Agent Papers Quadruple as NeurIPS Meets on Three Continents

RL with verifiable rewards for reasoning 38→69

Train reasoning LLMs with reinforcement learning and analyze mathematical reasoning gains.

The Reasoning Boundary Paradox: How Reinforcement Learning Constrains Language Models

SARL: Label-Free Reinforcement Learning by Rewarding Reasoning Topology

Efficient and adaptive thinking 35→68

Control reasoning length and self-verification so models think wisely rather than longer.

LYNX: Learning Dynamic Exits for Confidence-Controlled Reasoning

Beyond What Seems Necessary: Hidden Gains from Scaling Training-Time Reasoning Length under Outcome Supervision

Test-time compute scaling 26→65

Allocate inference-time compute through confidence, branching, and self-consistency strategies.

Test-Time Scaling in the Wild: Why Exploitation, Not Exploration, Is the Bottleneck

SPECS: Faster Test-Time Scaling through Speculative Drafts and Dynamic Switching

Long CoT dynamics and efficiency 16→37

Study the geometry, legibility, and efficiency of long reasoning chains.

Beyond Semantics: The Unreasonable Effectiveness of Reasonless Intermediate Tokens

Efficient Reasoning via Constrained Optimization in Latent Space

LLM pruning and scaling data 25→42

Prune LLMs after training and study data effects on scaling and collapse.

DataFlex: A Unified Benchmark and Evaluation Platform for Data-Centric Training of Large Language Models

UnlearningSoup: Is Repeated Tuning Necessary for Large Language Model Unlearning?

LLM unlearning and benchmarks 32→35

Evaluate LLMs on unlearning, fallacies, and procedure generalization, with adaptation to domains.

An Empirical Study on Noisy Data and LLM Pretraining Loss Divergence

It's All Training: A Fully Synthetic Single-Stage Recipe for LLMs

Uncertainty quantification for LLMs 11→30

Quantify confidence and uncertainty to benchmark and evaluate language models.

Why Are LLMs Confidently Wrong? Correcting Overconfident Errors via Causal Head Intervention

Hidden Measurement Error in LLM Pipelines Distorts Annotation, Evaluation, and Benchmarking

Statistical frameworks for LLM evaluation 14→30

Use conformal prediction, ranking models, and noise analysis for valid LLM evaluation.

Black-Box Uncertainty Quantification for Large Language Models via Ensemble-of-Ensembles

Brittlebench: Quantifying LLM robustness via prompt sensitivity

Mathematical and structural reasoning benchmarks 29→46

Benchmark LLM math and structural reasoning, with decomposition-based analysis.

SH$^2$: A Mathematician-Curated benchmark for Assessing Research-level Math Capabilities of LLMs

When Form Changes but Logic Doesn’t: Building Logic-invariant LLMs through Structures

Software engineering agents 17→37

Build and evaluate coding agents using execution feedback and code-graph models.

CSBench: A Comprehensive Benchmark for Evaluating Project-Level System Construction in Computer Science

EnvFaultBench: Benchmarking LLM Agents on Software Environment-Fault Troubleshooting

Agent instruction-following benchmarks 23→32

Evaluate and improve LLM agents' instruction following and decision-making.

PACE: A Proxy for Agentic Capability Evaluation

A Process-Level Evaluation of LLM Discovery Agents

Long-horizon LLM agents 17→30

Train and structure LLM agents for long-horizon planning, exploration, and in-context learning.

\$OneMillion-Bench: How Far are Language Agents from Human Experts?

ACED-Bench: Evaluating How LLM Agents Acquire and Act on Causal Evidence

Data commons and human baselines 11→39

Measure AI's societal footprint, data consent, and rigor of human baselines.

FraudBench: A Legal Evaluation of AI Deception on Realistic Tasks

RealityTest: How People Probe AI Identity and Whether Models Disclose It

Social-science view of AI evaluation 16→36

Argue for rigorous, socially grounded measurement and broader evaluation of AI systems.

The Window for Preventive Action Before AI Saturates Most Cognitive Benchmarks is Closing

AI Evaluation Should Require Standardized Item-Level Data Releases

Human-centered trust and agency 12→27

Model human trust, reliance, and agency when people work with AI systems and regulation.

Humans Correct Their Judgements Through Debate, but Weaker AI Models Not

Stop Using Plausibility as the Criterion for Explainable AI

Benchmarking and reproducibility practice 7→21

Promote common task frameworks, dataset documentation, and replication for scientific ML.

Peer Review of Applied Machine Learning Papers Must Include Code Execution

Stop Mechanizing Reform Heuristics as Scientific Quality Filters in AI Review

Knowledge editing in LLMs 12→28

Edit or update stored knowledge in language models via parameters and memory modules.

Exposing the Illusion of Erasure in Knowledge Editing for LLMs

Capacity Allocation at the Source: Sparse Target Optimization for LLM Knowledge Editing

Retrieval-augmented reasoning 9→27

Improve retrieval-augmented LMs with context compression, search-enhanced reasoning, and generative retrieval.

Knowing What is Missing: Efficient Conversational Memory via Explicit Evidence-Gap Tracking

Retriever-Free Retrieval-Augmented Reasoning via Corpus-Traversing MCTS

RAG robustness and evaluation 17→27

Benchmark, critique, and harden retrieval-augmented generation against noise and attacks.

Latent Abstraction for Retrieval-Augmented Generation

Conflict-Suppressed RAG: A Simple Decoding-Time Framework for Faithful Retrieval-Augmented Generation

Knowledge-augmented question answering 5→12

Inject graph or specialized knowledge into LLM prompts and benchmark question answering.

Medmarks: A Comprehensive Open-Source LLM Benchmark Suite for Medical Tasks

A Semantic-Sampling Framework for Evaluating Calibration in Open-Ended Question Answering

Fine-tuning for reasoning data 21→35

Improve reasoning via tailored losses, scaled instruction data, and efficient distillation.

Strong Post-Training from Permissive, Reasoning-Dominant, Web-Scale Pretraining

From Weeks to Hours: Fast and Principled SFT Curation for LLM

Efficient LLM fine-tuning systems 30→26

Cut memory and compute of fine-tuning through sparsity, projections, and distributed training.

One Algorithm, Two Goals: Dual Scoring for Parameter and Data Selection in LLM Fine-Tuning

Co-evolution: A "One-to-many" LLM Fine-Tuning Paradigm

Safety alignment after fine-tuning 11→22

Defend against harmful fine-tuning with post-hoc and in-training safety alignment.

When Safety Becomes An Outlier: Understanding the Retention of LLM Safety Behaviors

The Geometry of Alignment Collapse: When Fine-Tuning Breaks Safety

Fine-tuning generalization and delta compression 14→9

Study fine-tune deltas, emulation, and mixtures to preserve generality when specializing LLMs.

Alignment Dynamics in LLM Fine-Tuning

Adaptive and Neutral Theory of Evolution Strategies for Large Language Models

Robust and calibrated DPO 12→18

Make direct preference optimization robust to noisy preferences and distribution shift through calibration and data selection.

PORT: Preference Optimization via Robust Token-Level Reweighting

Learning Where It Matters: Geometric Anchoring for Robust Preference Alignment

LLM-as-judge and feedback data 8→18

Collect scaled AI feedback and study human-versus-LLM judgment in evaluation.

From Preference Data to Personalization: Tracing Sycophancy in Large Language Models

Auditing the Judge: Human-Grounded Bias Discovery, Quantification, and Mitigation in LLM Judges

Steering and correcting LLM behavior 10→17

Align LLMs through representation editing, influence functions, and intrinsic corpus signals.

MATO: Multi-objective Personalized Alignment with Test-time Optimization for Large Language Models

Theoretical Limits of Language Model Alignment

DPO design choices 9→11

Study data, scheduling, dynamic beta, and multi-reference variants of direct preference optimization.

MaPPO: Maximum a Posteriori Preference Optimization with Prior Knowledge

Ultra-DPO: Reimagining LLM Alignment as a Semi-Supervised Task

Chain-of-thought multimodal reasoning 28→88

Improves multimodal LLM reasoning with latent thinking, long chains, and structured scene-graph reasoning.

SLVR: Structured Latent Visual Reasoning via Human-like Reasoning Flows

MMFineReason: Closing the Multimodal Reasoning Gap via Open Data-Centric Methods

Spatial reasoning in VLMs 24→63

Improves grounded spatial reasoning and perspective understanding in vision-language models.

Lost on Campus: Evaluating Embodied Spatial Reasoning of Vision-Language Models in the Wild

MARBLE: an Agent Benchmark for Spatial Reasoning and Visual Abstraction

Multimodal LLM benchmarks 32→51

Builds comprehensive and evolving benchmarks for large multimodal models.

ARK: A Dual-Axis Multimodal Retrieval Benchmark along Reasoning and Knowledge

InterLV-Search: Benchmarking Interleaved Multimodal Agentic Search

VLM capability analysis 33→42

Probes and refines vision-language models, assessing priors, reasoning, and design choices.

DataComp-VLM: Improved open datasets for Vision-Language Models

In-Context Learning Can Help Vision Language Models Overcome Training Prior

Vision-language-action foundation models 22→108

Trains and benchmarks end-to-end VLA policies, with reasoning, unified representations and efficient token caching.

ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models

SimVLA: Attributing Gains in VLA Models Through Controlled Ablation

Memory and subgoal-based manipulation 10→15

Improves robot manipulation with memory, behavioral phasing, subgoal images and skill decomposition.

ForceLDM: A Force-aware Latent Dynamics Model for Contact-Rich Manipulation

Chain-of-Correction: Progress-Aware Policy Steering via Anchor-Grounded Predictive Reasoning

Pretraining and data augmentation for VLN 5→13

Improves navigation agents through pretraining, synthetic instructions, imitation learning and exploration with scene objects.

InfoNav: A Unified Value Framework Integrating Semantic Relevance and Information Gain for Zero-Shot Object Goal Navigation

Sparse Video Generation Propels Real-World Beyond-the-View Vision-Language Navigation

Embodied multimodal LLM agents 5→8

Applies pretrained multimodal LLMs and embodied chain-of-thought to interactive agents, with benchmarks.

Long-term Embodied Visual Tracking with Lightweight Vision-Language-Action Models

WorldVLN: Autoregressive World Action Model for Aerial Vision-Language Navigation

Multimodal fusion with missing modalities 15→34

Learns aligned multimodal representations robust to missing or imbalanced modalities.

Do More Modalities Always Help? A Geometric Perspective on Missing-Modality Robustness

Fusion or Confusion? Multimodal Complexity Is Not All You Need

Medical multimodal fusion 8→19

Combines clinical and imaging modalities with decomposition, symmetric consistency and graph structure.

When Are Multimodal Predictions Biologically Supported? A Diagnostic Evaluation Framework

Pan-FM: A Pan-Organ Foundation Model with Saliency-Guided Masking for Missing Robustness

Contrastive vision-language alignment 6→14

Refines contrastive and preference-based alignment between image and text embeddings.

LoMo: Local Modality Substitution for Deeper Vision-Language Fusion

Strong Helps Weak: Directional Cross-Modal Alignment Transfer in Multi-modal LLMs

Multi-modality image fusion 5→13

Fuses RGB, infrared, event and other modalities with transformers and semantic priors.

MODULE: A Mutual-Promoting Deep Unfolding Framework Towards Degradation-Robust Multi-modal Image Fusion

FusionNeXt: Sequence-First 3D Multi-Modal Fusion in the Era of LLMs

fMRI visual decoding 8→13

Decodes and synthesizes visual stimuli from brain recordings using semantic and cross-subject models.

Prototypes of the Mind: A Unified Framework for Probing the Visual Brain

BRIDGE: Brain-Vision Representation Integration through Depth and Granularity Encoding

Language-grounded visual concept learning 4→12

Learns visual concepts by tokenization, decomposition, and transfer from language models.

A General Concept-based Decomposition for Vision–Language Embeddings

Revealing the Gap in Human and VLM Scene Perception through Counterfactual Semantic Saliency

Medical vision-language models 21→12

Pretrains medical vision-language models for radiology reports and 3D imaging.

MedVTok: A General-Purpose Medical Visual Tokenizer

A World Model of Radiologist Reading for Medical Image Representation Learning

Vision-language pretraining 12→9

Pretrains unified vision-language models using image-as-language tokens and decoupled visual encoding.

Representation Forcing for Bottleneck-Free Unified Multimodal Models

HYDRA: Representation Harmonized Tokenization for Multimodal Generation and Understanding

Open-vocabulary alignment from VLMs 7→9

Refines vision-language alignment and similarity scoring to enable open-vocabulary recognition and segmentation.

GLINT: Sparsely Gated Vision-Language Alignment for Fine-Grained Radiology Representations

SPEXT: A Decoupled Multi-Spectral Foundation Model for Earth Observation and Vision-Language Grounding

Attribute-based zero-shot learning 5→7

Uses attribute-guided transformers, prototypes and semantic grounding to classify unseen classes.

Align3D-AD: Cross-Modal Feature Alignment and Dual-Prompt Learning for Zero-shot 3D Anomaly Detection

Voxel as Token: A New Perspective for Zero-shot Cross-subject Vision Decoding

Adapting VLMs for zero-shot classification 5→7

Improves zero-shot classification by adding text descriptions, LLM prompts and long-tail handling to VLMs.

Auditing Capsule Vision 2024: Within-Split Train-to-Validation Re-Exposure and a Kvasir-Channel Sensitivity Diagnostic

Cross-Layer Evolution Graph Learning for Fine-grained VLM Hallucination Detection

Recaptioning and text-supervised pretraining 2→7

Uses synthetic captions and interleaved image-text data to learn better visual representations.

Text as Partial Constraint: Core–Residual Alignment for Robust Vision–Language Learning

Aligning Forest and Trees in Images & Long Captions for Visually Grounded Understanding

Multimodal question answering datasets 3→4

Creates benchmarks and models for multi-hop, video, and text-based multimodal question answering.

Not Another Text Benchmark: Putting the “Visual" Back in Visual Question Answering for Large Video Models

Trust What Matters: Language-Conditioned Evidence Routing for Video-IMU Action Question Answering

Universal VQA models 3→2

Develops general and knowledge-based visual question answering with memory, reasoning transfer, and causal adapters.

