive optimization framework guided by excess risk decomposition, which dynamically estimates view-specific excess risk and adjusts loss weights to robustly handle supervision noise, enhancing both geometric fidelity and textural quality in monocular 3D Gaussian splatting.
147. ICODEN: Ordinary Differential Equation Neural Networks for Interval-Censored Data
Core Problem: Predicting time-to-event outcomes when event times are interval censored is challenging due to unobserved exact event times, and many existing survival analysis approaches rely on strong model assumptions or cannot handle high-dimensional predictors.
Key Innovation: Developed ICODEN, an ordinary differential equation-based neural network for interval-censored data that flexibly models the hazard function without requiring proportional hazards or prespecified parametric forms, demonstrating robust predictive accuracy in high-dimensional biomedical settings.
148. LightGTS-Cov: Covariate-Enhanced Time Series Forecasting
Core Problem: Time series foundation models often ignore exogenous covariates or incorporate them via simple concatenation, limiting their effectiveness in covariate-rich applications like electricity price or renewable energy forecasting.
Key Innovation: LightGTS-Cov, a covariate-enhanced extension of LightGTS that explicitly incorporates both past and future-known covariates via a lightweight MLP plug-in, residually refining forecasts and achieving superior performance in covariate-aware benchmarks and real-world energy applications.
149. Benchmarking Deep Learning and Statistical Target Detection Methods for PFM-1 Landmine Detection in UAV Hyperspectral Imagery
Core Problem: Lack of standardized benchmarks for UAV-based landmine detection using hyperspectral imaging, and the limitations of traditional evaluation metrics (ROC-AUC) for sparse targets.
Key Innovation: Presents a systematic benchmark of classical statistical and a proposed lightweight Spectral Neural Network for PFM-1 landmine detection, emphasizing precision-focused evaluation and releasing pixel-level ground truth masks.
150. C^2ROPE: Causal Continuous Rotary Positional Encoding for 3D Large Multimodal-Models Reasoning
Core Problem: Existing Rotary Position Embedding (RoPE) in 3D Large Multimodal Models (LMMs) disrupts continuity of visual features along the column dimension, causing spatial locality loss, and its 1D temporal causality leads to long-term decay in attention, progressively neglecting earlier visual tokens in long sequences.
Key Innovation: C^2RoPE, an improved RoPE that explicitly models local spatial continuity and spatial causal relationships for visual processing, introduces a spatio-temporal continuous positional embedding mechanism (integrating 1D temporal positions with Cartesian spatial coordinates and frequency allocation) and Chebyshev Causal Masking (determining causal dependencies by Chebyshev distance in 2D space), demonstrating effectiveness in 3D scene reasoning and visual question answering.
151. Neural Additive Experts: Context-Gated Experts for Controllable Model Additivity
Core Problem: The trade-off between interpretability (e.g., in Generalized Additive Models) and predictive accuracy in machine learning, where strictly additive models limit performance, and introducing interactions can obscure feature contributions.
Key Innovation: Proposes Neural Additive Experts (NAEs), a novel framework using a mixture of experts with a dynamic gating mechanism and targeted regularization to balance interpretability and accuracy, allowing for a smooth transition from additive models to those capturing intricate feature interactions while maintaining clear feature attributions.
152. Dynamic Frequency Modulation for Controllable Text-driven Image Generation
Core Problem: Modifying text prompts in text-guided diffusion models for semantic adjustments often leads to unintended global structure changes, and existing intervention methods rely on empirical feature map selection.
Key Innovation: Proposing a training-free dynamic frequency modulation method that manipulates the frequency spectrum of noisy latent variables to maintain structural consistency while enabling targeted semantic modifications in text-driven image generation, outperforming state-of-the-art methods.
153. AugVLA-3D: Depth-Driven Feature Augmentation for Vision-Language-Action Models
Core Problem: Vision-Language-Action (VLA) models primarily rely on 2D images, limiting their spatial understanding and action grounding in complex 3D robotic environments.
Key Innovation: Proposes AugVLA-3D, a novel framework that integrates depth estimation to enrich 3D feature representations, employing a depth estimation baseline to extract geometry-aware 3D cues and an action assistant module to constrain learned 3D representations with action priors, improving VLA model generalization and robustness.
