Skip to content
Computational Pathology Reading Roadmap

ICML 2026 — corpus atlas

Caption labels, not panels. A six-panel figure counts once. Observed figures, tables and section spans describe the inspected source versions; they are not quotas or acceptance thresholds. For visual purpose, use the examples and type guides.

168 / 168inspected PDFs / oral inventory records
5,000physical pages in those PDFs
8editorial contribution types
0separate files, excluded from totals

What the measured manuscripts look like

Observed extraction results, not a target length or figure quota. The middle 50% is the interval from the 25th to the 75th percentile. Missing boundaries are excluded from that span’s denominator, not replaced by zero. Caption counts are extraction-assisted and can miss nonstandard or image-only labels; inspect each paper’s notes.

MeasureMedianMiddle 50%Full rangeDenominator
Inspected PDF lengthphysical pages2621–3510–76168 papers
Before references / appendixestimated page-equivalents9.048.6–9.197.6–10.04168 papers
Attached appendix spanestimated page-equivalents14.818.96–23.650.88–61.91161 papers
Main figurescaption labels54–70–12168 papers
Main tablescaption labels21–40–13168 papers
Appendix figurescaption labels42–9.250–130168 papers
Appendix tablescaption labels41–7.250–30168 papers
Separate source-file lengthwhole-file physical pages, possibly including repeated main text; not appendix-only lengthNot resolvedNot resolved–Not resolvedNot resolved–Not resolved0 papers
Separate source-file figureswhole-file caption labels; may include overlapping main-text displaysNot resolvedNot resolved–Not resolvedNot resolved–Not resolved0 papers
Separate source-file tableswhole-file caption labels; may include overlapping main-text displaysNot resolvedNot resolved–Not resolvedNot resolved–Not resolved0 papers

Compare like contributions

Each measured paper has one primary editorial type so the totals sum to 168. Secondary contributions were not separately coded; empty secondary lists do not imply the absence of mixed contributions. These are classifications of the annotated argument, not official ICML tracks. The table uses the same extraction method and shows medians; the category guides below explain what these artifacts should accomplish.

Primary contributionPapersMain figures / tablesBody span ≈ pp.Appendix span ≈ pp.
Architecture & representation226 / 49.1112.9121/22 attached boundaries; separate PDF median Not resolved pp. (0 papers)
Training, objectives & optimization356 / 38.8311.9135/35 attached boundaries; separate PDF median Not resolved pp. (0 papers)
Theory & formal analysis383.5 / 0.59.0518.8837/38 attached boundaries; separate PDF median Not resolved pp. (0 papers)
Empirical analysis, diagnosis & auditing236 / 29.0414.9123/23 attached boundaries; separate PDF median Not resolved pp. (0 papers)
Datasets, benchmarks & measurement176 / 39.1423.9117/17 attached boundaries; separate PDF median Not resolved pp. (0 papers)
Efficiency, inference & systems165 / 38.9815.9115/16 attached boundaries; separate PDF median Not resolved pp. (0 papers)
Domain-grounded methods & physical systems85.5 / 39.049.918/8 attached boundaries; separate PDF median Not resolved pp. (0 papers)
Position & research agenda92 / 09.045.925/9 attached boundaries; separate PDF median Not resolved pp. (0 papers)

Search all 168 measurement records

Open a paper’s ledger for every detected top-level section, its estimated length and share of the PDF, and figure/table locations. Its existing study explains what those visuals and sections do.

Showing 168 of 168 records.