Selective Answering for Medical VQA via Parallel Independent Claim Verification

SynTeX-FL: Cross-Modal Text Transfer in Federated Learning for Medical Visual Question Answering

Multimodal LLM question answering 2→2

Augments multimodal LLMs for knowledge-based and text-rich VQA and evaluates them dynamically.

SIEVES: Selective Prediction Generalizes through Visual Evidence Scoring

CoMet: Context and Multiplicity Decomposition for Multimodal Uncertainty Estimation

Visual reasoning diagnostics 1→2

Diagnoses shortcuts and compositional reasoning in visual reasoning models using synthetic benchmarks.

SitCom: Scaling Egocentric Multi-Party Spoken Dialogue for Situated Communication Assistance

The World is Not Mono: Enabling Spatial Understanding in Large Audio-Language Models

Sparse attention for long-context LLMs 12→50

Selects or compresses tokens and keys to cut attention cost in long-context LLM inference.

DisSparse: Pipelined Top-$p$ Sparse Attention for Long-Context LLM Serving

Full Attention Strikes Back: Transferring Full Attention into Sparse within Hundred Training Steps

Transformer expressivity and learnability 21→42

Theoretically analyzes what algorithms and structured data transformers can represent and learn.

Transformers Converge to Invariant Algorithmic Cores

Predictive Geometry of Hidden Trajectories in Transformers

State space and recurrent sequence models 15→28

Builds SSM, long-convolution, and recurrent models for efficient long-sequence modeling.

Raven: High-Recall Sequence Modeling via Sparse Memory Routing

Approximate Memory Suffices for Universal Approximation with State Space Models

KV cache eviction and compression 16→23

Prunes and compresses key-value caches with adaptive budgets to cut LLM inference memory.

KV Cache Compression via Attention Output Distortion Minimization

EchoKV: Efficient KV Cache Compression via Similarity-Based Reconstruction

Layer-wise analysis of language models 14→33

Probes hidden representations and layer dynamics in pretrained language models.

Slower Generalization, Faster Memorization: A Sweet Spot in Algorithmic Learning

Dynamics of the Transformer Residual Stream: Coupling Spectral Geometry to Network Topology

Distributed LLM inference systems 14→30

Optimizes LLM serving and training with parallelism, KV reuse, batching, and GPU scheduling.

ReCoVer: Resilient LLM Pre-Training System via Fault-Tolerant Collective and Versatile Workload

Beyond the Trial-and-Error Loop: Hybrid Projection and Automated Tuning for Distributed Training

Sparse dictionary representations 6→20

Extracts interpretable features with sparse autoencoders and dictionary views of implicit neural representations.

Mechanistic Interpretability with Sparse Autoencoder Neural Operators

Sanity Checks for Sparse Autoencoders: Do SAEs Beat Random Baselines?

Neural scaling laws 19→16

Fits scaling laws for model size, data, weight decay, and merging in language and vision.

Decomposing and Reshaping Scaling Laws through Token Learning Times

Tokens-per-Parameter Coverage Is Critical for Robust LLM Scaling Law Extrapolation

Low-rank compression and merging 20→38

Compresses and adapts models with low-rank factorization and weight merging.

Dynamic Model Merging Made Slim

Stabilizing the Dynamic Low-Rank Training

LoRA variants and theory 14→29

Improves low-rank adaptation through new parameterizations, optimization fixes, and expressiveness analysis.

SDS-LoRA: Overcoming Anisotropic Gradient Scaling in Low-Rank Adaptation

Beyond LoRA vs. Full Fine-Tuning: Gradient-Guided Optimizer Routing for LLM Adaptation

Adapter-based parameter-efficient tuning 6→18

Adds lightweight adapters and reparameterized modules for memory- and parameter-efficient fine-tuning.

LOFT: Low-Rank Orthogonal Fine-Tuning via Task-Aware Support Selection

RoSA: Rotational Sparse Adaptation for Memory-Efficient Fine-Tuning

Layer-wise pruning of LLMs 12→10

Allocates sparsity across layers for one-shot or structured pruning of large language models.

The Sparsity Whisperer

TOM-Pruning: Target-aware Output Manifold for LLM Pruning

Post-training quantization of LLMs 6→31

Handles outliers and low-bit errors when quantizing large language models after training.

D$^2$Quant: Accurate Low-bit Post-Training Weight Quantization for LLMs

GSQ: Highly-Accurate Low-Precision Scalar Quantization for LLMs via Gumbel-Softmax Sampling

KV cache quantization 6→27

Compresses key-value caches with vector and mixed-precision quantization for long-context inference.

RankVQ: Low-Rank Parameterized Commutative Vector Quantization for KV Cache Compression

ResKV: Residual-based Channel-wise Unstructured Pruning for KV Cache Compression

Post-training quantization of ViTs 5→18

Reduces quantization error in vision transformers using Hessian, Fisher, and outlier-aware reconstruction.

DiRotQ: Rotation-Aware Quantization for 4-bit Diffusion Transformers

Post-Training Quantization with Gradient-Projected Fisher Approximation for Vision Transformers

Post-training weight quantization 2→6

Compresses weights by quantization with sharpness control, loss-aware objectives, and quantization noise.

ASVQ: What Reparameterization Is Just Enough for Efficient Codebook Learning?

Provable Quantization with Randomized Hadamard Transform

Expert routing and specialization 13→26

Improves expert specialization, soft routing, and interpretation in mixture-of-experts models.

Towards Principled Fine-Grained MoE Expert Pruning via Pseudo-Boolean Approximation

Token-Conditional Expert Dropout: Implicit Regularization for Stable MoE Pretraining

Efficient MoE language model systems 8→18

Restructures LLMs into MoE and speeds distributed MoE training with routing and scheduling.

Structure, Subspace and System: Push the Real Limit of Extremely Low-Bit Quantization for MoE-LLMs

dMoE: dLLMs with Learnable Block Experts

MoE LLM compression 3→5

Compresses mixture-of-experts LLMs through decomposition, delta decompression, and expert-level quantization.

From Experts to Sub-experts: Fine-grained Parameter-Efficient Fine-Tuning for MoE LLMs

FourierMoE: Fourier Mixture-of-Experts Adaptation of Large Language Models

Spiking transformers 10→14

Adapts transformer architectures to spiking networks with token sparsification and ANN-to-SNN conversion.

SpikingGamma: Temporally Precise Online SNN Training Through Smoothed Temporal Delays

FiTS: Interpretable Spiking Neurons via Frequency Selectivity and Temporal Shaping

Training deep spiking networks 5→12

Trains spiking neural networks with implicit differentiation, distillation, calibration, and energy-efficient search.

Spiking neural network initialization for scale-invariant maximization of entropy

Parallel Fixed-Point Spiking Neurons for Efficient Training of Spiking Neural Networks

Group symmetry equivariant networks 2→4

Builds networks equivariant to Lie groups, diffeomorphisms, and gauge symmetries.

Winning the Symmetry Lottery: Learning Invariance from Data Augmentations with Transformers

On Making $SE(2)$-Invariant Networks Optimal

Biologically inspired CNN explanation 1→3

Explains CNN behavior and links networks to human visual cortex and perceptual learning.

DD-CAM: Minimal Sufficient Explanations for Vision Models Using Delta Debugging

AuGhostmentation: The Eyes Never Stand Still—$\textit{Why Should CNNs?}$

Large and dynamic kernel design 1→2

Rethinks convolution kernel size, shape, and dynamic composition to enlarge receptive fields.

KernelDNA: Cross-Layer Kernel Sharing via Decoupled Neural Adapters

kVNN: Learnable Volterra Network Kernels

Bi-level differentiable NAS 0→6

Reformulates gradient-based architecture search optimization with sampling, mixed-level, and geometry-aware methods.

Structuring Open-Ended NAS: Semi-Automated Design Knowledge Structuring with LLMs for Efficient Neural Architecture Search

AIRA-Compose: Agentic Discovery of Neural Architectures

Differentiable architecture search (DARTS) 2→3

Relaxes architecture choice to continuous weights, improving DARTS robustness, memory use, and analysis.

Symplectic Reck: In-Situ Learning of Gaussian Quantum Operations

Cool Graphs: Active Property Search Towards Quantum Nano-Refrigerators

Efficient NAS search strategies 2→2

Improves architecture search with credit assignment, distillation, and generative modeling of architectures.

Agentic Neural Architecture Search

ACDP: Architecture-aware Cross-Dataset Performance Predictor for NAS

Generative protein structure design 17→38

Generates protein structures and sequences using flow matching, equivariant models and AlphaFold.

NanoFold: Designing Reproducible Protein Structure Benchmarks through Principled Sampling

Few-Step Cofolding with All-Atom Flow Maps

Flow matching generative modeling 7→19

Develops flow matching theory and variants, including equivariant and spectral versions.

MaterialsSaddles: 34 Million Transition States and a Flow-Matching Saddle-Point Predictor for Materials

Drift-React: One-step Generation of Reaction Pathways via SE(3) Drifting Fields

Protein interface and docking prediction 10→15

Predicts interfaces, ligand docking and conformations with equivariant 3D structure networks.

ACER: Towards Generalizable Protein-ligand Co-folding

PepDDG: Peptide–Protein Binding ΔΔ𝐺 Prediction via Information Channel Decomposition

Structure-based 3D molecule generation 10→13

Generates drug-like molecules in protein pockets and extends scaffolds with 3D generative models.

Hit Expansion via Localized Exploration of Synthesizable Chemical Space

On the Selectivity of Generative Models in Structure-Based Drug Design

Diffusion theory and learned samplers 13→63

Provides variational and non-asymptotic views of diffusion and learns diffusions for sampling.

Path-independent Flow Matching for Multi-parameter Generative Dynamics

Inference-Time Self-Aligned Drifting for Few-Step Flow Matching

Posterior sampling with diffusion guidance 11→21

Samples posteriors or unnormalized targets using gradient flows, Stein guidance and amortized samplers.

BridgeTwist: Twisting Schrödinger Bridges for Training-Free Conditional Sampling

Learning Energy-Based Models from Stochastic Interpolants using Spatiotemporal Differences

Score-based generative modeling analysis 3→10

Estimates data gradients and analyzes bias, generalization and evaluation of generative models.

Generative Modeling via Drifting

Stein Transport for Generative Modeling

Variational inference theory and gradients 4→8

Gives guarantees and lower-variance gradient estimators for variational inference.

Symmetry Guarantees Statistic Recovery in Variational Inference

Denoising Implicit Variational Inference

State-space inference for stochastic systems 9→21

Learns filters and amortized inference for nonlinear stochastic dynamical systems.

Beyond Single-Step Likelihood: Gibbs Variational Last Layers for Long-Horizon Dynamics Learning

The Dynamic-Probabilistic Consistency Gap in Chaotic Surrogate Modeling

Data assimilation from sparse observations 6→15

Reconstructs physical dynamics from sparse measurements using generative assimilation.

Structural Blindness in Latent Data Assimilation: Representation Geometry Misleads Sensor Design

Rethinking Forward Processes for Score-Based Nonlinear Data Assimilation in High Dimensions

Physics-informed dynamics learning 7→13

Learns physical laws using Hamiltonian, Lagrangian and geometric neural dynamics models.

Discovering Phase Space Structure in Learned Hamiltonian Systems

Generalizable Physics Simulation through Compositional Energy Minimization

Latent-space control and planning 6→13

Plans and controls using learned latent dynamics, predictive coding and controlled SDEs.

Markovian Dynamics Enforcer: Feasibility Preserving Correction on Learned Dyanmics Manifolds

ControlJEPA: Principled Trajectory Regularization via Lyapunov Tube Loss

Constraint-based structure learning with latents 5→15

Recover causal graphs under latent variables and mixed data using recursive, constraint-based algorithms.

A Characterization of Latent Variable Causal Models Consistent with Observational Data

Coarsening Linear Non-Gaussian Causal Models with Cycles

Counterfactual identification and bounds 7→15

Characterize and bound counterfactual quantities from observational and experimental data.

Improving Causal Explanations

Transporting the Past: An Optimal Transport View of Backtracking Counterfactuals

Latent-variable causal estimation 5→14

Learn structural causal models and deconfounded estimates using generative and latent-variable models.

Decomposing Effects in Neural Causal Models

Observable Neural ODEs for Identifiable Causal Forecasting in Continuous Time

Causal representation identifiability 8→14

Prove when causal variables can be recovered from observations under interventions or weak supervision.

Towards Identifiable Latent Additive Noise Models

Causal Concept Explanations for Deep Neural Models

Neural operators for general geometries 8→27

Learns PDE solution operators on irregular domains using neural fields and latent operators.

AOT-POT: Adaptive Operator Transformation for Large-Scale PDE Pre-training

Adaptive Coordinate Transforms for Neural Operators

Neural PDE solvers 12→21

Solves PDEs with convergence guarantees, coarse correction and latent spectral models.

Geometry Meets Physics: Data-Efficient Pre-Training for Unstructured Neural PDE Solvers

Compute-optimal data scaling for neural surrogates via multi-fidelity training

Physics-informed operator learning 9→19

Embeds physical constraints in DeepONets and diffusion models for fluid and graph PDEs.

Stable Resolution-Invariant Emulation in Entropically Controlled Kinetic Methods

TurtLES: A Large-Scale Benchmark for Turbulent 3D Neural PDE Surrogates

Physics-informed neural network training 10→10

Improves PINN training through sequential, accelerated and competitive learning.

Per-Loss Adapters for Gradient Conflict in Physics-Informed Neural Networks

PINNeval: A Comprehensive Evaluation Standard for Physics-Informed Neural Networks

Variational Bayesian deep learning 5→12

Parameterizes and analyzes variational posteriors for deep models, linking ensembles to Bayesian inference.

Robust Amortized Simulation-Based Inference via Learned Error Models

Maxitive Donsker-Varadhan Formulation for Possibilistic Variational Inference

Epistemic versus aleatoric decomposition 5→10

Critically analyzes how to define and separate epistemic and aleatoric uncertainty.

probly: Uncertainty-Aware Machine Learning

The Aleatoric-Epistemic Dichotomy of Uncertainty is Meaningful and Indispensable for Machine Learning

Nonparametric and PAC-Bayes Bayesian learning 3→10

Develops Bayesian nonparametric models and links PAC-Bayes theory with Bayesian inference.

Bayesian Additive Distribution Regression

Scale-Invariant Empirical-Bayes Laplace Approximation for ReLU Networks

Simulation-based inference 4→8

Estimates posteriors for simulators using learned statistics and active sequential estimation.