154. Interpretable Graph-Level Anomaly Detection via Contrast with Normal Prototypes
Core Problem: Deep graph-level anomaly detection (GLAD) methods are black-box, and existing explanation methods either lack references to normal graphs or rely on abstract latent vectors, limiting their reliability and interpretability in real-world applications.
Key Innovation: Proposes Prototype-based Graph-Level Anomaly Detection (ProtoGLAD), an interpretable unsupervised framework that provides explanations for detected anomalies by explicitly contrasting them with their nearest normal prototype graphs, which are iteratively discovered using a point-set kernel.
155. Self-Supervised Image Super-Resolution Quality Assessment based on Content-Free Multi-Model Oriented Representation Learning
Core Problem: Assessing the quality of super-resolved (SR) images (SR-IQA) obtained from real-world low-resolution (LR) images is challenging due to complex, unpredictable, and model-dependent degradations, especially in data-scarce domains.
Key Innovation: The paper introduces S3 RIQA, a self-supervised, no-reference SR-IQA approach that pretrains multiple SR model-oriented representations using a content-free contrastive learning framework, enabling domain-adaptive IQA for real-world SR applications and outperforming state-of-the-art metrics.
156. DMP-3DAD: Cross-Category 3D Anomaly Detection via Realistic Depth Map Projection with Few Normal Samples
Core Problem: Cross-category 3D anomaly detection for point clouds typically relies on category-specific training, limiting flexibility in few-shot scenarios where only a few normal examples are available.
Key Innovation: Proposes DMP-3DAD, a training-free framework for cross-category 3D anomaly detection that converts point clouds into multi-view realistic depth images, leverages a frozen CLIP visual encoder for representations, and performs anomaly detection via weighted feature similarity, achieving state-of-the-art performance in few-shot settings.
157. Time Series Foundation Models for Energy Load Forecasting on Consumer Hardware: A Multi-Dimensional Zero-Shot Benchmark
Core Problem: Evaluating whether zero-shot prediction capabilities of Time Series Foundation Models (TSFMs) translate to mission-critical applications like electricity demand forecasting, considering accuracy, calibration, and robustness, remains an open question.
Key Innovation: Presents a multi-dimensional benchmark evaluating four TSFMs against baselines for ERCOT hourly load data on consumer-grade hardware, demonstrating TSFMs' superior accuracy, stable performance with minimal context, and varying calibration, providing practical model selection guidelines for energy load forecasting.
158. Spatial-Morphological Modeling for Multi-Attribute Imputation of Urban Blocks
Core Problem: Accurate reconstruction of missing morphological indicators (FSI, GSI) in urban blocks is crucial for urban planning and data-driven analysis, and existing methods may not fully leverage both morphological and spatial patterns.
Key Innovation: Introduces the spatial-morphological (SM) imputer tool, which combines data-driven morphological clustering with neighborhood-based methods (IDW, sKNN) to reconstruct missing FSI and GSI values, demonstrating superior performance by leveraging both global morphological patterns and local spatial information.
159. LaSSM: Efficient Semantic-Spatial Query Decoding via Local Aggregation and State Space Models for 3D Instance Segmentation
Core Problem: Existing 3D instance segmentation methods from point clouds suffer from query initialization dilemmas and rely on computationally intensive attention mechanisms, leading to inefficiency and suboptimal performance.
Key Innovation: LaSSM introduces an efficient semantic-spatial query decoding method using a hierarchical semantic-spatial query initializer and a coordinate-guided state space model (SSM) decoder with local aggregation and a spatial dual-path SSM block, achieving comprehensive scene coverage, accelerated convergence, and efficient instance prediction with reduced computation.
160. PhyCritic: Multimodal Critic Models for Physical AI
Core Problem: Existing multimodal critic models are primarily trained for general visual domains, lacking optimization for physical AI tasks that require perception, causal reasoning, and planning.
Key Innovation: PhyCritic, a multimodal critic model optimized for physical AI through a two-stage RLVR pipeline (physical skill warmup and self-referential critic finetuning), which enhances physically oriented perception and reasoning, leading to improved judgment stability and physical correctness in physically grounded tasks.