Paper / categoryPDF pagesMain F / TAppendix F / TBody / appendix span ≈ pp.Notes
Do We Need Adam? Surprisingly Strong and Sparse Reinforcement Learning with SGD in LLMsEmpirical analysis, diagnosis & auditing184 / 62 / 18.41 / 2.91No automatic flagsSource correction retained
Learning Unmasking Policies for Diffusion Language ModelsTraining, objectives & optimization3510 / 024 / 59.1 / 21.91No automatic flags
Less is Enough: Synthesizing Diverse Data in LLM Feature Space with Sparse AutoencodersTraining, objectives & optimization547 / 411 / 158.87 / 39.92No automatic flags
OPUS: Towards Efficient and Principled Data Selection in Large Language Model Pre-training in Every IterationTraining, objectives & optimization425 / 52 / 69.18 / 28.45No automatic flags
Benchmarking at the Edge of ComprehensionDatasets, benchmarks & measurement333 / 215 / 109.39 / 20.91No automatic flags
daVinci-Dev: Agent-native Mid-training for Software EngineeringTraining, objectives & optimization184 / 63 / 09.34 / 4.56No automatic flags
Strategic Navigation or Stochastic Search? How Agents and Humans Reason Over Document CollectionsDatasets, benchmarks & measurement648 / 317 / 229.22 / 50.88No automatic flags
VenusBench-Mobile: A Challenging and User-Centric Benchmark for Mobile GUI Agents with Capability DiagnosticsDatasets, benchmarks & measurement266 / 54 / 38.72 / 15.91No automatic flags
dnaHNet: A Scalable and Hierarchical Foundation Model for Genomic Sequence LearningArchitecture & representation196 / 12 / 89.11 / 5.91No automatic flags
FLIP2: Expanding Protein Fitness Landscape Benchmarks for Real-World Machine Learning ApplicationsDatasets, benchmarks & measurement544 / 132 / 188.57 / 41.96No automatic flags
Protein Autoregressive Modeling via Multiscale Structure GenerationArchitecture & representation216 / 54 / 108.81 / 8.91No automatic flags
Protein Fold Classification at Scale: Benchmarking and PretrainingDatasets, benchmarks & measurement196 / 53 / 49.14 / 6.88No automatic flags
DiScoFormer: Plug-In Density and Score Estimation with TransformersArchitecture & representation2111 / 54 / 010.04 / 7.91No automatic flags
LASER: Learning Active Sensing for Continuum Field ReconstructionDomain-grounded methods & physical systems266 / 1112 / 39.28 / 13.91No automatic flags
Multimodal Nested Learning for Decoupled and Coordinated OptimizationTraining, objectives & optimization215 / 34 / 18.36 / 9.68No automatic flags
Riemannian Metric Matching for Scalable Geometric Modeling of DistributionsTraining, objectives & optimization205 / 15 / 09.17 / 8.91No automatic flags
Position: Don’t Just "Fix it in Post": A Science of AI Must Study Training DynamicsPosition & research agenda170 / 00 / 08.79 / UnresolvedNo attached appendix boundary detected
A Systematic Study of Behavioral Cloning for Scientific Data AnnotationEmpirical analysis, diagnosis & auditing466 / 133 / 199.04 / 33.92No automatic flags
AI Engram: In Search of Memory Traces in Artificial IntelligenceTheory & formal analysis269 / 49 / 59.21 / 13.96No automatic flagsSource correction retained
Guaranteed Optimal Compositional Explanations for NeuronsTheory & formal analysis431 / 49 / 59.04 / 31.91No automatic flags
Are VLMs Seeing or Just Saying? Uncovering the Illusion of Visual Re-examinationEmpirical analysis, diagnosis & auditing274 / 98 / 99.16 / 14.91No automatic flagsSource correction retained
CLEAR: Context-Aware Learning with End-to-End Mask-Free Inference for Adaptive Video Subtitle RemovalDomain-grounded methods & physical systems146 / 34 / 09.11 / 2.91No automatic flagsSource correction retained
Motion Attribution for Video GenerationTraining, objectives & optimization225 / 24 / 58.19 / 9.96No automatic flags
PhotoAgent: Exploratory Visual Aesthetic Planning with Large Vision ModelsDomain-grounded methods & physical systems316 / 27 / 48.71 / 18.91No automatic flags
Asymmetric Perturbation in Solving Bilinear Saddle-Point OptimizationTheory & formal analysis355 / 04 / 48.75 / 23.65No automatic flagsSource correction retained
Mixtures Closest To A Given Measure: A Semidefinite Programming ApproachTheory & formal analysis234 / 00 / 18.19 / 10.91No automatic flags
On the Convergence Rate of LoRA Gradient DescentTheory & formal analysis195 / 03 / 17.73 / 7.91No automatic flags
Revenue Guarantees of No-Swap-Regret Dynamics in First Price AuctionsTheory & formal analysis223 / 01 / 07.92 / 10.91No automatic flags
Controlled LLM Training on Spectral SphereTraining, objectives & optimization169 / 32 / 08.26 / 4.91No automatic flags