Guidance For Prior Change via Density Ratio Estimation

Mesh Invaiant Infinite Dimensional Adaptive MCMC for Latent Gaussian Processes

Distributional treatment effects 4→11

Estimates distributional and kernel treatment effects with doubly robust tests.

Counterfactual Estimation under Composite Treatments via Progressive Distribution Alignment

Semiparametric Efficient Tests for Interpretable Distributional Treatment Effects

Latent-variable causal inference 9→7

Infers individual effects using deep latent-variable and structural-equation models.

Bayesian Causal Stress Testing: Posterior Fragility of Treatment-Effect Conclusions

Generalized Bayes for Causal Inference

Meta-learners and partial identification 5→6

Estimates conditional treatment effects with proximal learning and bounds under partial identification.

Assessing the robustness of heterogeneous treatment effects in survival analysis under informative censoring

Structure-Adaptive Estimation of Heterogeneous Treatment Effects with Kernel Methods

Benchmarking treatment effect models 3→5

Evaluates limits and representation learning methods for heterogeneous treatment effect estimation.

Distribution-First Framework for Learning Risk-Sensitive Individualized Treatment Rules

Decision-Aware Proximal Bridge Learning for Optimal Treatment Selection

Local and high-dimensional Bayesian optimization 6→13

Make Bayesian optimization scale with local search, structure, and analyses of high-dimensional behavior.

Bayesian Optimization with Fisher Information Geometry: Gradient Bounds and Trust-Region Methods

Automated Kernel Discovery Towards Understanding High-dimensional Bayesian Optimization

Acquisition function design 2→8

Develop new acquisition criteria and stopping rules for sequential Bayesian optimization.

BONSAI: Bayesian Optimization with Natural Simplicity and Interpretability

BO-Arena: An Evolving Benchmark for High-Dimensional Bayesian Optimisation

Offline black-box optimization with surrogates 4→5

Optimize designs from static datasets using learned surrogates, generative models, and language-model embeddings.

In-Context Black-Box Optimization with Unreliable Feedback

Posterior Inference in Latent Space for Scalable Constrained Black-box Optimization

Multi-objective Bayesian optimization 3→3

Approximate Pareto fronts with batch, diversity, and preference-aware acquisition.

Simple Extensions of Single-Objective Acquisition Functions and Hedge Strategies for Multi-Objective Bayesian Optimization

MOBO-CAPS: Multi-objective Bayesian Optimization with Cardinality-Aware Pareto Selection

Masked diffusion language model decoding 10→64

Accelerate and improve parallel decoding of masked diffusion language models.

Corrective Diffusion Language Models

Attention-Discounted Adaptive Sampler for Masked Diffusion Language Models

Diffusion theory and design space 18→56

Analyze convergence of score-based diffusion and optimize noise schedules and training design.

Diffusion Model's Generalization Can Be Characterized by Inductive Biases toward a Data-Dependent Ridge Manifold

Revisiting Diffusion Model Predictions Through Dimensionality

Training-free diffusion acceleration 29→31

Speed diffusion inference through caching, pruning, early exit, and adaptive step selection.

Dual-Rate Diffusion: Accelerating diffusion models with an interleaved heavy-light network

SARA: Step-Adaptive Rank Adjustment for Diffusion Inference Acceleration

Guidance for conditional diffusion 10→27

Improve classifier-free and rectified guidance for conditional generation.

Decompose the Distillation: Interpretable Single-Pass Guidance for Diffusion Models

Emergence of Distortions in High-Dimensional Guided Diffusion Models

High-resolution diffusion synthesis 13→20

Scale diffusion models to high resolutions with position extrapolation and efficient architectures.

PixelDiT2: Representation-Grounded Pixel Diffusion Transformers

UltraDiff:Transferring High-Fidelity Priors to Compressed Latent Spaces for High-resolution Image Generation

Latent diffusion with learned tokenizers 10→17

Apply diffusion and masked modeling in learned latent spaces for images and language.

End-to-End Training for Unified Tokenization and Latent Denoising

Masked Generative Pretraining Improves Cross-Dataset Transfer in Pixel-Space Diffusion

Generalizable fake image detection 11→13

Detect AI-generated images with large-scale datasets, multi-generator training, and semantic-pixel features.

MIRROR: Manifold Ideal Reference ReconstructOR for AI-Generated Image Detection

BIAS-ID: A Framework for Analyzing Transformation Biases in AI-Generated Image Detectors

Next-scale and token autoregressive images 8→12

Generate images autoregressively over scales or continuous tokens with improved tokenizers.

Context-Aware Autoregressive Image Generation for Emerging Reasoning Properties

Pixel-space Autoregressive Image Synthesis via Spectrum Serialization and Flow-based Refinement

Layout-to-image control 9→21

Control spatial arrangement with layouts, instance maps, and spatio-textual representations.

Imagine Before You Draw: Visual Prompt Engineering for Image Generation

CyCLeGen: Cycle-Consistent Layout Prediction and Image Generation

Compositional text-to-image guidance 10→19

Improve object composition in text-to-image diffusion through attention guidance and token-level supervision.

ORCA: Hunting Compositional Failures in Text-to-Image Diffusion

Concurrent Image Understanding and Generation: Self-Correcting Coupled Markov Jump Process

T2I evaluation and controllable adapters 9→18

Benchmark text-to-image models with human feedback and add controllable adapters.

Arena-T2I Hard: Benchmarking and Improving Faithfulness with Dependency-Aware Checklist Rewards

DynT2I-Eval: A Dynamic Evaluation Framework for Text-to-Image Models

Inference-time T2I improvement 11→13

Improve fidelity, safety, and diversity of text-to-image models through noise optimization and guidance.

Breaking the Static: Dynamic Text Conditioning for Diverse Image Generation

Steering Away from Memorization: Reachability-Constrained Reinforcement Learning for Text-to-Image Diffusion

Efficient restoration architectures 7→10

Design efficient, progressive, and frequency-based networks for general image restoration.

Learning Where and What to Restore for Composite Image Restoration

Continuous Expert Assembly: Instance-Conditioned Low-Rank Residuals for All-in-One Image Restoration

Image harmonization and color transformation 4→5

Harmonize composites and decompose illumination using pixel-wise color transforms.

PolarScale: A Physics-Grounded Benchmark for Radiometrically Consistent RGB-to-Stokes Estimation

Bricker to BRACE: A Bracket Exposure RAW Dataset and Restoration Model for Flicker-Banding

Pansharpening and spectral fusion 1→5

Fuse panchromatic, multispectral, and hyperspectral images with spatial-spectral networks.

Euler-Mamba: Learning Resolution-Invariant State Evolution on Polar Manifolds for Asymmetric Pansharpening

Manifold Prior Guided Deep Unfolding for Hyperspectral Image Reconstruction

Learned image compression 2→4

Compress images with learned hyperprior entropy models and lossless or lossy coding.

Bound-Conditioned Latent Inference for Progressive Image Compression

DC-Ocean: Deep Latent Compression for Global High-Resolution Ocean Forecasting

Instruction-based diffusion editing 16→16

Edit images from text instructions or exemplars using trained diffusion models.

Mission Impossible: Diagnosing and Fixing Non-Operative Instruction Following in Image Editing

Inline Critic Steers Image Editing

Texture and sketch synthesis 9→5

Synthesize stylized images, sketches, and textures with controllable disentanglement.

TopoRefine: Topology-Aware Correspondence and Residual Refinement for Training-Free Subject-Consistent Generation

Barycentric Guidance: Turning Foundation Image Editors into Continuous Affective Controllers

Localized and attribute-specific editing 4→5

Edit specific objects and facial attributes while preserving background and conditions.

Steering Fields: Adaptive Vector Fields for Safe Image Generation and Beyond

Text-Vision Co-Instructed Image Editing

Diffusion face generation and editing 1→4

Generate and edit identity-consistent faces using diffusion and 3D priors.

On the Feasibility of Identity Manipulation for Diffusion-Based Face Privacy Preservation

Non-Colliding Biometric Identities for Digital Entities: Geometry, Capacity, and Million-Scale Virtual Identity Provisioning

Perceptual and real-world SR 6→11

Improve realism and semantic accuracy of super-resolution with perceptual losses and semantic guidance.

NSARM: Next-Scale Autoregressive Modeling for Robust Real-World Image Super-Resolution

PRISM: Prior Rectification and Uncertainty-Aware Structure Modeling for Diffusion-Based Text Image Super-Resolution

Diffusion-based super-resolution 4→5

Accelerate diffusion super-resolution with one-step distillation and residual shifting.

When Latents Forget Pixels: Restoring Fidelity in Diffusion Transformer Super-Resolution

Efficient One-Step Diffusion Restoration Model with Compact Token Compression and Linear Attention

Efficient lightweight super-resolution 3→3

Reduce super-resolution cost with lightweight networks, selective processing, and mixtures of experts.

FluxFlow: Conservative Flow-Matching for Astronomical Image Super-Resolution

SR-Prominence: A Crowdsourced Protocol and Dataset Suite for Perceptually-Weighted Super-Resolution Artifact Evaluation

Stabilized GAN training 0→2

Improve GAN stability, convergence, and limited-data training with new objectives and discriminators.

Magnitude-preserving Layers Enable Efficient GANs

Improving Generative Adversarial Networks with Self-Distillation

Robust and efficient agent execution 11→88

Benchmarks tool-using agent robustness and speeds execution with plan caching and compilation.

EvoCodeBench: Evaluating Coding Agents in Multi-Turn Iterative Interactions

Rethinking Parallel Multi-Agent Systems: A Cost-Aware Framework for Efficient Coordination

Self-evolving language agent learning 13→85

Improves language agents via search, step-level value models, reinforcement learning, and self-distillation.

EvoMemBench: Benchmarking Agent Memory from a Self-Evolving Perspective

When Should Agents Remember? Falsification-Gated Self-Evolution for LLM Agents

Agentic benchmarks and orchestration 17→84

Benchmarks long-horizon agent memory, reliability, and orchestration of sub-agents.

AgentAbstain: Do LLM Agents Know When Not to Act?

UniClawBench: A Universal Benchmark for Proactive Agents on Real-World Tasks

LLM multi-agent decision making 19→68

Designs collaboration, hierarchy, and credit assignment for LLM-based multi-agent teams.

COOP$^2$: Defining, Observing, and Repairing Cooperation in LLM Multi-Agent Systems

Recursive Semantic Divergence for LLM Agent Consistency

Human-AI coordination agents 8→18

Trains agents that adapt to and collaborate with human partners, including in games like Hanabi.

Human–AI Collaboration Requires a High-Order Dynamic Abstraction Substrate

JobBench: Aligning Agent Work With Human Will

Embodied agent learning 10→11

Builds embodied robotic agents using vision-language distillation, affordance exploration, and generalist policies.

ManiTaskGen: A Comprehensive Task Generator for Benchmarking and Improving Vision-Language Agents on Embodied Decision-Making

Toward Embodied World Agents via Embodied-Planning Dataset and Interactive World Models

World-model-based agents 5→10

Builds learned or code-based world models for planning and generalist agent behavior.

Baba in Wonderland: Online Self-Supervised Dynamics Discovery for Executable World Models

Imagine, Don't Narrate: The Generative Bottleneck in World Models of Interaction Dynamics

Imitation and skill learning 2→6

Learns long-horizon skills and representations from demonstrations, instructions, and goal-reaching via world models.

ASH: Agents that Self-Hone via Embodied Learning

Self-Improving World Modelling with Latent Actions

Cooperative MARL algorithms 9→12

Proposes parameter-sharing, primal-dual, and benchmarked algorithms for cooperative multi-agent reinforcement learning.

Coordination Connectivity: Shared Initialization Shapes the Joint-Policy Landscape in MARL

Coordinating Hundreds of RL Agents through Scalable Inference-Time Search

MARL benchmarks and grouping 5→11

Introduces benchmarks and agent grouping or experience-sharing schemes for cooperative MARL.

Dr. MAS: Stable Reinforcement Learning for Multi-Agent LLM Systems

Self-Supervised Goal-Reaching Results in Multi-Agent Cooperation and Exploration

Offline and networked MARL 6→5

Learns multi-agent policies from fixed datasets or with networked, scalable value decomposition.

SamaDICE: Safe Multi-Agent Reinforcement Learning with Stationary Distribution Correction Estimation

Finite-Sample Convergence in Networked Average Reward MARL: Decentralization Pitfalls and Entropy Remedies

Learned inter-agent communication 1→4

Learns when, what, and how to communicate among agents under bandwidth and delay limits.

Adaptive Communication Range for Scalable Cooperative Multi-Agent Reinforcement Learning

From Local to Global: Progressive Consensus via Hierarchical Communication in Multi-Agent Reinforcement Learning

Cooperation and social dilemmas 3→9

Studies how learning agents cooperate or collude in dilemmas and sustain cooperation mechanisms.

Mitigating Retaliatory Algorithmic Collusion in Repeated Games

Training Generalizable Collaborative Agents via Strategic Risk Aversion

Learning in Markov games 5→8

Proves convergence and hardness of decentralized policy learning in zero-sum and general-sum Markov games.

Independent Learning of Nash Equilibria in Partially Observable Markov Potential Games with Decoupled Dynamics

NashDreamer: Model-Based Reinforcement Learning for Zero-Sum Imperfect-Information Games

Learning with strategic agents 2→4

Designs contracts, menus, and mechanisms for learning in principal-agent and strategic interactions.

Network Intervention by Polling Strategic Agents

Welfare, Improvability, and Variance: A Principal-Agent Approach to Optimal Benchmark Item Aggregation

Opponent modeling in competitive games 1→2

Shapes and adapts to opponents in two-player games via auto-curricula and strategy representations.

Differentiable Belief-based Opponent Shaping

The Attacker in the Mirror: Breaking Self-Consistency in Safety via Anchored Bipolicy Self-Play

RL for LLM reasoning 23→62

Apply reinforcement learning to improve multi-step reasoning in language models.

Don’t Let Gains FADE: Breaking Down Policy Gradient Weights in RL

DORA: A Scalable Asynchronous Reinforcement Learning System for Language Model Training

Scaling RL with verifiable rewards 11→40

Study scaling, exploration, and learning dynamics of RL with verifiable rewards.