161. SurfPhase: 3D Interfacial Dynamics in Two-Phase Flows from Sparse Videos
Core Problem: Experimentally measuring and reconstructing 3D interfacial dynamics in two-phase flows, especially sharp, deformable liquid-vapor interfaces, from sparse camera views is challenging for classical and existing neural rendering methods.
Key Innovation: Proposes SurfPhase, a novel model that reconstructs 3D interfacial dynamics by integrating dynamic Gaussian surfels with a signed distance function and leveraging a video diffusion model to synthesize novel-view videos, demonstrating high-quality view synthesis and velocity estimation from sparse observations.
162. Co-jump: Cooperative Jumping with Quadrupedal Robots via Multi-Agent Reinforcement Learning
Core Problem: Individual legged robots are constrained by physical actuation limits, making it difficult to perform tasks like high jumps beyond their solo capabilities, especially under high-impulse contact dynamics and decentralized settings without explicit communication.
Key Innovation: Introducing Co-jump, a framework enabling two quadrupedal robots to cooperatively jump significantly higher than solo capabilities through decentralized Multi-Agent Reinforcement Learning (MAPPO) with a progressive curriculum, achieving precise coordination solely via proprioceptive feedback and demonstrating robust performance in simulation and physical hardware.
163. Deep Bootstrap
Core Problem: Traditional bootstrap methods decouple the estimation of conditional distribution, sampling, and nonparametric regression, potentially limiting efficiency and accuracy, especially for high-dimensional or multimodal distributions.
Key Innovation: Proposing a novel deep bootstrap framework for nonparametric regression based on conditional diffusion models, which integrates these components into a unified generative framework for efficient sampling and accurate estimation with theoretical guarantees.
164. Bayesian Signal Component Decomposition via Diffusion-within-Gibbs Sampling
Core Problem: Estimating components of interest from noisy linear superpositions of multiple signals is a crucial pre-processing step in signal processing, requiring a flexible Bayesian framework that can incorporate both model-driven and data-driven prior knowledge in a unified manner.
Key Innovation: Developing a Bayesian framework for signal component decomposition that combines Gibbs sampling with plug-and-play (PnP) diffusion priors, allowing flexible incorporation of prior knowledge and separate learning/combination of component priors, with provable posterior sampling properties and superior performance over existing approaches.
165. Viewpoint Recommendation for Point Cloud Labeling through Interaction Cost Modeling
Core Problem: Semantic segmentation of 3D point clouds, essential for applications like autonomous driving, requires extensive labeled datasets, but the manual labeling process is time-consuming for annotators due to viewpoint tuning and lasso selection.
Key Innovation: Proposes a viewpoint recommendation approach that models the time cost of lasso selection in point clouds using Fitts' law, recommending viewpoints that minimize labeling time, and integrates this into a data labeling system to enhance annotation efficiency.
166. A Gibbs posterior sampler for inverse problem based on prior diffusion model
Core Problem: The difficulty of posterior sampling in ill-posed inverse problems where the observation system is linear with additive noise and the prior density is modeled by a diffusion process learned from a large set of examples.
Key Innovation: Introducing a novel and effective Gibbs algorithm for posterior sampling in this specific class of inverse problems, demonstrating its simplicity, effectiveness, and guaranteed convergence under clearly identified conditions, confirmed by numerical simulations.
167. Renet: Principled and Efficient Relaxation for the Elastic Net via Dynamic Objective Selection
Core Problem: The standard Elastic Net suffers from shrinkage bias, leading to suboptimal prediction accuracy, and existing relaxation implementations for debiasing are naive and can violate Karush-Kuhn-Tucker conditions, especially in high-dimensional, low signal-to-noise ratio, and high-multicollinearity regimes.
Key Innovation: Renet, a principled generalization of Relaxed Lasso to Elastic Net, which addresses shrinkage bias through an adaptive relaxation procedure that dynamically dispatches between convex blending and efficient sub-path refitting, enforcing sign consistency and providing robust debiasing, outperforming standard Elastic Net and Adaptive Elastic Net while maintaining computational efficiency.
168. LCIP: Loss-Controlled Inverse Projection of High-Dimensional Image Data
Core Problem: Current inverse projection methods are limited to generating a fixed surface-like structure in data space, which poorly covers the richness of the high-dimensional space and restricts their utility for tasks like data augmentation, classifier analysis, or data imputation.