Detecting the Semantic Fixed Point: A Geometric Framework for Efficient InferenceEfficiency, inference & systems343 / 107 / 108.4 / 22.91No automatic flagsSource correction retained
ECHO: Elastic Speculative Decoding with Sparse Gating for High-Concurrency ScenariosEfficiency, inference & systems215 / 14 / 78.92 / 9.91No automatic flags
MuonSSM: Orthogonalizing State Space Models for Sequence ModelingArchitecture & representation225 / 72 / 59.18 / 9.86No automatic flagsSource correction retained
Don’t Force the Fit: Bounded Log-Likelihood Loss for Enhanced Reasoning in Large Language ModelsTraining, objectives & optimization234 / 34 / 58.54 / 11.91No automatic flagsSource correction retained
Evaluating Robustness of Reasoning Models on Parameterized Logical ProblemsDatasets, benchmarks & measurement259 / 11 / 119.04 / 12.91No automatic flags
TG-RAG: A Retrieval-Augmented Framework for Reasoning Guidance in Specialized DomainsDomain-grounded methods & physical systems174 / 44 / 29.04 / 4.91No automatic flags
RAGEN-2: Reasoning Collapse in Agentic RLEmpirical analysis, diagnosis & auditing339 / 35 / 78.17 / 19.57No automatic flagsSource correction retained
Position: Anthropomorphic Misalignment Research Needs Stronger EvidencePosition & research agenda253 / 00 / 19.15 / 9.91No automatic flagsSource correction retained
Monitoring MonitorabilityDatasets, benchmarks & measurement656 / 128 / 39.28 / 52.91No automatic flagsSource correction retained
The Obfuscation Atlas: Mapping Where Honesty Emerges in RLVR with Deception ProbesEmpirical analysis, diagnosis & auditing305 / 410 / 199.4 / 17.91No automatic flags
VALUEFLOW: Toward Pluralistic and Steerable Value-based Alignment in Large Language ModelsDatasets, benchmarks & measurement638 / 434 / 119.54 / 49.91No automatic flags
From Feasible to Practical: Pareto-Optimal Synthesis PlanningDomain-grounded methods & physical systems215 / 41 / 79.04 / 8.91No automatic flags
ReViT: Rotational-equivariant Vision Transformers for Neural PDE SolversArchitecture & representation397 / 210 / 89.24 / 25.92No automatic flags
Rex: A Family of Reversible Exponential (Stochastic) Runge-Kutta SolversTheory & formal analysis739 / 111 / 89.19 / 57.91No automatic flagsSource correction retained
Towards Sub-Second Molecular Docking as a Structural Primitive: A Quantized Consistency Diffusion FrameworkEfficiency, inference & systems303 / 10 / 39.67 / 16.91No automatic flagsSource correction retained
Chebyshev Policies and the Mountain Car Problem: Reinforcement Learning for Low-Dimensional Control TasksArchitecture & representation235 / 29 / 88.74 / 12.91No automatic flags
Maximum Likelihood Reinforcement LearningTraining, objectives & optimization568 / 224 / 89.65 / 32.9No automatic flagsSource correction retained
Minimax Optimal Strategy for Delayed Observations in Online Reinforcement LearningTheory & formal analysis392 / 21 / 08.27 / 27.91No automatic flagsSource correction retained
Optimal Decision-Making Based on Prediction SetsTheory & formal analysis232 / 01 / 08.67 / 10.91No automatic flagsSource correction retained
Learning Human-Robot Collaboration via Heterogeneous-Agent Lyapunov Policy OptimizationTraining, objectives & optimization158 / 30 / 68.54 / 3.91No automatic flags
Optimal and Scalable MAPF via Multi-Marginal Optimal Transport and Schrödinger BridgesEfficiency, inference & systems314 / 114 / 39.1 / 19.6No automatic flags
RoboMME: Benchmarking and Understanding Memory for Robotic Generalist PoliciesDatasets, benchmarks & measurement554 / 423 / 289.21 / 41.91No automatic flags
Unsupervised Partner Design Enables Robust Ad-hoc TeamworkTraining, objectives & optimization278 / 214 / 58.61 / 13.91No automatic flags
Diffract: Spectral View of LLM Domain AdaptationEmpirical analysis, diagnosis & auditing267 / 015 / 58.6 / 14.91No automatic flags
Foundations of Equivariant Deep Learning: Unifying Graph and Sheaf Neural NetworksTheory & formal analysis230 / 00 / 29.1 / 12.65No automatic flagsSource correction retained
On Minimum Depth and Width of Floating-Point Neural Networks for Representing Floating-Point FunctionsTheory & formal analysis191 / 10 / 18.17 / 8.91No automatic flagsSource correction retained
How Many Different Outputs Can a Transformer Generate?Theory & formal analysis305 / 17 / 29.54 / 16.91No automatic flagsSource correction retained
Any-Order GPT as Masked Diffusion Model: Decoupling Formulation and ArchitectureEmpirical analysis, diagnosis & auditing215 / 19 / 48.75 / 9.91No automatic flagsSource correction retained