Not Only Where, But When: Temporal Scheduling for RLVR

Reward Budgeting Reduces Premature Convergence in Reinforcement Learning for LLM Reasoning

GRPO stability and variants 10→31

Analyze and fix group-relative policy optimization issues such as noise, collapse, and trust regions.

Skill-Coupled Policy Optimization with Calibrated Group-Wise Advantage Estimation

TIC-GRPO: Provable and Efficient Optimization for Reinforcement Learning from Human Feedback

RL fine-tuning of generative models 11→27

Fine-tune diffusion and multimodal generative models with policy-gradient and reward objectives.

TMPO: Trajectory Matching Policy Optimization for Diverse and Efficient Diffusion Alignment

Beyond the Dirac Delta: Mitigating Diversity Collapse in Reinforcement Fine-Tuning for Image Generation

Model-based RL with guarantees 9→27

Analyze and design model-based RL algorithms with provable robustness and sample efficiency.

Foundation Pareto Flow Policy for Multi-Objective Reinforcement Learning

Likelihood-Free Generative Policy Optimization

Regularized scalable deep value learning 11→26

Stabilize deep value-based RL through regularization, overfitting control, and scaling.

Designing Reinforcement Learning for Diffusion Models: A Unified Path-Space View

What Kind of Diffusion Models Do We Need in Online Reinforcement Learning?

Safety layers and shielding 3→10

Add safety editors, shields, or constraint-conditioned policies to keep agents within safe behavior.

Sampling-Based Safe Reinforcement Learning

Feasible Policy Optimization for Safe Reinforcement Learning

Temporal-difference and discounting theory 8→9

Analyze TD learning, discount factors, and value estimation error for prediction and control.

Long-Horizon Q-Learning: Accurate Value Learning via n-Step Inequalities

Almost Sure Convergence of Linear Temporal Difference Learning with Arbitrary Features

Dexterous manipulation transfer 9→21

Transfer manipulation skills with residual learning, benchmarks, and cross-embodiment data.

Scaling Whole-Body Loco-Manipulation through Compositional Data Synthesis

Humanoid Horizon: Extending Task Horizon in Whole-Body Loco-Manipulation via Parallel Training, Dynamic Starting, and Reward Gating

In-context and adaptive RL 12→15

Enable fast adaptation in RL via in-context learning, memory, and domain transfer.

Hindsight Relabeling is All You Need for Reach-Avoid Learning

Reinforcement Learning Agents Are Swimmers

RL for combinatorial routing 10→10

Solve routing and NP-hard problems with learned RL policies and search.

Rectified Policy Rollouts with Hierarchical Expert Guidance for Neural Combinatorial Optimization

Beyond Prediction: Steering VLM Agents with Retrospective World Modeling

Learned state and action representations 4→10

Learn compact representations of states, actions, or contexts to speed up and generalize RL.

Representation Learning Enables Scalable Multitask Deep Reinforcement Learning

When Do Learned State Representations Break Sensitivity Analysis?

Offline RL sample complexity theory 3→21

Prove sample-complexity bounds for offline RL under limited data coverage.

Trajectory-Consistent Diffusion Policies for Offline Reinforcement Learning

From Static Policies to Adaptive Priors in Offline Reinforcement Learning

Simple practical offline RL 7→13

Improve offline RL with minimalist algorithms, reweighting, and symmetry-based data efficiency.

Jump Start Your Policy Learning with Lessons from 145,000 Training Runs

Variational Consequence-Driven Offline Reinforcement Learning

Offline-to-online fine-tuning 7→10

Pretrain on offline data then fine-tune online with calibration and balanced exploration.

Diffusion Subgoal Planning for Long-Horizon Offline Goal-Conditioned Reinforcement Learning

AdaptFlow: State-Anchored Goal Conditioning with Flow Matching for Offline Goal-Conditioned RL

Diffusion models for offline decision making 3→8

Use diffusion sampling and representation learning for offline and offline-to-online decisions.

Horizon Adaptive Offline Policy Learning via Value Stitching

Uncertainty-Guided Reward Labeling for Reinforcement Learning under Limited Feedback

Offline RL sample complexity 10→9

Prove near-optimal offline RL guarantees with pessimism and variance reduction.

The Type Theory of Stationary MDPs: Rare Events and Uncertainty Quantification

Offline-Online Reinforcement Learning for Linear Mixture MDPs

Average-reward and regularized MDPs 7→7

Develop theory for regularized, average-reward, and non-stationary MDPs.

Augmented Lagrangian Method for Last-Iterate Convergence for Constrained MDPs

Decision-Focused Learning in MDPs: An Occupancy Measure Approach

Robust MDP policy gradient 7→7

Optimize policies in robust MDPs with rectangular uncertainty.

On the Complexity of Discounted Robust MDPs with $L_p$ Uncertainty Sets

Revisiting Subgradient Dominance in Robust MDPs: Counterexamples, Hardness, and Sufficient Conditions

POMDP policy optimization 5→6

Solve POMDPs with Monte Carlo, transformers, and policy-based methods.

Multi-Environment POMDPs with Finite-Horizon Objectives

From Likelihood Convergence to Parameter Convergence in POMDPs

Offline imitation from imperfect data 4→7

Imitate from suboptimal or limited demonstrations using distribution correction and offline data.

Reconciling Safety and Performance via Dual-Expert Offline Imitation Learning

Higher-Order Action Supervision Makes A Strong Policy Class

Imitation combined with RL 6→7

Mix demonstrations with reinforcement learning, including hierarchical and self-correcting policies.

World Action Verifier: Self-Improving World Models via Forward-Inverse Asymmetry

LiFT: Likelihood-Free Tree-Structured Policy Optimization for Flow-Based VLAs

Behavior cloning theory and models 4→7

Analyze behavior cloning and build flexible imitative or foundation-model-based policies.

Learning When to Stop: Selective Imitation Learning Under Arbitrary Dynamics Shift

Behavior Cloning is Not All You Need: The Optimality of On-Policy Distillation for Noisy Expert Feedback

Theory of inverse RL 1→4

Give theoretical guarantees and exploration strategies for offline and active inverse reinforcement learning.

Loop-Free Inverse Reinforcement Learning via Sequential Value Recovery

Inverse Reinforcement Learning with Just Classification and a Few Regressions

Conformal prediction sets 8→31

Construct prediction sets with coverage guarantees and improved efficiency or conditional validity.

Multi-Variable Conformal Prediction: Optimizing Prediction Sets without Data Splitting

Generative Conformal Prediction with Optimized Coverage Allocation

Algorithms with explicit predictors 2→11

Model prediction error and predictor quality to analyze algorithms using ML predictions.

Testable and Actionable Calibration for Full Swap Regret

The Minimax Rate of Second-Order Calibration

Adaptive testing and experiments 9→11

Sequential decision and testing procedures for validity, treatment effects and instance selection.

What Survival Benchmarks Don’t Tell You: Impact of Model Selection and Dataset Regimes

Subdata Selection: A Unified Framework for Optimal Selection and Statistical Efficiency Assessment

Optimal decision tree learning 4→10

Compute optimal or sparse decision trees with algorithmic and generalization guarantees.

Multistage Defer Trees for Hybrid Interpretability: If at First You Can't Succeed, Tree Again

Compressing Collections of Trees with Decision Equivalence

Learning halfspaces under noise 7→10

Develops efficient and label-optimal algorithms for halfspaces under noise and misspecification.

Statistical Query Lower Bounds for Smoothed Agnostic Learning

Proper Agnostic Learning of Functions of Halfspaces

Sample complexity of learning distributions 5→9

Derives sample bounds for learning Gaussians, simplices and predictors, including gradient-based algorithms.

Optimal Dimension-Free Sampling for Regularized Classification

Tight $L_\infty$ Sample Complexity for Low-Degree and Sparse Boolean Polynomials

Optimal mean estimation and subsampling 4→7

Develops optimal mean estimators and subsampling or resampling procedures with sharp guarantees.

Statistical Limits of Affine Equivariant Estimation

Sub-Gaussian Confidence Intervals for Heavy-Tailed Data: Characterizing the Limits of Inference

Sample complexity of ranking and selection 4→7

Bounds samples for item selection, ranking, boosting and auction learning.

Learning Distributions from Multiple Data Providers

Majority-of-Three is an Optimal PAC Learner

Empirical generalization measures 2→6

Measure and predict generalization of deep networks through experiments and scaling laws.

Generalization Bounds for Neural Networks with Sparse Connectivity

Does Compression Imply Generalization? A Minimum Description Length Perspective

RKHS kernel methods 4→6

Analyze and accelerate kernel methods with Hilbert space theory, Nystrom and kernel thinning.

Random features for Grassmannian kernel approximation with bounded rank-one projections

Beyond Eigenfunctions: Divergence Principal Functions for Representation Learning

High-dimensional asymptotic generalization 2→6

Derive exact asymptotic error of models in proportional high-dimensional limits using random matrix tools.

Sharp feature-learning transitions and Bayes-optimal neural scaling laws in extensive-width networks

Learning with Synthetic Data via SGD in High-Dimensional Linear Regression

Kernel ridge regression analysis 2→6

Characterize learning curves and test error of kernel ridge regression.

Prediction-only distillation with optimal mixing in ridge-regularized linear and logistic regression

Pooling Versus Ensembling for Ridge Regression Under Covariate Shift

Multi-objective optimization 3→15

Finds Pareto solutions with learned partitions, preference-guided scalarization and evolutionary methods.

Balanced Multi-Task Learning from an Optimality-Gap Perspective

ParetoM$^3$: Learning on the Pareto Set under Preference Guidance via Min-Max-Min Optimization

Unsupervised neural combinatorial optimization 7→9

Learns solvers for combinatorial problems via unsupervised relaxation, divide-and-conquer and learned branching.

PACE: Partial-state Amortized Constraint Editing for Neural Combinatorial Optimization

SchedDiff: Diffusion-Based Priority Refinement for Job Shop Scheduling

Semidefinite relaxations for verification 1→7

Tightens neural network verification with SDP relaxations, bound propagation and branch-and-bound.

On Lipschitz Explosion in Deep Neural Networks with Normalization: Consequences for Optimization and Robustness

Certified but Private: Scalable Zero-Knowledge Proofs for the Formal Verification of Neural Networks

Constrained neural network optimization 3→6

Embeds linear or orthogonal constraints into network layers and analyzes constrained training and routing.

Neural Slack Variables for Shape Constraints

Soft-Radial Projection for Constrained End-to-End Learning

Low-rank approximation theory 6→12

Analyzes error and perturbation bounds for SVD and low-rank estimation of matrices.

Fast and Stable Gradient Approximation for Bilinear Forms of Hermitian Matrix Functions

Debiasing Random Oblique Projections for Subsampled OLS and Fast CUR in High Dimensions

Geometry of low-rank and SPD structure 2→9

Studies rank, random matrix geometry and positive-definite manifolds in network training and learning.

Free Decompression with Algebraic Spectral Curves

Manifold-weighted neural networks

Matrix completion and factorization 2→8

Recovers low-rank matrices from partial entries via alternating minimization and sparse regularization.

Improving the Optimization Landscape of Matrix Completion with $\epsilon$-close Surrogates

Robust Noisy Inductive Matrix Completion with Local Linear Convergence

Structured low-rank learning 4→6

Applies low-rank structure to ranking, fairness and probability matrix problems with guarantees.

Necessary but Not Sufficient: Spectral Tests for Value-Linear Attention Surrogates

Approaching I/O-optimality for Approximate Attention

Zero-sum game learning dynamics 6→8

Analyze complexity and learning dynamics for computing equilibria in two-player zero-sum games.

Payoff-Only Learning in Games via Best-Response Differential Inclusions

Stabilizing RL+Search for Imperfect-Information Extensive-Form Games

Fair division of indivisible goods 2→7

Study envy-free and efficient allocation of goods and chores with approximation guarantees.

Online Fair Division Meets Reordering Buffers

Epistemic Pairwise Maximin Share

Learned auction design 4→6

Learn revenue-optimal auction and pricing mechanisms from samples, including with deep networks.

Data Auctions for Retrieval Augmented Generation

Extending Myerson's Optimal Auctions to Correlated Bidders via Neural Network Interpolation

Committee voting rules 2→6

Analyze proportionality and welfare of voting rules and committee selection.

Computing Thiele Rules on Interval Elections and their Generalizations

Polynomial-Time Algorithm for Thiele Voting Rules with Voter Interval Preferences

Sinkhorn and entropic OT algorithms 4→12

Analyze fast solvers for entropic and unbalanced optimal transport.

A Provably Convergent and Practical Algorithm for Gromov–Wasserstein Optimal Transport

A Grid Efficient Transport Optimization Model for the Wasserstein Barycenter Problem

Transport map estimation 2→11

Estimate and regularize transport maps with statistical and low-rank guarantees.

Geometry-Aware Optimal Transport: Fast Intrinsic Dimension and Wasserstein Distance Estimation

A Biconvex Formulation for Stable Transport of Mixture Models with a Unique Solution

Sliced Wasserstein variants 2→8

Develop sliced and tree-sliced Wasserstein distances with faster approximations and kernels.

Optimizing Computational-Statistical Runtime for Wasserstein Distance Estimation

Tree-Sliced Orlicz Integral Probability Metric

Neural optimal transport 3→7

Learn transport maps and distributions with neural, generative and loss-based formulations.

Supervised Distributional Reduction via Optimal Transport and Dependence Maximization

Brenier Meets Adversarial Training: Optimal Transport Geometry for Robust Learning

Coresets and dynamic clustering 5→8

Use coresets and dynamic data structures for k-center and robust clustering.

Beyond Worst-Case Coreset Bounds for $k$-Clustering via Determinantal Sampling

Coresets for Clustering Using Noisy Comparisons and Few Distance Queries

Other 9→5

Individuals Matter: Improving Deep Multi-View Clustering via Explicit Single-View Enhancement

Robust Multi-view Clustering against Imperfect Information

Clustering approximation algorithms 3→4

Design provable algorithms for hierarchical and general clustering objectives, often via convex relaxation.

Hierarchical Agglomerative Clustering via Relaxed Representatives

Classification Fields: Arbitrarily Fine Recursive Hierarchical Clustering From Few Examples

k-means++ seeding analysis 1→4

Analyze and speed up k-means++ seeding and Lloyd-type algorithms.