Key Innovation: LCIP, a new loss-controlled inverse projection method that allows users to 'sweep' the data space under intuitive control, working generically for any projection technique and dataset, and demonstrated effectively for image manipulation and style transfer.
169. Proficient Graph Neural Network Design by Accumulating Knowledge on Large Language Models
Core Problem: Large Language Models (LLMs) struggle with specialized, data-sensitive tasks like designing Graph Neural Networks (GNNs) due to inherent knowledge gaps in modeling graph properties and architectures, and external noise from misleading inputs, hindering meta-level proficiency.
Key Innovation: Introduces DesiGNN, a knowledge-centered framework that systematically converts past model design experience into structured, fine-grained knowledge priors for meta-learning with LLMs. It aligns empirical property filtering with adaptive elicitation of literature insights, enabling proficient GNN design for unseen datasets with minimal search cost.
170. HypeRL: Hypernetwork-Based Reinforcement Learning for Control of Parametrized Dynamical Systems
Core Problem: Optimal control of parametric dynamical systems in applied sciences and engineering is computationally infeasible for traditional numerical methods, especially when control/state variables are high-dimensional or parameter-dependent.
Key Innovation: Proposes HypeRL, a deep reinforcement learning (DRL) framework using hypernetworks to approximate optimal control policies directly. It learns an optimal feedback control strategy that generalizes across parameter variations by embedding parametric information into the value function and policy NNs, demonstrating effectiveness on complex parametric control problems.
171. On Transferring Transferability: Towards a Theory for Size Generalization
Core Problem: Many modern learning tasks require models that can take inputs of varying sizes, and there's a need for a general framework to understand and ensure transferability of model performance from low-dimensional to higher-dimensional inputs.
Key Innovation: A general framework for transferability across dimensions, showing it corresponds to continuity in a limit space, and providing design principles for creating new transferable models, supported by instantiations on existing architectures.
172. Semantic-Enhanced Time-Series Forecasting via Large Language Models
Core Problem: Existing LLM-based time series forecasting methods struggle with the intrinsic modality gap between linguistic knowledge and time series data patterns, limiting semantic representation, and are weak at modeling short-term anomalies.
Key Innovation: Proposes SE-LLM, which embeds inherent periodicity and anomalous characteristics of time series into the semantic space to enhance token embedding, and includes a plugin module for self-attention to model both long-term and short-term dependencies efficiently.
173. Improving Long-Range Interactions in Graph Neural Simulators via Hamiltonian Dynamics
Core Problem: Existing Graph Neural Simulators (GNSs) struggle to capture long-range interactions and suffer from error accumulation during autoregressive rollouts when simulating complex physical systems.
Key Innovation: Information-preserving Graph Neural Simulators (IGNS), a graph-based neural simulator built on Hamiltonian dynamics principles, guaranteeing information preservation, extending to port-Hamiltonian systems, and incorporating a warmup phase, geometric encoding, and multi-step training objective to improve accuracy and stability in complex dynamical systems.
174. Accelerating Streaming Video Large Language Models via Hierarchical Token Compression
Core Problem: Streaming Video Large Language Models (VideoLLMs) face significant real-time deployment challenges due to high computational costs from processing dense visual tokens and redundant processing of temporally similar frames during ViT encoding and LLM pre-filling.
Key Innovation: Streaming Token Compression (STC), a plug-and-play hierarchical framework with STC-Cacher and STC-Pruner, optimizes both ViT encoding and LLM pre-filling stages, significantly reducing latency and memory overhead while retaining high accuracy in streaming VideoLLMs.
175. EvoXplain: When Machine Learning Models Agree on Predictions but Disagree on Why -- Measuring Mechanistic Multiplicity Across Training Runs
Core Problem: Machine learning models, even when achieving high predictive accuracy, may rely on different internal logic, leading to unstable or multiple explanations that undermine trust and understanding of their mechanisms.
Key Innovation: EvoXplain is a diagnostic framework that measures the stability of model explanations across repeated training runs, revealing when single instance explanations obscure the existence of multiple admissible predictive mechanisms and reframing interpretability as a property of the training pipeline.
176. SnapGen++: Unleashing Diffusion Transformers for Efficient High-Fidelity Image Generation on Edge Devices
Core Problem: Diffusion transformers (DiTs), despite their high standards in image generation, are impractical for on-device deployment due to their significant computational and memory costs.