Error Propagation Mechanisms and Compensation Strategies for Quantized Diffusion ModelsTraining, objectives & optimization182 / 32 / 78.1 / 7.91No automatic flags
High-accuracy and dimension-free sampling with diffusionsTheory & formal analysis330 / 00 / 07.89 / 21.91No automatic flags
The Flexibility Trap: Rethinking the Value of Arbitrary Order in Diffusion Language ModelsEmpirical analysis, diagnosis & auditing169 / 23 / 49.22 / 3.96No automatic flagsSource correction retained
Mitigating Reward Hacking in RLHF via Bayesian Non-negative Reward ModelingTraining, objectives & optimization226 / 54 / 69.04 / 8.96No automatic flagsSource correction retained
DR Tulu: Reinforcement Learning with Evolving Rubrics for Deep ResearchTraining, objectives & optimization566 / 433 / 139.3 / 42.91No automatic flagsSource correction retained
Simultaneous Speech-to-Speech Translation Without Aligned DataTraining, objectives & optimization208 / 36 / 38.66 / 7.91No automatic flags
Video-Based Optimal Transport for Feedback-Efficient Offline Preference-Based Reinforcement LearningTraining, objectives & optimization237 / 312 / 108.36 / 9.91No automatic flagsSource correction retained
Position: Irresponsible AI: big tech’s influence on AI research and associated impactsPosition & research agenda182 / 00 / 09.3 / 0.91No automatic flagsSource correction retained
Activation Oracles: Training and Evaluating LLMs as General-Purpose Activation ExplainersEmpirical analysis, diagnosis & auditing376 / 018 / 39.12 / 25.91No automatic flagsSource correction retained
Large Language Models Develop Novel Social Biases Through Adaptive ExplorationEmpirical analysis, diagnosis & auditing416 / 26 / 99.2 / 26.91No automatic flagsSource correction retained
Mechanistic Data Attribution: Tracing the Training Origins of Interpretable LLM UnitsEmpirical analysis, diagnosis & auditing336 / 29 / 69.25 / 20.91No automatic flagsSource correction retained
Bad Seeing or Bad Thinking? Rewarding Perception for Multimodal ReasoningTraining, objectives & optimization204 / 40 / 28.1 / 7.91No automatic flags
DroneDINO: Towards Heterogeneous Routed Mixture of Experts for Drone-based Unified Object DetectionArchitecture & representation104 / 70 / 08.1 / UnresolvedNo attached appendix boundary detected
Lottery Prior: Randomized Neural Compression for Zero-Shot Inverse ProblemsTraining, objectives & optimization376 / 610 / 149.15 / 24.92No automatic flagsSource correction retained
Scalable Event Cloud Network for Event-based ClassificationArchitecture & representation184 / 114 / 29.04 / 5.77No automatic flagsSource correction retained
Position: Stop Automating Peer Review Without Rigorous EvaluationPosition & research agenda326 / 16 / 69.11 / 17.91No automatic flagsSource correction retained
Position: The ML Community Must Build an AI-Augmented Peer-Review EcosystemPosition & research agenda183 / 20 / 18.92 / 5.92No automatic flagsSource correction retained
Incentivizing Truthfulness and Collaborative Fairness in Bayesian LearningTheory & formal analysis383 / 116 / 39.18 / 26.91No automatic flagsSource correction retained
Is Your LLM Overcharging You? Tokenization, Transparency, and IncentivesTheory & formal analysis253 / 08 / 19.28 / 11.91No automatic flagsSource correction retained
Jailbreak Foundry: From Papers to Runnable Attacks for Reproducible BenchmarkingDatasets, benchmarks & measurement434 / 17 / 78.75 / 29.91No automatic flagsSource correction retained
Quantifying Frontier LLM Capabilities for Container Sandbox EscapeDatasets, benchmarks & measurement326 / 011 / 108.39 / 19.91No automatic flagsSource correction retained
Robust Harmful Features Under Jailbreak Attacks: Mechanistic Evidence from Attention Head Specialization in Large Language ModelsEmpirical analysis, diagnosis & auditing338 / 29 / 118.7 / 20.47No automatic flags
When the Prompt Becomes Visual: Vision-Centric Jailbreak Attacks for Large Image Editing ModelsDatasets, benchmarks & measurement277 / 312 / 58.31 / 15.91No automatic flagsSource correction retained
Exact Functional ANOVA Decomposition for Categorical Inputs ModelsTheory & formal analysis183 / 70 / 19.13 / 6.91No automatic flags
Joint Learning in the Gaussian Single Index ModelTheory & formal analysis222 / 00 / 08.78 / 11.91No automatic flagsSource correction retained
SVRG and Beyond via Posterior CorrectionTraining, objectives & optimization245 / 04 / 19.04 / 8.91No automatic flags
A Causal Decomposition Approach for Fair Contextual Multi-Armed BanditsTheory & formal analysis512 / 12 / 19.13 / 37.91No automatic flagsSource correction retained