A Batched Hartigan's $k$-Means Clustering

Mini-batch kernel $k$-means

Fairness-constrained classification 10→13

Learn classifiers under group or individual fairness constraints, with guarantees on fairness and accuracy.

An Information-theoretic Framework for Auditing Unfairness in Training Data

FairSplit: Decomposing the Embedding Space for Fair Classification

Fair allocation and selection 4→4

Design fair rankings, resource allocation and subset selection algorithms with provable fairness guarantees.

A Comprehensive View of Fairness through Distributional Stability

Intersectional Fairness via Mixed-Integer Optimization

Post-processing and transport for fairness 4→3

Adjust or transform classifiers, often via optimal transport, to remove discrimination while preserving utility.

Overcoming the Resolution Limit: Significance-Aware Regularization for Intersectional Fairness

Position: Fair Representations Cannot Hold What They Promise

Causal and policy fairness in decisions 1→2

Frames fairness in decision-making through causal analysis, preference-based notions and optimal fair policies.

Strategic Causal Policy Learning: Welfare, Safety, and Fairness Thresholds

The Panel Complexity of Sortition: Is 12 Angry Men Enough?

Dense features and foundation priors 23→33

Learns dense or tabular features using synthetic priors, hierarchical contrast and feature re-embedding.

TabBioMed: A Large-Scale Benchmark for Biomedical Tabular Learning

MorphoHELM: A Comprehensive Benchmark for Evaluating Representations for Microscopy-Based Morphology Assays

Contrastive representation theory 30→27

Analyzes unsupervised and contrastive representation learning, disentanglement and invariance.

Early Detection of Backbone Divergence via Neighborhood Structure in Learned Embeddings

Elastic Representations via Hyperbolic Geometry

Group-robust classification 12→13

Improves worst-group and subclass performance via distribution balancing, priors and invariant augmentation.

Model Cascades with Provable Per-Class Quality

Ensemble Selective Classification

Learning from unlabeled and positive-unlabeled data 7→13

Trains classifiers from positive-unlabeled or multiple unlabeled datasets with consistent risk estimators.

Generative Cross-Entropy: A Strictly Proper Loss for Data-Efficient Classification

Rank-Transformed Dissimilarity Profiles for High-Dimensional Classification

Invariance-based self-supervised learning 3→15

Learns invariant representations through pretext invariance and causal or functional regularization.

Complete or Sparse: A Tale of Two Identifiabilities

DINOv3

Self-supervised neuroimaging representations 9→9

Learns representations of brain signals and imaging with self-supervised objectives.

Scaling Neural Motor Decoding via Decoupled Behavioral Pretraining

Leveraging unlabelled data for generalizable neural population decoding

Data-efficient and robust self-supervision 8→7

Improves self-supervised training with data selection, geometric harmonization and defenses against attacks.

Breaking Noise Shortcuts in Self-Supervised Learning via Noise-Aligned View Generation

Generate in Reconstruction Space, Match in Semantic Space: Transport Geometry for One-Step Generation

Self-supervised speech and brain decoding 2→6

Uses wav2vec-style speech models to model and decode brain activity.

Reconstructing the Vocal Tract with Differentiable Acoustic Simulation

OpenWhistle: A Large-Scale Longitudinal Dataset and Benchmark of Bottlenose Dolphin Vocalizations

Rethinking distillation objectives 6→29

Revisits distillation losses via label smoothing, probability mass allocation and empirical analysis.

Distilling What Matters: Confidence-Aware Selective Distillation for Large Language Models

SEAD: Competence-Aware On-Policy Distillation via Entropy-Guided Supervision

Few-shot and metric distillation 1→7

Distills compact students with limited data using grafting, regression and metric learning.

Understanding Layer Patching in Model Size Interpolation

DeltaPrompts: Escaping the Zero-Delta Trap in Multimodal Distillation

Teacher properties and robustness 1→5

Studies what makes teachers effective, including adversarial, random and noisy-input teachers.

LARK: Learnability-Grounded Trajectory Selection for Efficient Reasoning Distillation

Subliminal Learning Is Steering Vector Distillation

Self and data-free distillation 1→4

Uses self-distillation, data-free settings, decoupled losses and ensemble distillation for efficient students.

Rethinking Dataset Distillation for Classification: Do Distilled Sets Outperform Coresets?

DD-Ranking: Rethinking the Evaluation of Dataset Distillation

Unified multi-class anomaly detection 2→12

Trains one anomaly detector across classes using reconstruction, cross-modal constraints and class-generalizable features.

Towards Explainable Industrial Anomaly Detection via Knowledge-Guided Latent Reasoning

InCLAD: A Continual Learning Benchmark for Industrial Visual Anomaly Detection

Anomaly detection with contaminated data 5→10

Handles noisy training data and distribution shifts using clustering, ensembles and outlier exposure.

Vendi Anomaly Scores for Efficient and Accurate Anomaly Detection

We Need to Rethink Benchmarking in Anomaly Detection

Spurious-aware OOD detection 6→9

Addresses spurious correlations and scaling in OOD detection with gradients and prototypes.

A Systematic Analysis of Out-of-Distribution Detection Under Representation and Training Paradigm Shifts

Is Complex Training Necessary for Long-Tailed OOD Detection? A Re-think from Feature Geometry

Provable and feature-based OOD detection 4→4

Designs OOD scores and outlier training with theoretical guarantees using typical features and prototypes.

Adaptive Covariance and Multi-Layer Alignment for Out-of-Distribution Detection

Strengthen Out-of-Distribution Detection via Adaptive Mahalanobis Gap

Conditional distribution alignment 9→7

Aligns conditional or disentangled distributions for domain adaptation, including universal and generalized shift.

Pseudo-Labeling for Unsupervised Domain Adaptation with Kernel GLMs

Language-Induced Priors for Domain Adaptation

Optimization-based domain generalization 4→6

Studies failure modes and gradient-guided or meta-learning methods for domain generalization.

Learning Informative Invariant Representations via Hierarchical Latent Decomposition

Beyond Average Flatness: Domain-wise Flatness for Domain Generalization

Geometric alignment and synthetic-to-real adaptation 4→4

Aligns geometry and statistics, simplifies images, and adapts for pose or anytime settings.

D-GAP: Improving Out-of-Domain Robustness via Dataset-Agnostic and Gradient-Guided Augmentation in Amplitude and Pixel Spaces

HIFC-IQA: Train-Free Cross-Domain Image Quality Assessment via Dual-Process Cognition

Statistical alignment and hypothesis testing 2→4

Applies covariance, kernel and mutual-information alignment and statistical tests to adaptation.

DETS: An Interval-Censored Evidential Sampling Framework for Cross-Domain Scientific Discovery

Uncertainty Aware SURE Transfer Learning for Classification Problems

Realistic and stable test-time adaptation 5→11

Tackles layer selection, long-horizon drift and sequential streams in practical test-time adaptation.

Controlling Temporal Pseudo-Label Marginals for Stable Online Test-Time Adaptation

Test-Time Learning with an Evolving Library

Pseudo-label test-time adaptation 9→7

Uses pseudo-labels, feedback and proxy data for robust adaptation on noisy streams.

Efficient Test-time Adaptation through Candidate Verification and Divergence Shifts

Training-Free Active Test-Time Adaptation for Vision-Language Models

Test-time adaptation of vision-language models 6→5

Adapts CLIP-style models at inference using caches, diffusion and weight averaging.

Complementary Cache Guidance with Gradient Disentanglement for Continuous Test-Time Adaptation

DEDCA: Test-Time Adaptation for Generalized AI-Generated Image Detection

Robust optimization for distribution shift 2→2

Trains for robustness via risk extrapolation, selective augmentation and adaptive risk minimization.

Dynamic Context Modeling for Longitudinal Mental Health Monitoring under Distribution Shift

"What is Different Between These Datasets?" A Framework for Explaining Data Distribution Shifts

Zero-shot robustness and transfer 7→13

Studies and improves generalization of foundation models to new groups, domains and data.

NOCE-Net: Representation Learning for Battery Operational Context via Nested Sequence Modelling

Learning Transferable Cross-Day Representations for Few-Shot Neural Decoding

Robust fine-tuning of zero-shot models 1→4

Improves robustness and prompt ensembles of zero-shot models, with theory for zero-shot prediction.

A Minimal Interpretable Architecture for Zero-Shot Reconstruction of Dynamical Systems

Hyperspherical Local Margin Retraction for Zero-Shot Instance-Wise Machine Unlearning

Transductive and semantic-dictionary zero-shot 5→2

Uses unlabeled test data and learned semantic dictionaries for zero-shot and CLIP transfer.

Transcoda: End-to-End Zero-Shot Optical Music Recognition via Data-Centric Synthetic Training

WildBox: A Dataset and Benchmark for Aerial Monocular 3D Detection of African Savanna Wildlife

Fine-grained and cross-domain few-shot baselines 0→2

Uses baselines, feature reconstruction and prototypical refinement for fine-grained or cross-domain few-shot tasks.

DDGE: Disentangled Dirichlet Geodesic Evaluation for Robust Few-Shot Learning

Intervention-Guided Image-Free Classifier Expansion for Fine-Grained Recognition

Noise-aware loss and pairwise methods 3→5

Reduces noise effects through pairwise similarity, open-set noise and noise detection.

GradTrack: Detecting Noisy Labels via Temporal Trajectories of Class-wise Gradient Misalignment

Dual-Granularity Learning for Regression with Continuous Noisy Labels

Loss correction and noise robustness theory 7→3

Corrects losses and analyzes how overfitting and subclass structure affect noisy training.

Fisher information and the geometry of memorization in neural networks

Not All Instances Should Contribute Equally: Sparse Support Grounding via Unbalanced Transport for Heterogeneous MIL

Contrastive learning on noisy data 0→3

Applies contrastive and diffusion-based training to noisy, imbalanced or weakly labeled recognition.

Robust-by-Design Distributional Learning from Contaminated Samples

Recovering Clean Evaluation Metrics from Contaminated Benchmarks

Noisy cross-modal hashing and multi-label 1→3

Handles noisy labels in cross-modal retrieval and multi-label classification via robust hashing and loss handling.

Memorization Is Folding: Topological Signatures of Noisy-Label Learning

Unveiling Memorization–Generalization Coexistence: A Case Study on Arithmetic Tasks with Label Noise

Confidence-based pseudo-label selection 9→7

Selects reliable pseudo-labels using uncertainty, local variance and bias-adaptive classifiers.

AdapMatch: Adaptive Bias Decoupling for Semi-Supervised Partial Label Learning under Unknown Class Distributions

Trajectory-Matching Meta Pseudo-Labeling for Semi-Supervised Learning

Semi-supervised medical segmentation 3→2

Combines foundation models, density-aware contrast and variance analysis for medical image segmentation.

Beyond Stabilization: Dual-EMA Teachers for Global–Local Semantic Learning in Semi-Supervised Medical Image Segmentation

Parametric Meets Non-parametric: Bridging Dual Prediction Spaces for Semi-supervised Medical Image Segmentation

Efficient video diffusion models 18→52

Builds video diffusion models with latent decomposition, autoregressive distillation and training-free long-video generation.

Dynamics-Aware Sparse Attention for Efficient Autoregressive Video Diffusion

VideoMLA: Low-Rank Latent KV Cache for Minute-Scale Autoregressive Video Diffusion

Trajectory-guided motion control 16→44

Controls video generation motion with sparse trajectories, latent trajectory guidance and tracking signals.

TKCAM: Text and Keyframe to Camera Trajectory Generation

DynaTokens: Teaching Dynamics to Camera-Controlled Video Models at Test Time

Autoregressive long video generation 10→34

Generates long videos autoregressively with continuous tokens, propagation and rich perception, plus evaluation suites.

Learning-to-Memorize: Dynamic Context Management for Long-Horizon Autoregressive Video Generation

PlasticMem: Adding Temporal Reasoning to Diffusions for Consistent Long Video Generation

World-model video generation 12→33

Uses video generators as interactive world simulators and benchmarks their physical consistency.

LongScape: Advancing Long-Horizon Embodied World Models with Context-Aware MoE

EA-WM: Event-Aware Generative World Model with Structured Kinematic-to-Visual Action Fields

Long video understanding 18→62

Handles hour-long videos using memory, key-frame reasoning and training-free or streaming approaches.

Elephant in the Fridge: Constant-Memory Frame Packing for Long Video Understanding

TeMPO: Frame-Causal Token Compression for Efficient Video Large Language Model

Event and anomaly reasoning in video 17→26

Reasons about events, anomalies and object state changes in video with datasets and grounded reasoning.

CollabVR: Collaborative Video Reasoning with Vision-Language and Video Generation Models

OpenCoF: Learning to Reason Through Video Generation

Video LLMs with fine-grained understanding 22→23

Builds video language models for object-level and temporal understanding, with memory and benchmarks.

Reinforced Evidence-Aware Long Video Understanding

BoxTuning: Object-Aware Visual Prompting for Multimodal Model Fine-Tuning

Video question answering 2→8

Answers questions about videos using memory prompting, sampling, tool use and compositional benchmarks.

AdaAlloc: Adaptive Visual Token Allocation for Long-Video Question Answering

FrameRouter: Frame Budget Routing for Long-Video Understanding on Video-MME-v2 and Beyond

Layered video decomposition and dynamics 6→6

Decomposes videos into layers, objects and effects to model scene dynamics and generate videos.

Learning to Correct Geometry in Generated Videos

Can 4D Foundation Models Remember?

Self-supervised video representation learning 3→6

Learns video features from temporal self-supervision, correspondence and reconstruction without labels.

TTVidT: Decoupling the Temporal Axis for Efficient Motion-Centric Video Pretraining

V1-Inspired Dynamic Vision System: A Bio-Plausible Video Embedding Framework with Decoupled Shape–Color Pathways and Long-Range Spatiotemporal Perception

Stochastic video prediction 5→5

Predicts future frames with learned dynamics, hierarchical models and example guidance.

Rethinking the State Update Gate for Long-Sequence Recurrent 3D Reconstruction

HIST3R: Rectified State Decomposition from History for Streaming 3D Reconstruction

Matching-based video object segmentation 2→5

Propagates object masks via feature matching, instance queries and efficient linear attention.

Beyond Token Representations: Explicit Visual Object Grounding for Video Reasoning Segmentation

GMOS: Grounding Moving Object Segmentation in 3D Space and Time

Action anticipation and latent actions 2→11

Predicts future actions with LLMs, graph growth and latent action world models.