Key Innovation: SnapGen++ is an efficient DiT framework tailored for mobile and edge devices, combining a compact DiT architecture with adaptive global-local sparse attention, an elastic training framework for optimizing sub-DiTs, and Knowledge-Guided Distribution Matching Distillation for high-fidelity, low-latency generation.
177. Multi-Objective Bayesian Optimization for Networked Black-Box Systems: A Path to Greener Profits and Smarter Designs
Core Problem: Designing complex industrial systems requires balancing multiple competing objectives (e.g., profitability, resilience, sustainability) within networked black-box models, and existing multi-objective optimization methods often impose rigid structural assumptions or fail to capture cyclic dependencies and feedback loops.
Key Innovation: MOBONS, a novel Bayesian optimization-inspired algorithm, provides a unifying approach to grey-box multi-objective optimization by leveraging network representations to efficiently optimize general function networks, including those with cyclic dependencies, while incorporating constraints and preserving sample efficiency.
178. LighthouseGS: Indoor Structure-aware 3D Gaussian Splatting for Panorama-Style Mobile Captures
Core Problem: Accurate camera pose and 3D point estimation for novel view synthesis from panorama-style mobile captures are challenging, especially in textureless indoor scenes, due to rotation-dominant motion and narrow baseline, leading to inconsistent geometry and optimization instability.
Key Innovation: Introduces LighthouseGS, a 3D Gaussian Splatting framework that leverages rough geometric priors and indoor planar structures, proposing a plane scaffold assembly initialization, stable pruning strategy, and geometric/photometric corrections to achieve photorealistic rendering from mobile captures.
179. Synthetic Homes: An Accessible Multimodal Pipeline for Producing Residential Building Data with Generative AI
Core Problem: Computational models for energy modeling research require extensive, often inaccessible, expensive, or privacy-sensitive residential building data.
Key Innovation: A modular multimodal framework and pipeline that uses generative AI to produce realistic residential building data from publicly accessible images and information, reducing reliance on costly or restricted data sources.
180. Conformal Prediction for Compositional Data
Core Problem: The absence of well-established methods for constructing valid prediction sets for compositional data within Dirichlet regression models, especially considering the unique geometry of the compositional space.
Key Innovation: Investigation and evaluation of three conformal prediction-based strategies (quantile residuals, approximate HDR, and grid-based HDR discretization) for constructing valid predictive regions in Dirichlet regression models, demonstrating their robustness, improved coverage, and reduced prediction region area for compositional data.
181. Research on drift limits for floating offshore substations based on the mechanical strength of dynamic cables
Core Problem: The inherent conflict between load-bearing capacity and draft limitations in Floating Offshore Substations (FOSSs), leading to a lack of mature design guidelines and established displacement criteria to protect critical submarine cables.
Key Innovation: Simulated a Cable–FOSSs model under dynamic wave loads, measured tensile tension and bending curvature, analyzed maximum FOSSs displacement at cable failure, and proposed optimization schemes for cable attachments, providing key design criteria for FOSSs.
182. ESA Arctic+ Salinity Product v4: Enhanced Retrievals Near the Ice-Edge
Core Problem: Existing SMOS sea surface salinity (SSS) products in the Arctic suffer from contamination and reduced accuracy near the ice edge, limiting data coverage and quality.
Key Innovation: Developed the ESA Arctic+ Salinity Product v4, incorporating specific algorithms at Level 1 and Level 2 to reduce ice-edge contamination, resulting in a 25% increase in coverage and a 24-43% reduction in RMS difference compared to in situ measurements, significantly enhancing SSS retrieval accuracy near ice edges.
183. Spatial Ergodicity of Doppler Characteristics in Polarimetric Ocean Radar Scattering: A Numerical Study
Core Problem: Understanding the spatial ergodicity of Doppler characteristics in polarimetric ocean radar scattering is crucial for accurate sea surface current retrieval, especially considering different polarizations and sea states.
Key Innovation: Numerically investigated the spatial ergodicity of Doppler shift and width using the full Apel wave spectrum and the second-order small-slope approximation (SSA-2) model, determining that co-polarization achieves spatial ergodicity at smaller illumination sizes (>= 1/4 largest gravity wave wavelength) than cross-polarization (>= 1/2 largest gravity wave wavelength), with limited effect from wind direction.