CAT-Q: Cost-efficient and Accurate Ternary Quantization for LLMsEfficiency, inference & systems193 / 1311 / 29.04 / 6.91No automatic flagsSource correction retained
POET-X: Memory-efficient LLM Training by Scaling Orthogonal TransformationEfficiency, inference & systems156 / 101 / 58.2 / 4.9No automatic flagsSource correction retained
ReQAT: Achieving Full-Precision Reasoning Accuracy with 4-bit Floating-Point Quantization-Aware TrainingTraining, objectives & optimization227 / 74 / 109.21 / 7.91No automatic flags
Skip a Layer or Loop It? Learning Program-of-Layers in LLMsEfficiency, inference & systems228 / 40 / 710.04 / 9.91No automatic flagsSource correction retained
From Abstraction to Instantiation: Learning Behavioral Representation for Vision-Language-Action ModelArchitecture & representation187 / 32 / 49.04 / 5.91No automatic flags
From Pixels to Tokens: A Systematic Study of Latent Action Supervision for Vision-Language-Action ModelsEmpirical analysis, diagnosis & auditing256 / 55 / 178.84 / 13.91No automatic flagsSource correction retained
Pretrained Vision-Language-Action Models are Surprisingly Resistant to Forgetting in Continual LearningEmpirical analysis, diagnosis & auditing228 / 56 / 79.69 / 8.92No automatic flagsSource correction retained
XR-1: Towards Versatile Vision-Language-Action Models via Learning Unified Vision-Motion RepresentationsArchitecture & representation457 / 411 / 159.31 / 29.91No automatic flags
Position: The Alignment Community is Unintentionally Building a Censor’s ToolkitPosition & research agenda140 / 10 / 08.21 / UnresolvedNo attached appendix boundary detected
Information Flow Reveals When to Trust Language ModelsEmpirical analysis, diagnosis & auditing206 / 21 / 119.04 / 7.91No automatic flags
Modeling Hierarchical Thinking in Large Reasoning ModelsEmpirical analysis, diagnosis & auditing263 / 22 / 79.33 / 13.91No automatic flagsSource correction retained
Reward-free Alignment for Conflicting ObjectivesTraining, objectives & optimization276 / 20 / 08.32 / 13.91No automatic flags
A Recursive Decomposition Framework for Causal Structure Learning in the Presence of Latent VariablesTheory & formal analysis324 / 09 / 49.25 / 18.91No automatic flagsSource correction retained
DISCO: Mitigating Bias in Deep Learning with Conditional Distance CorrelationTraining, objectives & optimization304 / 38 / 49.23 / 17.91No automatic flagsSource correction retained
Disentangling Latent Risk Pathways via Bayesian Hypergraph InferenceArchitecture & representation224 / 67 / 49.1 / 9.91No automatic flags
On the Identifiability of Poisson Branching Structural Causal Model Under Latent ConfoundingTheory & formal analysis306 / 14 / 69.32 / 18.88No automatic flagsSource correction retained
CoEvol-NO: State and Coordinate Co-Evolution with an Error-Driven Predictor-Corrector Paradigm for Neural Operator TransformerArchitecture & representation273 / 105 / 99.15 / 14.86No automatic flags
Geometric Flow Grounding: A Unified Manifold Decoupling Framework for Dynamics Discovery and VerificationArchitecture & representation276 / 62 / 59.04 / 14.91No automatic flags
Solving Time-Dependent Differential Equations with Physical Dynamical SystemsEfficiency, inference & systems177 / 11 / 08.6 / 4.91No automatic flags
Training-Free Bayesian Filtering with Generative EmulatorsDomain-grounded methods & physical systems244 / 28 / 08.21 / 11.91No automatic flagsSource correction retained
DPO Unchained: Your Training Algorithm is Secretly Disentangled in Human Choice Theory (and Its Loss’ Convexity is Dispensable)Theory & formal analysis241 / 13 / 09.25 / 10.92No automatic flagsSource correction retained
High-Accuracy Sampling for Diffusion Models and Log-Concave DistributionsTheory & formal analysis350 / 00 / 08.12 / 23.91No automatic flagsSource correction retained
Rational TransductorsArchitecture & representation366 / 13 / 19.15 / 22.91No automatic flags
To Grok Grokking: Provable Grokking in Ridge RegressionTheory & formal analysis234 / 03 / 08.75 / 11.91No automatic flags
ConFlux: Multivariate Time Series in Flux, One Unified Forecast in ConfluenceArchitecture & representation316 / 420 / 98.1 / 19.91No automatic flags
From Text to Forecasts: Bridging Modality Gap with Temporal Evolution Semantic SpaceArchitecture & representation178 / 20 / 18.65 / 5.91No automatic flags
Markov Chain Monte Carlo without Evaluating the Target: an Auxiliary Variable ApproachTheory & formal analysis352 / 311 / 49.06 / 23.91No automatic flagsSource correction retained
Path-dependent Discrete Amortized InferenceArchitecture & representation267 / 27 / 38.3 / 12.91No automatic flags