Generalizing Action-Conditioned Latent World Models with Video Model Rewards

Perception for Action in Latent World Models

Disentangled action recognition 5→6

Recognizes actions by decomposing dynamics, components and human-object interactions.

DiA: Directional Adapter

TrAction: Action Recognition with Sparse Trajectories

Temporal action segmentation 1→2

Segments activities in untrimmed videos with unsupervised, weakly or semi-supervised clustering and transformers.

EpiStream: Utility-Aware Temporal Abstraction for Dense-Action Streams

SAX: Advancing Video Diffusion Models for Sequential Action Execution

Fully supervised temporal action detection 1→2

Detects action boundaries using temporal convolution, multi-task recycling and context aggregation.

Adaptive Residual Quantization for Memory-Efficient Temporal Action Segmentation

HyrCap: Hybrid Rank-Calibration of Action Proposals for Temporal Event Understanding

Hand-object interaction generation 3→17

Generates 3D hand-object and grasp motion with contact guidance and diffusion models.

FlowHOI: Flow-based Semantics-Grounded Generation of Hand-Object Interactions for Dexterous Robot Manipulation

Towards Scalable Egocentric HOI for Humanoids: Benchmarking Whole-Body Dexterous Interaction with Tactile Prediction

Learning 3D reconstruction from images 13→15

Reconstructs 3D structure from single or multiple views with learned priors and foundation-model features.

Horizon-Stream: Long-Horizon Attention for Streaming 3D Reconstruction

TurboVGGT: Fast Visual Geometry Reconstruction with Adaptive Alternating Attention

Diffusion-based novel view synthesis 8→14

Generates consistent novel views from one or few images using multi-view and video diffusion priors.

NAMVIS: Next-Scale Autoregressive Multi-View Image Synthesis

GenRec: Knowing Where to Reconstruct and Where to Generate

Generative indoor scene synthesis 10→11

Generates 3D indoor layouts using scene graphs, diffusion, grammars and language-driven spatial reasoning.

HoloCode: A Code-Centric Multi-Agent Framework for Image-to-3D Scene Generation

WorldAct: Activating Monolithic 3D Worlds into Interactive-Ready Object-Centric Scenes

Compact and efficient Gaussian splatting 17→44

Prunes, sparsifies and optimizes Gaussian primitives for faster, smaller 3DGS rendering.

Provable Pruning for Efficient 3D Gaussian Splatting via Coresets

ReorgGS: Equivalent Distribution Reorganization for 3D Gaussian Splatting

Regularization and primitive design for 3DGS 20→16

Improves splatting quality through regularization, new kernels and rendering optimizations.

Diffeomoprhism-Informed 3D Gaussian Splattings via Screen-Space Optimal Transport (OT)

Deformable 2D Gaussian Splatting

Sparse-view Gaussian splatting for novel views 9→11

Uses depth, epipolar and view priors for generalizable novel view synthesis from few images.

StereoSplat: Metric-Scale Novel View Synthesis via Stereo-Grounded Gaussian Splatting

FRUC: Feedforward Dynamic Scene Reconstruction from Uncalibrated Collaborative Driving Views

Deformable 4D Gaussian splatting 13→9

Models dynamic scenes with deformable or explicit motion Gaussians for real-time rendering.

No Pose, No Problem in 4D: Feed-Forward Dynamic Gaussians from Unposed Multi-View Videos

GGQR: Gaussian-Grounded Query Refinement for Feed-Forward 4D Gaussian Splatting

Feed-forward 3D reconstruction 4→12

Reconstructs scenes and structure-from-motion in a single pass with learned priors, planes and dense matching.

FlashSSM-3D: A Distilled State Space Model for Lightning-Fast Dense 3D Reconstruction

Glob3R: Global Structure-from-Motion with 3D Foundation Models

Scene coordinate regression for localization 3→10

Localizes cameras by regressing scene coordinates or maps, with voting and 3D surface prediction.

LINK: Learning to Localize from Known to Unknown Scenes

Trustworthy and Efficient Map-free LiDAR Localization via Scan-Pose Alignment and Flow Matching

Foundation models for monocular geometry 6→10

Estimates geometry from single images with inductive biases, open-domain training and synthetic-real adaptation.

$R^3$: 3D Reconstruction via Relative Regression

Less Decoder is More Encoder: Geometric Representation Learning from Novel View Synthesis

Cost-volume multi-view stereo 1→9

Infers depth from multiple views with learned cost volumes, recurrent regularization and geometry-aware networks.

Understanding Multi-View Transformers

Déjà View: Looping Transformers for Multi-View 3D Reconstruction

Implicit neural representation design 3→8

Studies architectures, spectral bias and Fourier features for implicit neural representations.

ProPolar: Progressive Polar Decomposition for Implicit Neural Representations

Closed-Form Implicit Neural Representations

Textured mesh generation 4→7

Generates textured and PBR 3D meshes with generative models and sparse high-resolution representations.

MeshFIM: Local Low-Poly Mesh Editing via Fill-in-the-Middle Autoregressive Generation

Mesh BDF: Barycentric Dominance Field for 3D Native Mesh Generation

3D shape generative modeling 1→5

Generates 3D shapes with deep generative models, latent grids, primitives and text conditioning.

SILSA: Sliding-Window Slice Latents for Topology-Preserving High-Resolution 3D Generation

Flow-Corrected Shape Optimization: Taming Manifold Drift in High-Dimensional 3D Models

Neural implicit surface reconstruction 5→5

Reconstructs surfaces from multi-view images with signed or unsigned distance fields and volume rendering.

Neural Refraction Fields for Image Verification

Pygmalion Effect in Vision: Image-to-Clay Translation for Reflective Geometry Reconstruction

Physics and sensor-based motion capture 3→9

Recovers human motion from video or inertial sensors using physical constraints and object interaction.

EgoMo3R: Joint Egocentric Motion and Scene Reconstruction

Ego-HMB: Human Motion Bridging from Egocentric Images via Motion Bridge Diffusion Model

Category-level object pose estimation 2→5

Estimates object pose for unseen instances within categories using correspondence learning and universal features.

SchemaPose: RGB-Based Category-Level Pose Estimation with Parametric Category Schema

F2G-Pose: Geometry-Aware Foundation Feature Lifting for Direct RGB-D Category-Level Object Pose Estimation

Human mesh recovery from images 0→4

Recovers 3D body meshes end-to-end with pose calibration, point guidance and diffusion.

FactorizedHMR: A Hybrid Framework for Video Human Mesh Recovery

EIHMR: Collaborative Human-Camera Estimation for Global Human Mesh Recovery

Depth-based hand pose regression 2→4

Estimates 3D hand pose from depth or RGB images using CNN regression, dense prediction and neural rendering.

ComPose: When to Trust Hands for Object Pose Tracking

HOPE: Hand-Object Pressure Estimation from Monocular Videos

Inverse rendering with neural fields 4→5

Recovers geometry, reflectance and lighting from multi-view images for relighting and view synthesis.

TTB: Test-time MLP Baking for Efficient Rendering of Decoder-only View Synthesis Models

Resolving Representation Ambiguity in Feedforward Novel View Synthesis Transformer via Semantic-Spatial Decoupling

Sparse-input NeRF regularization 2→3

Improves NeRF view synthesis with geometry, depth and ray priors, and in-the-wild robustness.

RASS: Risk-Audited Budget Selection for Compact NeRF Benchmark Subsets

Spik-NeRF v2: Pushing the Limit of Spiking Neural Radiance Fields with $ \pm $I-LIF

Diffusion-based graph generation 9→25

Generate graphs with discrete and score-based diffusion and flow matching.

When Edge Independence Fails: Joint Graph Diffusion with Latent Sociability Priors

Hypergraph Generation with Latent Diffusion

Semi-supervised node representation learning 12→14

Learn node embeddings and classifiers on graphs with GCNs and inductive methods.

Invariant Hyperbolic Unfolding: Radial Canonicalization for Label-Free Cross-Graph Link Prediction

Hyperbolic Graph Neural Networks Under the Microscope: The Role of Geometry–Task Alignment

Graph anomaly detection 4→10

Detect anomalous nodes and graphs with generalist, generative, and diffusion-based methods.

Graph Anomaly Detection as Dynamical Transport: Training-Free Scoring via Empirical Bayes

Towards Generalist Graph-Level Anomaly Detection

Graph structure learning 7→10

Learn or refine graph structure to improve node classification.

When Graph Structure Provably Helps Classification: Non-Asymptotic Recovery Guarantees

Information Loss and Disparate Effects in Network Embeddings

GNN generalization theory 10→14

Analyze size generalization and statistical properties of GNNs, with benchmarks.

Distilling Graph Geometry: Knowledge Gap from GNNs to MLPs

From Link Prediction to Linear Community Detection: A Finite-Sample Guarantee for Graph Pretraining

Expressive equivariant message passing 10→11

Analyze limits of message passing and build more powerful equivariant or subgraph GNNs.

Beyond Coordinates: Encoding Graph Structure via Contextual Distribution and Relational Similarity

Graph Cascades: Contagion-Based Mesoscopic Rewiring for Structure-Aware Graph Machine Learning

Physics-inspired GNNs 2→9

Apply equivariant and diffusion-based GNNs to physical simulation.

Beyond Local Neighborhoods: Fractional Diffusion with Levy Flights on Simplicial Complexes for Link Prediction

Oversmoothing as Representation Degeneracy in Neural Sheaf Diffusion

GNN explainability 4→8

Evaluate and generate explanations for GNN predictions.

Robust Surrogate Modeling for Explainable Graph Neural Networks

The Graph Concept Bottleneck: Decoding Combinatorial Reasoning in GNNs for Interpretability

Continuous-time dynamic graph learning 10→16

Model link prediction on continuous-time dynamic graphs and benchmark temporal graph learning.

S3Former: Sequential, Structural, and Statistical Fusion for Continuous-Time Dynamic Graphs

Hierarchical Regime-Conditioned Dynamics for Spatiotemporal Graphs

Spatio-temporal graph traffic forecasting 0→6

Forecast traffic with spatio-temporal graph neural networks.

Learning Continuously Evolving Spatio-Temporal Explanations for Traffic Flow Forecasting

ExoC2T: An exogenous-driven spatio-temporal learning framework for cross-city transfer

Social and behavior graphs 3→5

Model social interactions with graphs for trajectory and behavior prediction.

Querying Counterfactuals on Tissue Graphs with Supervised Disentanglement

NicheIB: Targeted Spatial Bottlenecks for Predictive Microenvironment Discovery

Temporal graph neural architectures 7→4

Design state-space, motif-based, and spiking architectures for dynamic graphs.

A Topological Encoder Decoder Framework for Temporal Graph Learning

Weisfeiler and Leman Follow the Arrow of Time: Expressive Power of Message Passing in Temporal Event Graphs

Neural combinatorial optimization on graphs 4→11

Guide combinatorial and integer-programming solvers with graph neural networks.

AT-SKM-Net: An Accelerated Trainable Sampling Kaczmarz-Motzkin Framework for Linear Hard-Constraint Feasibility on Dynamic Graphs

Structurally Separated DAG Learning with Multi-Scale Normalized Closure

Graph comparison and random features 6→7

Compare graphs with optimal transport and efficient graph random features.

Query-Limited Community Recovery in Stochastic Block Models

Prune, Don’t Rebuild: Efficiently Tuning $\alpha$-Reachable Graphs for Nearest Neighbor Search

Continuous DAG structure learning 5→6

Learn DAG structure via continuous optimization and active experimental design.

Graph Distance Based on Cause-Effect Estimands with Latents

A Hierarchy of Entropy-Shapley Games for Multivariate Predictive Uncertainty

Neural and local graph clustering 5→4

Cluster graphs with neural modularity, flow diffusion, and correlation clustering.

A Graph Foundation Model for Unified Clustering

Do Graph Neural Networks Learn Generalizable Algorithms for Clustering Graphs?

Stochastic gradient last-iterate analysis 7→19

Analyzes convergence of SGD and adaptive stepsizes under noise, interpolation, and nonsmooth settings.

In-Expectation Convergence of Stochastic Gradient Methods under Heavy-Tailed Noise

Normalized SGD in the Convex Regime: First High-Probability Guarantees and Momentum Extension

Coordinate and Adam convergence analysis 8→16

Proves convergence under Lojasiewicz and nonuniform smoothness for coordinate descent, Adam, and steepest descent.

Mirror Descent Beyond Euclidean Stability: An Exponential Separation in Initialization Sensitivity

Follow the Regularized Leader Does Not Converge in Constrained Optimization

Nonconvex minimax and constrained methods 6→11

Proposes first-order and gradient-free algorithms for nonsmooth, constrained, and minimax nonconvex problems.

On the Parallel Optimality of Exponentiated Gradient Descent

Min-Max Optimization Requires Exponentially Many Queries

Stochastic convex optimization complexity 5→11

Establishes sample and oracle complexity of gradient methods for stochastic convex problems.

Stochastic Zeroth-Order Optimization Under Heavy-Tailed Noise

Stochastic Optimization with Random Search

Gradient flow and mean-field dynamics 8→12

Studies training dynamics of wide networks via gradient flow, mean-field limits, and teacher-student models.

Feature learning in high-dimensions under structured covariance: Scaling laws in quadratic networks

Gradient Descent on Two ReLU Neurons: Global Landscape and Bifurcation Dynamics

Shallow ReLU network convergence 5→10

Proves gradient descent convergence and generalization for shallow and deep ReLU or linear networks.

Mildly Overparameterized ReLU Networks on Orthogonal Data: Incremental Learning and Implicit Bias

Convergence of Near-Linear Width ReLU Networks with Unbalanced Initialization

Edge-of-stability dynamics 14→10

Characterizes gradient descent at large step sizes where sharpness hovers at the stability threshold.

Classification at the Edge of Stability: Unifying Self-Stabilization and Convergence Rates

Edge of Stochastic Stability: Revisiting the Edge of Stability for SGD

Neural tangent kernel analysis 6→6

Uses NTK and linearization theory to analyze training, feature learning, and inductive bias in wide networks.

How does feature learning change the function space evolution?

The Weight Gram Matrix Captures Sequential Feature Linearization in Deep Networks

Adam convergence and adaptive learning rates 4→12

Analyzes Adam-family convergence and adapts learning rates using hypergradients.