184. Refining the Lagrangian approach for moisture source identification through sensitivity testing of assumptions using BTrIMS1.1
Core Problem: Lagrangian models for moisture tracking rely on assumptions that are seldom thoroughly tested, leading to potential inaccuracies in moisture source identification, which is crucial for understanding precipitation patterns and extreme weather events.
Key Innovation: Developed BTrIMS1.1 by systematically testing and refining key assumptions (number of parcels, release height, vertical movement, identification methods) in Lagrangian moisture source identification, improving accuracy and applicability for future climate change studies.
185. Implementation of a multi-layer snow scheme in the GloSea6 seasonal forecast system: impacts on land–atmosphere interactions and climatological biases
Core Problem: Traditional single-layer snow schemes in land surface models inadequately represent the insulating effect of snowpack, leading to climatological biases (cold/warm) in seasonal forecast systems and poor simulation of land-atmosphere interactions.
Key Innovation: Implemented a multi-layer snow scheme in the GloSea6 seasonal forecast system, which improved the simulation of Northern Hemisphere snow seasonality, delayed snowmelt onset, enhanced soil moisture memory, and mitigated near-surface warming biases and precipitation errors.
186. Multi-year La Niña–El Niño transition influenced Earth’s extreme energy uptake in 2022–2023
Core Problem: The specific causes of Earth's extreme energy uptake in 2022–2023 and the precise role of internal climate variability, such as the La Niña-El Niño transition, in shaping Earth's energy imbalance remain unclear.
Key Innovation: Multi-model climate simulations and satellite observations revealed that the transition from a multi-year La Niña to El Niño was the key driver, explaining approximately 75% of Earth's extreme energy uptake in 2022–2023, highlighting the dominant influence of internal climate variability.
187. Preserving water under megacities is crucial — and urgent
Core Problem: The urgent need to preserve water resources beneath megacities, which are under increasing pressure.
Key Innovation: Highlighting the critical importance and urgency of urban water preservation as a key environmental and societal challenge.
188. Gas vortex discovery in butterfly microcavities for constructing ultrasensitive gas sensors
Core Problem: Gas sensors typically face a sensitivity-stability trade-off in trace-gas detection, often due to reactive surface modifications or inefficient gas-solid interaction time.
Key Innovation: Discovery of gas vortex effects in butterfly wings to prolong molecular residence time, applied to gas sensor design using periodic microcavities to achieve ultralow detection limits and long-period stability through geometric fluidic control rather than material chemistry.
189. Robotic conformal 4D printing of liquid crystal elastomers
Core Problem: Existing 4D printing methods for liquid crystal elastomers (LCEs) are limited by the 2D alignment of mesogens, restricting the range of achievable shape transformations and applications.
Key Innovation: Development of a robotic direct-ink-writing conformal 4D printing technology that enables deposition of LCEs onto complex, nonplanar 3D substrates, unlocking new design spaces for reversibly shape-changing structures and enabling applications like on-demand protective coatings and structural repair.
190. Navigating high-dimensional processing parameters in organic photovoltaics via a multitier machine learning framework
Core Problem: Optimizing organic photovoltaic (OPV) performance requires navigating high-dimensional, interdependent processing parameters governing bulk heterojunction morphology.
Key Innovation: Developed a three-tiered machine learning framework using gradient boosting regression trees, leveraging a standardized database, to achieve robust optimization of OPV photoactive layers with high accuracy and generalization, identifying optimal multiparameter configurations.
191. Nonlinear optical extreme learner via data reverberation with incoherent light
Core Problem: Artificial neural networks have growing energy and computational demands, and optical neural networks suffer from weak optical nonlinearities.
Key Innovation: Demonstrated a low-power, incoherent-light-compatible optical extreme learner that leverages 'data nonlinearity' from optical pattern reverberations, achieving nonlinear transformations at extremely low optical power and outperforming linear digital networks in classification tasks.
192. A novel data-driven health status assessment model based on multiple criteria appraisal recommendation with three-parameter interval grey number
Core Problem: Accurate health status assessment from large-scale historical data is challenging, as existing multiple criteria appraisal recommendation (MCAR) models struggle to effectively extract knowledge from data and represent inherent data uncertainties.