Midtraining Bridges Pretraining and Posttraining DistributionsEmpirical analysis, diagnosis & auditing266 / 47 / 58.57 / 12.91No automatic flagsSource correction retained
ThreadWeaver: Adaptive Threading for Efficient Parallel Reasoning in Language ModelsEfficiency, inference & systems365 / 52 / 68.38 / 24.91No automatic flagsSource correction retained
TokSuite: Measuring the Impact of Tokenizer Choice on Language Model BehaviorDatasets, benchmarks & measurement462 / 211 / 309.07 / 30.21No automatic flags
WeDLM: Reconciling Diffusion Language Models with Standard Causal Attention for Fast InferenceEfficiency, inference & systems185 / 24 / 39.14 / 6.5No automatic flags
Mind Your Margin and Boundary: Are Your Distilled Datasets Truly Robust?Training, objectives & optimization156 / 41 / 68.22 / 4.91No automatic flags
On the Difficulty of Learning a Meta-network for Training Data SelectionEmpirical analysis, diagnosis & auditing453 / 42 / 29.14 / 31.91No automatic flagsSource correction retained
The Signal is in the Steps: Local Scoring for Reasoning Data SelectionTraining, objectives & optimization316 / 44 / 189.36 / 16.91No automatic flags
Transforming Weather Data from Pixel to Latent SpaceArchitecture & representation238 / 18 / 08.76 / 10.91No automatic flags
Distributional Inverse Reinforcement LearningTraining, objectives & optimization325 / 38 / 59.19 / 19.91No automatic flagsSource correction retained
On the Role of Computation in Reinforcement LearningTheory & formal analysis157 / 10 / 28.25 / 2.91No automatic flagsSource correction retained
Second-Order Smooth Planning with Optimal-Transport Bellman SmoothingTheory & formal analysis400 / 20 / 09.04 / 29.92No automatic flagsSource correction retained
Stabilizing the Q-Gradient Field for Policy Smoothness in Actor-Critic MethodsTraining, objectives & optimization231 / 85 / 179.19 / 11.91No automatic flagsSource correction retained
Position: There are futures that benchmark-driven AI cannot seePosition & research agenda130 / 00 / 09.04 / UnresolvedNo attached appendix boundary detected
CausalGame: Benchmarking Causal Thinking of LLM Agents in GamesDatasets, benchmarks & measurement4012 / 412 / 189.23 / 23.91No automatic flagsSource correction retained
Characterizing, Evaluating, and Optimizing Complex ReasoningTraining, objectives & optimization686 / 312 / 38.36 / 54.91No automatic flags
Rare Event Analysis of Large Language ModelsEmpirical analysis, diagnosis & auditing235 / 02 / 09.12 / 9.91No automatic flags
Compositional Generalization Requires Linear, Orthogonal Representations in Vision Embedding ModelsTheory & formal analysis6810 / 035 / 39.21 / 53.92No automatic flagsSource correction retained
Non-Euclidean Gradient Descent Operates at the Edge of StabilityTheory & formal analysis276 / 013 / 08.75 / 15.91No automatic flags
PRISM: Gauge-Invariant Tangent-Space Differentially Private LoRATheory & formal analysis354 / 517 / 39.08 / 22.38No automatic flags
Robust Contextual Optimization with Missing CovariatesTheory & formal analysis626 / 23 / 59.16 / 50.91No automatic flags
FlatLand: Personalized Graph Federated Learning via Tailored Lorentz SpaceArchitecture & representation347 / 43 / 59.19 / 20.92No automatic flagsSource correction retained
MV-FGAD: Towards Efficient and Effective Federated Graph Anomaly Detection via Multi-view LearningEfficiency, inference & systems2010 / 35 / 48.62 / 7.91No automatic flagsSource correction retained
PhenoBrain: Phenotype-Conditioned Long-Range Communication for Multi-Modal Brain Network AnalysisDomain-grounded methods & physical systems136 / 12 / 08.8 / 2.91No automatic flags
Towards Hierarchy–Uniformity Equilibrium: Recovering Semantic Depth in Hypergraph Contrastive LearningTraining, objectives & optimization276 / 25 / 49.19 / 14.91No automatic flags
3ViewSense: Spatial and Mental Perspective Reasoning from Orthographic Views in Vision-Language ModelsTraining, objectives & optimization245 / 59 / 68.78 / 12.91No automatic flags
Agent0-VL: Exploring Self-Evolving Agent for Tool-Integrated Vision-Language ReasoningTraining, objectives & optimization184 / 48 / 18.1 / 5.91No automatic flags
Holi-Spatial: Evolving Video Streams into Holistic 3D Spatial IntelligenceDatasets, benchmarks & measurement1710 / 56 / 09.13 / 4.91No automatic flags
SpatioLM: Towards General Physical Spatial Intelligence in Vision-Language ModelsArchitecture & representation277 / 86 / 89.16 / 13.91No automatic flags