Towards Understanding the Power and Limits of the Muon Optimizer: A River-Valley Perspective

Why Muon Outperforms Adam: A Curvature Perspective

Natural gradient and Fisher approximations 3→8

Develops and critiques Kronecker-factored and structured approximations of natural gradient and second-order updates.

Matrix-Free Natural Gradient via Stochastic Truncated Metrics

Calibrated Target Noise Recovers Curvature from the Gradients

Memory-efficient LLM optimizers 8→8

Designs large-batch, low-memory, and normalized optimizers for training large language models.

Sharp Capacity Scaling of Spectral Optimizers in Learning Associative Memory

Sign-Based Optimizers Are Effective Under Heavy-Tailed Noise

Differentiating through embedded optimization 10→6

Backpropagates through optimization layers and learns optimizers or proximal operators.

A Differentiable Interior-Point Method in Single Precision

Unbiased First-Order Randomized Smoothing for Differentiable Simulation

Local and asynchronous SGD 4→6

Analyze asynchronous, local, and minibatch SGD for communication-efficient distributed training.

Do We Need Asynchronous SGD? On the Near-Optimality of Synchronous Methods

Rescaled Asynchronous SGD: Optimal Distributed Optimization under Data and System Heterogeneity

Other 0→5

LOSCAR-SGD: Local SGD with Communication-Computation Overlap and Delay-Corrected Sparse Model Averaging

Can Model Merging Improve Aggregation in DiLoCo?

Byzantine-robust learning 0→3

Defend distributed training against adversarial workers with robust aggregation.

From Inexact Gradients to Byzantine Robustness: Acceleration and Optimization under Similarity

RESIST: Resilient Decentralized Learning Using Consensus Gradient Descent

Decentralized optimization algorithms 0→2

Design and analyze decentralized and distributed optimization methods with faster convergence.

Faster Rates For Federated Variational Inequalities

Stochastic Matching via Local Sparsification

Online learning with feedback graphs 9→22

Design best-of-both-worlds experts and bandit algorithms with structured feedback and limited information.

From Average Sensitivity to Small-Loss Regret Bounds under Random-Order Model

Adaptive Realizable Regression from Two Online Learners

Learning for pricing and auctions 11→13

Develop regret-minimizing pricing and bidding algorithms for repeated auctions with strategic buyers.

Learning to Bid in Repeated Second-Price Auctions with Dynamic Values and Aggregated Feedback

The Power of Menus in Dynamic Pricing: Near Optimal Regret and Equivalence Results

Online decision-making with feedback structure 6→11

Analyze online prediction under delayed, bandit, strategic, or dynamical feedback.

The Stability of Online Algorithms in Performative Prediction

Online Decision-Focused Learning under Semi-Bandit Feedback

Constrained and unconstrained OCO algorithms 7→10

Design online convex optimization algorithms handling constraints, unbounded domains, and switching costs.

From Non-Convex to Strongly Convex: Curvature-Adaptive FTPL for Online Optimization

Capacity-Constrained Online Convex Optimization with Delayed Feedback

Optimal stochastic bandit algorithms 9→13

Characterize regret-optimal algorithms for stochastic, batched, and combinatorial bandits.

Nash Social Welfare for Multi Armed Bandits: Trajectory-wise Expected and High Probability Regret

Near-Optimal Learning in Parametric Bandits with Action-Dependent Coarsened Feedback

Best-arm identification in linear bandits 3→7

Identify the best arm with optimal sample complexity for linear and fixed-budget settings.

Best Arm Identification in Generalized Linear Bandits via Hybrid Feedback

On Minimizing Regret in Fixed-Confidence $\varepsilon$-Best Arm Identification

Pure exploration algorithms 1→7

Design efficient pure-exploration and thresholding algorithms with batch and combinatorial structure.

Instance-Dependent Bandit Convex Optimization in One Dimension

On the Optimal Sample Complexity of Offline Multi-Armed Bandits with KL Regularization

Fixed-confidence versus fixed-budget identification 0→7

Develop optimal elimination and top-two algorithms for confidence and budget settings.

Does 1/2-Tsallis-INF Also Work Well for Best-Arm Identification?

Sharp Non-Asymptotic Analysis of the Penalized Challenger in $\beta$-EB-TCI for Bernoulli Bandits

Oracle-efficient contextual bandits 8→15

Build practical contextual bandit algorithms using regression oracles and handling misspecification or privacy.

Active Context Selection Improves Simple Regret in Contextual Bandits

Probabilistic Guarantees for Adversarial Linear Contextual Bandits

Linear and generalized linear bandits 3→7

Derive improved regret bounds for stochastic linear bandits with efficient updates.

Regret minimization in Linear Bandits with offline data via extended D-optimal exploration.

Logistic Bandits with $\tilde{O}(\sqrt{dT})$ Regret without Context Diversity Assumptions

Causal and structured bandits 3→6

Exploit causal graphs and auxiliary feedback to learn good interventions.

Multi-Objective Causal Bandits: Minimal Intervention Space and Policy-Level Learning

Learning to target with network interference

Ranking bandits 2→4

Learn to rank online from click feedback with unimodal and semi-bandit methods.

Learning from Ranking Feedback: Improved Regret Bounds via Independence Preserving Rank Breaking

Learning with Multiple Correct Answers - Regret Bounds under Different Feedback Models

Average-reward and latent MDPs 2→5

Derive regret guarantees for constrained, average-reward, factored, and latent MDPs.

Towards instance-dependent regret optimality in Episodic MDPs with Posterior Sampling

Learning Weakly Communicating Average-Reward CMDPs: Strong Duality and Improved Regret

Linear function approximation in RL 1→3

Establish efficient optimistic algorithms for RL with linear and generalized linear features.

Information-Directed Offline-to-Online Reinforcement Learning

Reinforcement Learning with Multi-Step Lookahead Information Via Adaptive Batching

Non-stationary and structured MDPs 0→2

Learn unknown, non-stationary, or structured MDPs with provable efficiency.

DARLING: Detection Augmented Reinforcement Learning with Non-Stationary Guarantees

Informed Posterior Sampling: More Efficient Online Learning with Few Offline Demonstrations in Average-Reward MDPs

Regret bounds with function approximation 3→2

Prove instance-dependent and low-switching regret bounds for RL with function approximation.

Lyapunov-Driven Optimistic Learning for Online Scheduling with Multi-Stage Tasks

Zero-Violation Regret for Cooperative Markov Games with Coupled Instantaneous Hard Constraints

Probabilistic and conformal forecasting 2→15

Produce calibrated probabilistic or conformal forecasts for series, including hierarchical and non-stationary data.

Dynamic Dual-Feedback Conformal Inference for Time Series Forecasting

TimeES: Probabilistic and Deterministic Time Series Forecasting via Evolutionary Spectra

Diffusion models for time series 4→11

Generate and forecast time series using conditional diffusion models.

ToLD: Efficient Time Series Forecasting via Tokenized Truncated Latent Diffusion

Bridging Diffusion and Autoregression for Flexible Time Series Synthesis

Dynamical system reconstruction 2→10

Learn latent nonlinear or linear dynamical systems from time series data.

Topological Periodicity Test (TopPT) via Confidence Bound of Time-Delay Embeddings

Time series analysis with Gumbel dynamics

Causal discovery from time series 2→7

Recover causal structure from temporal data under non-stationarity, latent confounding and Granger-style assumptions.

Hyper Hawkes Processes: Interpretable Models of Marked Temporal Point Processes

Flexible Intensities Matter: A comprehensive re-evaluation of Classical and Neural Temporal Point Processes

Time series foundation models 8→17

Adapt pretrained time-series models for zero-shot forecasting with in-context tuning and adapters.

Kairos: Toward Adaptive and Parameter-Efficient Time Series Foundation Models

Learning the Context of Errors: Black-Box Online Adaptation of Time Series Foundation Models

LLM-based time series forecasting 6→16

Repurpose language models for forecasting through prompting, retrieval and alignment.

Rethinking Post-Training Recipes for Multimodal Time-Series Forecasting

TiRex-2: Generalizing TiRex to Multivariate Data and Streaming

Time-series language reasoning 9→8

Combine time series and text for clinical reasoning and multimodal embedding.

MMTA: Benchmarking Multimodal Temporal Analysis with Time Series, Text, and Vision

Revitalizing Medical Time Series with Vision-Informed Retrieval: A Vision-Language Perspective

Time series representation learning 10→4

Learn time series embeddings with contrastive and non-contrastive objectives and augmentation.

Time-Frequency Decoupled Cross-Scale Partial Optimal Transport for Time Series Domain Adaptation

Spectral Energy Allocation Enables Source-Free Domain Adaptation in Time Series Forecasting

Attention for time series forecasting 13→20

Forecast long series with spectral, multi-scale and adaptive transformers and normalization.

FISC: Time-Series Forecasting via First-Layer Statistical Calibration Constraints

WILD: Widely Linear Conditioning for Time Series Forecasting

Graph-based multivariate forecasting 3→10

Forecast multivariate series by learning channel dependence with graph neural networks.

Rubato: Signature Attention for Irregular Multivariate Time Series Forecasting

Multivariate Time Series Forecasting needs Cross Variable Loss

Lightweight long-term forecasters 9→10

Forecast long horizons with lean models exploiting periodicity, memory and mixing.

DSR-TSF: Spectrum-Driven Dynamic Routing for Efficient Long-Horizon Time Series Forecasting

TStruct: Learning Shared Temporal Structures for Long-Term Time-Series Forecasting

Spatiotemporal forecasting models 2→6

Model time series with spatiotemporal graphs, convolutions, state-space and diffusion pretraining.

Progressive Memory Transformer: Memory-Aware Attention for Time-Series

Spectral Transformer Neural Processes

Time series anomaly detection evaluation 4→14

Evaluate and improve anomaly detection on time series with metrics, foundation models and LLMs.

TENET: Time-point Encoding Network for Multivariate Time Series Anomaly Detection

Supervision Recovery for Time Series Anomaly Detection via Counterfactual Pairing

Multivariate anomaly detection 6→4

Detect anomalies across series using graph dynamics, causality and diffusion reconstruction.

Flow+Diff: Unifying Flow and Diffusion Representations for Multivariate Time Series Anomaly Detection

COMET: Codebook-based Online-adaptive Multi-scale Embedding for Time-series Anomaly Detection

Local-DP and private estimation 10→15

Estimate means and statistics under local or central privacy with minimax rates.

Settling Pure Differentially Private Covariance Estimation

Privacy Guarantees in Posterior Sampling under Contamination

Privacy accounting and mechanisms 8→12

Develop tighter composition bounds and noise mechanisms for differential privacy.

Sampling-Free Privacy Accounting for Matrix Mechanisms under Random Allocation

Privacy Amplification Persists under Unlimited Synthetic Data Release

Differentially private synthetic data 4→11

Generate private synthetic data or text and in-context examples under differential privacy.

Ask in the Crowd: Differentially Private LLM Inference via Dummy-Augmented Shuffling

Re-examining Low Rank adaptation for private LLM fine-tuning

Private online and user-level learning 4→10

Analyze private learning in online, shuffle and user-level settings.

Computationally Efficient Replicable Learning of Parities and Applications

Private Prediction via Shrinkage

Non-IID federated learning 9→19

Counter data heterogeneity with drift decoupling, distillation and re-labeling.

FedPeel: Peeling Stabilized Layers for Robust Heterogeneous Federated Learning

Federated Dataset Simulation: Inducing Label-Free Heterogeneity Across Tasks

Federated prompt and LoRA tuning 10→19

Adapt foundation models federatedly with prompts and low-rank adapters.

FedReCall: Recalling Client-Specific Directions in Federated LoRA Fine-tuning

Fed-SB: A Silver Bullet for Extreme Communication Efficiency and Performance in (Private) Federated LoRA Fine-Tuning

Federated learning with pretrained models 9→5

Handle client imbalance and long-tailed data with pretrained and CLIP models and consistency.

FedSEM: Mitigating Cross-Client Evidence Drift in Federated Multiple Instance Learning

Ripple in Still Water: Zero-Shot Clustering in Heterogeneous Federated Learning with Wavelet Scattering Transform

Federated semi-supervised and graph learning 6→5

Learn from unlabeled or graph data across clients with pseudo-labeling.

One-Shot Federated Graph Learning via High-Fidelity Proxies and Transferability-Guided Collaboration

Federated Graph Learning with Local Message Compensation

Defending federated learning against poisoning 4→7

Detect and mitigate malicious clients and backdoors in federated learning with provable defenses.

Consistency-Verified Backdoor Defense for Federated Graph Learning via Cross-Layer Drift

Improved Robust Verifiable Federated Learning Based on Packed Secret Sharing

Membership inference and auditing 2→6

Attack and audit models for privacy leakage with membership inference and statistical tests.

Provable Robustness against Backdoor Attacks via the Primal-Dual Perspective on Differential Privacy

A Unified Framework for Adversary-Aware Differential Privacy Bounds

Gradient inversion attacks 0→3

Reconstruct training data from shared gradients and defend against such leakage.

Leveraging Soft Prompts for Privacy Attacks in Federated Prompt Tuning

FedTrace: Generated-Content-Based Watermark Verification for Traitor Tracing in Federated Learning

Encryption-based private learning 2→3

Use homomorphic encryption and private synthetic data for privacy-preserving learning.

Auditing Privacy Leakage in Tabular Foundation Model Embeddings

Jaguar: Fast Private CNN Inference with Power-of-Two Homomorphic Arithmetic

Client participation analysis 2→8

Prove convergence of federated averaging under client heterogeneity and arbitrary participation.

FedAdaVR: Adaptive Variance Reduction for Robust Federated Learning under Limited Client Participation

Mitigating Data Heterogeneity Effect in Client-Reshuffling-Based Federated Learning

Federated optimization algorithms 2→4

Develop faster local-update and composite or minimax optimizers for federated learning.

Constant Term Shrinkage for Federated Learning

Breaking the $\sqrt{d}$ Communication Barrier in Federated Sampling with Adaptive Hamiltonian Monte Carlo

Efficient local training in FL 3→2

Cut federated training cost with low precision, few rounds and decoupled local learning.