Key Innovation: Proposes a novel data-driven health status assessment model that improves MCAR by embedding interval rule-bases for knowledge extraction, defining three-parameter interval grey numbers (TPIGN) with a new distance to capture data uncertainties, and using an interval evidential reasoning (IER) algorithm for accurate appraisals, demonstrating high accuracy and robustness on benchmark datasets.
193. Blast pressure variability from a vehicle-borne improvised explosive device and its effect on the risk of progressive collapse
Core Problem: Accurately characterizing the spatial variability of blast pressure from vehicle-borne improvised explosive devices (VBIEDs) and assessing its impact on the risk of progressive collapse in reinforced concrete buildings, as conventional blast modeling is often overly conservative.
Key Innovation: Probabilistic characterization of spatial variability of airblast for VBIEDs, including correlation on adjacent structural components, demonstrating that modeling VBIED blast loads reduces collapse probabilities by over 50% compared to bare charge assumptions, with the Alternate Path method reducing damage risks by over 99%.
194. Knowledge-data-model-driven multimodal few-shot learning for hyperspectral fine classification: Generalization across sensor, category and scene
Core Problem: Previous few-shot learning methods for hyperspectral fine classification struggle to identify unseen or rare land cover species (e.g., one sample/shot) and lack generalization across inevitable cross-sensor, cross-category, and cross-scene variations, limiting their real-world application for fine-grained mapping.
Key Innovation: Proposes 'Knowing-Net,' a knowledge-data-model-driven multimodal few-shot learning network. It leverages prior spectral knowledge for cross-sensor image reconstruction, embeds multimodal data (textual descriptions, natural images) for unseen classes, and uses a cross-alignment mechanism and a sliding discriminant window to enhance generalization and robustness for fine-grained land-cover mapping.
195. Improving national forest attribute maps of Sweden with machine learning
Core Problem: Existing modeling approaches for nationwide forest attribute mapping can be enhanced, and the potential of machine learning (ML) to leverage airborne laser scanning (ALS) and field inventory data for improved accuracy needs exploration, especially by incorporating spatially correlated surrounding data.
Key Innovation: Explores extreme gradient boosting (XGBoost) and convolutional neural networks (CNN) to relate ALS data with Swedish National Forest Inventory (NFI) field data, including surrounding spatially correlated ALS data. This approach achieved significant improvements in prediction accuracy across five forest variables (e.g., height, diameter, biomass), with further enhancements from incorporating surrounding information, forming a promising basis for large-scale forest monitoring.
196. Numerical manifold method for the transient HMC fully coupled model in triple-layer composite liners
Core Problem: Existing numerical methods (like FEM) struggle with accurately modeling transient hydro-mechanical-chemical (HMC) fully coupled behavior and interface continuity in weakly discontinuous porous media like triple-layer composite landfill liners, hindering accurate simulation of leachate migration.
Key Innovation: Developed a Numerical Manifold Method (NMM) for a transient HMC fully coupled model in triple-layer composite liners, overcoming FEM drawbacks by constructing approximations that exactly satisfy interface continuity conditions for displacement, pore pressure, and pollutant concentration, accurately simulating solute migration.
197. Determination of well stability and sand risk minimization parameters for gas condensate field conditions using geomechanical and CT-based approaches
Core Problem: Ensuring well stability and minimizing sand production risk in complex reservoir rocks, particularly in Arctic shelf gas condensate fields, requires a comprehensive understanding of deformation, fracture, and filtration processes.
Key Innovation: Integrated geomechanical and CT-based studies (mechanical properties, true triaxial physical modeling, triaxial sand production, digital CT-analysis) to determine drawdowns for wellbore stability, optimal gravel filter screen parameters, and localize failure initiation points in weakly cemented formations.
198. A dual attention-based deep learning model for lithology identification while drilling
Core Problem: Inaccurate and limited real-time lithology identification during drilling due to traditional models' poor feature extraction and adaptability to complex geological conditions.
Key Innovation: Development of a dual attention-based deep learning model (LSTM with dual attention) optimized by the crayfish optimization algorithm (COA) to significantly improve the accuracy and adaptability of lithology identification while drilling.