Expressivity-Efficiency Tradeoffs for Hybrid Sequence ModelsTheory & formal analysis258 / 18 / 18.18 / 14.81No automatic flagsSource correction retained
How much can language models memorize?Empirical analysis, diagnosis & auditing2011 / 19 / 48.89 / 8.46No automatic flagsSource correction retained
Prescriptive Scaling Reveals the Evolution of Language Model CapabilitiesEmpirical analysis, diagnosis & auditing317 / 322 / 18.69 / 17.91Table numbering discontinuity; see source noteSource correction retained
Procedural Pretraining: Warming Up Language Models with Abstract DataTraining, objectives & optimization308 / 115 / 138.7 / 17.91No automatic flagsSource correction retained
τ^2-Bench: Evaluating Conversational Agents in a Dual-Control EnvironmentDatasets, benchmarks & measurement476 / 35 / 29.18 / 35.84No automatic flagsSource correction retained
Measuring Agents in ProductionEmpirical analysis, diagnosis & auditing478 / 022 / 59.04 / 32.91No automatic flagsSource correction retained
CVE-Factory: Scaling Expert-Level Agentic Tasks for Code Security VulnerabilityDatasets, benchmarks & measurement325 / 41 / 79.21 / 19.91No automatic flags
OMAC: A Holistic Optimization Framework for LLM-Based Multi-Agent CollaborationTraining, objectives & optimization294 / 44 / 168.54 / 17.91No automatic flags
DOUBT: Decoupled Object-level Understanding and Bridging via vMF-based Trustworthiness for Hallucination Detection in MLLMsDomain-grounded methods & physical systems244 / 30 / 89.23 / 10.91No automatic flags
Learning to Theorize the World from ObservationArchitecture & representation437 / 315 / 249.3 / 30.91No automatic flags
On the Limits of LLM Adaptability: Impact of Model-Internalized Priors on Annotation Task PerformanceEmpirical analysis, diagnosis & auditing255 / 71 / 199.04 / 11.91No automatic flags
Towards Long-Horizon Interpretability: Efficient and Faithful Multi-Token Attribution for Reasoning LLMsEfficiency, inference & systems265 / 52 / 58.38 / 15.91No automatic flagsSource correction retained
Faster Activation Functions at the Edge for Post-Training SpeedupsEfficiency, inference & systems138 / 40 / 09.04 / UnresolvedNo attached appendix boundary detectedSource correction retained
FlashSinkhorn: IO-Aware Entropic Optimal Transport on GPUEfficiency, inference & systems365 / 33 / 219.16 / 23.91No automatic flags
FlashSketch: Sketch-Kernel Co-Design for Fast Sparse Sketching on GPUsEfficiency, inference & systems514 / 1130 / 08.27 / 40.96No automatic flagsSource correction retained
SoftJAX & SoftTorch: Empowering Automatic Differentiation Libraries with Informative GradientsEfficiency, inference & systems358 / 015 / 49.29 / 21.91No automatic flags
Position: AI Should Facilitate Democratic Deliberation at ScalePosition & research agenda152 / 00 / 09.13 / UnresolvedNo attached appendix boundary detectedSource correction retained
Equilibrium Pricing in Oligopolistic Data MarketsTheory & formal analysis100 / 00 / 08.15 / UnresolvedNo attached appendix boundary detectedSource correction retained
Nash Equilibria in Games with Playerwise Concave Coupling Constraints: Existence and ComputationTheory & formal analysis256 / 00 / 18.29 / 13.91No automatic flags
What Preferences Can—and Cannot—Predict in Multi-Agent Online LearningTheory & formal analysis475 / 010 / 09.23 / 34.91No automatic flags
Position: AI/ML Deepfake Research is Misaligned with AI Generated Non-Consensual Intimate Imagery (AIG-NCII)Position & research agenda152 / 10 / 17.6 / 0.88No automatic flagsSource correction retained
GoodDiffusion: Proactive Copyright Protection for Diffusion Bridge Models via Learnable Sample-specific SignaturesTraining, objectives & optimization294 / 28 / 78.83 / 15.92No automatic flagsSource correction retained
Orthogonal Concept Erasure for Diffusion ModelsTraining, objectives & optimization224 / 64 / 79.21 / 9.91No automatic flags
Privacy-Aware Video Anomaly Detection through Orthogonal Subspace ProjectionArchitecture & representation336 / 80 / 219.25 / 19.91No automatic flags
A Random Matrix Theory Perspective on the Consistency of Diffusion ModelsTheory & formal analysis525 / 033 / 08.8 / 40.91No automatic flagsSource correction retained
Equivalence of Context and Parameter Updates in Modern Transformer BlocksTheory & formal analysis255 / 17 / 09.08 / 13.91Figure numbering discontinuity; see source noteSource correction retained
Focus and Dilution: The Multi-stage Learning Process of AttentionTheory & formal analysis543 / 02 / 09.26 / 41.91No automatic flags
Which Algorithms Can Graph Neural Networks Learn?Theory & formal analysis762 / 25 / 89.18 / 61.91No automatic flags