Federate the Router: Learning Language Model Routers with Sparse and Decentralized Evaluations

Principled Federated Random Forests for Heterogeneous Data

Communication-efficient heterogeneous FL 5→2

Reduce communication via dynamic regularization, sparse training and heterogeneity-aware client sampling.

Energy Is All We Need: Beyond FLOPs in Model-Heterogeneous Federated Learning

ERIS: Enhancing Privacy and Scalability in Federated Learning via Federated Shard Aggregation

LLM jailbreak and injection defense 9→18

Study jailbreaks and prompt injections and train or evaluate safer LLMs.

Systematic Scaling Analysis of Jailbreak Attacks in Large Language Models

MemPoison: Uncovering Persistent Memory Threats and Structural Blind Spots in LLM Agents

Adversarial prompts for language models 11→11

Attack and defend language models with adversarial prompts and robustness benchmarks.

Statistical Estimation of Adversarial Risk in Large Language Models under Best-of-N Sampling

Safeguarding LLMs via Model-Agnostic Latent Safety Signals from Dark Knowledge

Vision-language model robustness 8→9

Attack and harden vision-language models with multimodal adversarial search and prompt tuning.

Even Sailors Need Calm Seas: Taming the Geometry of VLMs for Fast Adversarial Fine-Tuning

Cross-Model Transfer Attacks against Large Vision-Language Models via Model Diversity Enrichment and Stochastic Parameter Sampling

Membership inference and inversion 6→9

Attack and defend models through membership inference, model inversion and reconstruction.

Assessing Per-Sample Membership Inference Vulnerability without Retraining

Estimating Model-Level Membership Inference Vulnerability Without Reference Models

Robustness-accuracy trade-off analysis 0→4

Study adversarial training and its evaluation, including trade-offs with accuracy and weak-attack pitfalls.

Learning Better Certified Models from Empirically-Robust Teachers

Measuring Robustness and Efficiency in a Connectome-Constrained Fly Visual System Model on a Collision-Detection Task

Randomized smoothing certification 1→4

Certify robustness by smoothing classifiers with noise, extending to new norms and tasks.

Multi-Constrained Randomized Smoothing Certificates

Halt Fast! Early Stopping for Certified Robustness

Adversarial training refinements 0→3

Reweight examples and classes or tune training to improve robustness and convergence.

A Simple Class-Agnostic Approach to Enhance Fair Adversarial Training

GUARD: Scalable Gradient-based Unlearning with Adversarial Robustness Defense

Robust overfitting and regularization 3→3

Explain and reduce robust overfitting and improve generalization of adversarially trained models.

Rethinking Incompleteness: Formalizing Protocol Divergence and Train-Once Learning for Robust IMVC

Are Well-Trained Surrogates Optimal? Rethinking the Surrogate Role with Instability for Unlearnable Examples

Backdoors in large multimodal models 5→14

Attack and analyze backdoors in LLMs, vision-language and action models.

Not Suppressing or Purifying: Backdoor Containment via Expert Quarantine and Shutdown in LLMs

Backdoor Purification for LoRA-Tuned LLMs via Null-Space Projection

Trigger-based backdoor attacks 4→8

Design poisoning-based backdoor attacks with hidden or clean-label triggers on vision models.

RPP: A Certified Poisoned-Sample Detection Framework for Backdoor Attacks under Dataset Imbalance

Backdoor Attacks Rerouted: BatchNorm as a Sink for Adversarial Signals

Detecting poisoned samples 2→4

Identify backdoor data or inputs and unlearn triggers, including in pretrained language models.

Clean Data Can Still Carry Backdoors: Support-Persistent Backdoors for Model Reuse

When Sanitization Becomes the Trigger: Defense-Triggered Backdoor Attacks

Transfer attacks on face recognition 3→5

Improve transfer attacks with momentum and checkpoints, including against face recognition.

Anchoring Adversarial Trajectories to Data Manifolds: A Bilevel Transfer Optimization Framework

AnchorRep: Defending LLMs Against Cross-Model Adversarial Transfer via Representation Repulsion

Data-free substitute training 3→4

Attack black-box models by training substitutes without data or stealing models.

BiShield-TEE: On the (In-)Security of Unilateral Weight Obfuscation in On-Device TEE-Shielded LLM Partition

FILOsofer: A TEE-Shielded Model Partitioning Framework based on Fisher Information-Guided LoRA Obfuscation

Robustness benchmarks and defenses 1→2

Benchmark adversarial robustness and defend through architecture, denoising and loss design.

Adversarial Attack and Defense for Machine Learning in Statistical Physics

Seeing but Not Detecting: Privacy-Preserving Scene Text Attack via Hybrid Adversarial Policy Learning

Transferable adversarial examples 3→2

Boost transfer of adversarial examples across models with gradient and input-diversity techniques.

Surrogate Calibration for Transferable Adversarial Attacks against Black-Box MLLMs

What to Perturb, How to Propagate: A Graph-Guided Transferable Attack on VLP Models

Open-vocabulary semantic segmentation 4→10

Segment arbitrary categories using CLIP, mask adaptation, and diffusion without dense labels.

Open-Vocabulary 3D Part Segmentation with Semantic Propagation Hawkes Process

SetCon: Towards Open-Ended Referring Segmentation via Set-Level Concept Prediction

Interactive and referring segmentation 3→8

Segment objects from clicks or language expressions with refinement pipelines.

Auditing Correlated Failures in Frozen-Feature Pretrained-Encoder Pools for Medical Segmentation

U-Bench: A Comprehensive Understanding of U-Net through 100-Variant Benchmarking

Object-level representation learning 6→7

Learn object-centric recognition features through contrastive pretraining and context debiasing.

Reasoning-based Spatial Prior (RSP): Learning Spatial Priors from Multimodal LLMs for Object Detection

HiLoc: A Hierarchical Representation Method for Spatial Localization in Multimodal Large Language Models

Remote sensing segmentation and change 8→7

Build datasets and models for geospatial segmentation and change detection.

In-Context Learning for Remote Sensing Vision: A Semantic-Aware Rotation-Robust Diffusion Framework

EviSAM3: Evidence-Driven SAM3 for Referring Remote Sensing Image Segmentation

Region-based and query-based detectors 1→6

Design detectors using region proposals, learnable queries, and transformer pretraining.

ContextShift: A Controlled Benchmark for Context Dependence in Object Detection

iDETR: Implicit DETR for Tiny Object Detection

Context modeling for semantic segmentation 5→6

Capture context with structured models, prototypes, and CRFs for semantic segmentation.

Fractal-G: Topology-Aware Heterogeneous Graphs for Medical Image Segmentation

Beyond Pixel-wise Supervision: Local Structure Regularization for Semantic Segmentation

Domain-specific segmentation and detection 4→5

Adapt context-aware networks to radar, maps, HDR, and medical imaging.

ReSMap: Recasting Satellite Priors for Robust and Accurate Online HD Map Construction

Beyond Chamfer Distance: Granular Order-aware Evaluation Metric For Online Mapping

Context for hard detection conditions 3→4

Use context and attention to detect camouflaged, low-light, and one-shot objects.

Reviving the Discarded High-Resolution Feature for Transformer-Based Tiny Object Detection

PRISM: Priority-Guided Scanning in the Wavelet Domain for UAV Maritime Small Object Detection

Monocular 3D object detection 0→3

Improve single-image 3D detection with auxiliary contexts, distillation, and depth-aware losses.

RefineAny3D: Depth Refinement as Semantic Alignment for Monocular 3D Detection

Rethinking Geometric Depth in Monocular 3D Object Detection: A Projection-Consistent Reformulation

Multi-camera BEV detection 0→3

Detect 3D objects from multiple cameras with BEV representations and cross-view attention.

RESBev: Making BEV Perception More Robust

Towards Effective and Transferable Physical Camouflage against Multi-View BEV-based 3D Perception in Autonomous Driving

LiDAR perception for driving 5→2

Build efficient LiDAR detectors, anomaly datasets, and generative models for driving perception.

Virtual Task Prompting for Multi-Task Scene Understanding

WayPOP: A Panoramic Open-Set Panoptic Tracking Benchmark

CAM refinement and prototypes 0→3

Refine class activation maps with prototypes, SAM, and co-training for segmentation.

Improved Weakly Supervised Semantic Segmentation with A Relationally Optimized Prototype Memory Bank Framework

When Noise Meets Long-Tail: Feature-Threshold Dual Calibration for Robust Pseudo-Labeling

Other 3→2

TrajLoc: Trajectory-Attention Localization for Multi-Object Motion Control

Self-Supervised Doppler-Guided RF Odometry

Association cues for online MOT 1→2

Improve detection-to-track association with appearance, ID prediction, and weak cues.

TrackFish3D: Self-Supervised 3D Tracking of Schooling Fish from Multi-view Videos

ViewSAM: Learning View-aware Semantics for Weakly Supervised Cross-view Referring Multi-Object Tracking

Low-rank and model-merging adaptation 9→15

Controls forgetting in continual fine-tuning through low-rank updates, prompts and weight merging.

PM-LoRA: Scalable Continual Learning via Progressive Merging of Low Rank Adapters

PaLoRA: Paced Low-Rank Adaptation for Continual Learning

Continual pretraining of language models 4→9

Analyzes and mitigates forgetting during continual pretraining and knowledge updating of large models.

CapTrack: Multifaceted Evaluation of Forgetting in LLM Post-Training

Combating Catastrophic Forgetting in Continual Domain Adaptation via Knowledge Recasting

Mixture-of-experts for multimodal continual tuning 5→6

Expands vision-language models with expert adapters and benchmarks continual multimodal learning.

Dynamic Cross-Modal Prompt Generation for Multimodal Continual Instruction Tuning

Not Every Image Teaches Vision: Visual-Necessity-Gated Continual Learning for Multimodal Large Language Models

Machine unlearning by selective forgetting 0→6

Removes specific data influence from trained networks without full retraining.

Unlearning That Lasts: Utility-Preserving, Robust, and Almost Irreversible Forgetting in LLMs

Forgetting to Improve: Principled Data Removal in Active Learning

Theoretical analysis of forgetting 7→8

Studies why and when networks forget, using linear models, teacher-student setups and complexity results.

Lost or Hidden? A Concept-Level Forgetting in Supervised Continual Learning

STABLE: A Continual Learning Optimizer with Adaptive Drift Control

Plasticity-preserving continual reinforcement learning 3→7

Keeps policies adaptable using architectural design, policy subspaces and composition across tasks.

From Anti-Forgetting to Fast Adaptation: Continual Reinforcement Learning with World Models

Learning to Continually Learn via Meta-learning Agentic Memory Designs

Unified and unsupervised continual frameworks 3→5

Proposes general frameworks, benchmarks and architecture-growing methods for task-agnostic and unsupervised continual learning.

Re-evaluating Continual Learning with Few-Shot Adaptation

FedVaccine: Knowledge Recall after Spatial-Temporal Catastrophic Forgetting in Federated Continual Learning via Gradient-Based Vaccine

Gradient projection and plasticity preservation 2→4

Constrains updates to subspaces or preserves network trainability to limit forgetting.

Distance-Dependent Connectivity Shapes Continual Learning by Synaptic-Resource-Delimited Separation of Neural Dynamics

Predicting Plasticity in Deep Continual Learning: A Theoretical Perspective

Exemplar-efficient class-incremental learning 3→10

Retains old classes with compact exemplars, embedding distillation and pretrained models.

Hyperbolic Displacement-Constrained Adaptation for CLIP-Based Class-Incremental Learning

Class-Incremental Learning via LoRA-based Elastic Ensemble of Experts

Class-balanced knowledge retention 2→5

Preserves per-class knowledge and decision boundaries when new classes arrive.

Subspace-Guided Continual Learning: Hessian Based Stable–Plastic Decomposition for Exemplar-Free Class-Incremental Learning

A Theoretical Bridge Between Long-Tailed Recognition and Continual Learning

Domain-incremental learning with memory 2→2

Adapts to sequentially arriving domains using memory, domain-aware representations and randomization.

Anchoring Physiological Invariance, Expanding Domain Plasticity: Prior-Stabilized Dynamic Adaptation for Continual rPPG Measurement

Cross-Domain Knowledge Separation and Positive Transmission for Noisy Domain Incremental Learning

Continual test-time adaptation 2→2

Adapts models online to shifting target domains using adapters, prompts and knowledge fusion.

Reliability-Budgeted Edge–Cloud Adaptation for Continual Multimodal Dehazing on UAV

CTRL: Continual Test-Time Reinforcement Learning for Large Language Models

Point-language and open-vocabulary 3D 6→12

Align 3D point clouds with language for zero-shot and open-vocabulary understanding.

Hierarchical Graph Alignment for Cross-Modal 3D Scene Grounding

GraphWrit3R: End-to-End Writing of Scene Graphs for Multi-Modal 3D Scenes

Equivariant point cloud networks 7→6

Build rotation-equivariant and invariant architectures for point clouds.

SO(3)-Equivariant Learning on CAD Boundary Representations

Fast Alignment of Embeddings: Computational Guarantees for Anisotropic Procrustes-Wasserstein

Self-supervised point cloud pretraining 5→3

Pretrain point representations via cross-modal, contrastive and diffusion objectives.

RILA: A Radar-Native Structured Interface from Sparse mmWave Point Clouds to Large Language Models

mmLIP: mmWave Radar-Language Interactive Pretraining via Point Confidence

Category-level pose estimation 4→3

Estimate object pose and part motion from point clouds.

RVLoss: Runoff Vote Loss for Self-Supervised LiDAR Scene Flow Estimation

Motion Cues from Image-based Point Tracking for LiDAR Scene Flow Estimation

Learned feature-based registration 3→3

Register point clouds with learned features, multi-view and image cues, and unsupervised training.

Register Anything Model for Generalizable and Robust Point Cloud Registration

RSPReg: Reliability-aware Structural Prototype Learning for Point Cloud Registration

Point cloud diffusion generation 1→3

Generate 3D shapes with diffusion models over point clouds and latent spaces.

Geometry-Centered 3D Latent World Models for Growing Surfaces

HilbertGen-3D:Hilbert--Multifractal Conditioning for Topology-Aware 3D Generation

Point cloud semantic segmentation 2→2

Segment point cloud scenes with weak supervision and joint semantic-instance learning.

PARI: Policy-Driven Active Residual Intervention for Weakly Supervised Point Cloud Segmentation

Relational Feature Distillation for Lightweight 3D Point Cloud Segmentation