Measurement method

Main-file measurements use published PMLR physical pages, size-aware mixed-case heading candidates and adaptive two-/one-column reading order. Attached appendix envelopes run from the supported boundary through the end of the file, including any internal bibliography. A long appendix is not additional allowed main text.

Caption labels are counted once even when a display has multiple panels; a repeated caption or reference to a table is not another table. Section envelopes can include floating displays, whitespace and references sharing a page. These are layout estimates rather than exact text proportions.

The nine Position Track papers remain a separate editorial contribution type. Absence of an architecture figure or numerical benchmark is not a defect in a position argument. Selected gallery examples teach display choices; they do not establish their corpus-wide frequency.

Section spans run from one detected top-level heading to the next in physical-page coordinates. They include whitespace and floats, not exact text lengths, occupied-ink area or official page-limit accounting. Their percentages use the entire inspected PDF; the title, abstract and teaser can precede the first heading.

Missing boundaries remain unresolved rather than zero. Separately inspected files have separate denominators. Caption counts are extraction-assisted; contents entries, repeated labels, references and image-only captions need source checks. Corrections remain associated with the exact source URL.

Contribution labels are editorial, not official tracks. Selected examples do not measure visual-style prevalence, and accepted-paper observations cannot show which writing choices caused acceptance.

Download structural metadata · Inspect individual sources and studies