ICML 2026

6,487 papers

Top Research Directions

Research directions most represented by ICML 2026 papers.

Scaling Language Model Reasoning Beyond Verifiable Rewards 79
Group Relative Policy Optimization for Large Reasoning Models 63
General Purpose Reasoning from Specialized Reinforcement Learning 62
Overcoming the Iterative Inference Latency of Diffusion Models 50
Shifting from Static Benchmarks to Dynamic Evaluation 47
Controlling Language Models Through Latent Space Interventions 45
Resolving Superposition with Sparse Autoencoders 42
Autonomous Code Refinement via Execution Feedback Loops 41
The Performance Cliff in Ultra Low Bit Quantization 41
Activation Steering for Inference Time Model Control 39
Mitigating Expert Collapse in Sparse Mixture of Experts 37
Proxy Objective Exploitation in Language Model Alignment 37
The Decoupling of Visual Perception and Logical Reasoning 36
Mitigating Computational Overthinking in Large Reasoning Models 35
Input Adaptive Compute Allocation for Language Models 33

Top Authors

AuthorPapers
Junchi Yan19
Min Zhang19
Tat-Seng Chua15
Haoxuan Li14
Philip Torr14
Yong Li14
Kun Zhang13
Li Shen13
Tianlong Chen13
Xiang Wang13
Xin Geng13
Yejin Choi13
Bo An12
Mihaela van der Schaar12
Xunliang Cai12
Bo Han11
Ge Zhang11
Han Liu11
Jiang Bian11
LEI BAI11

Impactful Papers

Papers from ICML 2026 that shaped research clusters.

Foundational

Papers

Title
What Makes a Strong Model? A Unified Spectral Analysis of Knowledge Transfer over High-dimensional Linear Regression
What Makes Effective Supervision in Latent Chain-of-Thought: An Information-Theoretic Analysis
What Makes Synthetic Data Effective in Image Segmentation
What Makes Value Learning Efficient in Residual Reinforcement Learning?
What Preferences Can—and Cannot—Predict in Multi-Agent Online Learning
What Really Improves Mathematical Reasoning: Structured Reasoning Signals Beyond Pure Code
What Reward Structure Enables Efficient Sparse-Reward RL? A Proof-of-Concept with Policy-Aware Matrix Completion
What You Think is What You See: Driving Exploration in VLM Agents via Visual-Linguistic Curiosity
When Actions Go Off-Task: Detecting and Correcting Misaligned Actions in Computer-Use Agents
When Agents Go Rogue: Activation-Based Detection of Malicious Behaviors in Multi-Agent Systems
When AI Agents Compete for Jobs: Strategic Capabilities and Economic Dynamics of AI Labour Markets
When AI Benchmarks Plateau: A Systematic Study of Benchmark Saturation
When and How Human Curation Backfires: Preference Alignment under Multi-Model Self-Consuming Loop
When Attributes Disagree: Gradient Conflict in Image Aesthetic Assessment
When Benign Inputs Lead to Severe Harms: Eliciting Unsafe Unintended Behaviors of Computer-Use Agents
When Can We Trust Survival Model Evaluation ?
When Can You Poison Rewards? A Tight Characterization of Reward Poisoning in Linear MDPs
When Data Is Scarce: Scaling Sparse Language Models with Repeated Training
When Diffusion Language Models Hesitate: Detecting and Correcting Visual Hallucinations via Confidence Fluctuation
When Distance Distracts: Representation Distance Bias in BT-Loss for Reward Models
When Do Diffusion Models Learn to Generate Multiple Objects?
When Does Adaptation Win? Scaling Laws for Meta-Learning in Quantum Control
When Does Predictive Inverse Dynamics Outperform Behavior Cloning?
When Does Sparsity Mitigate the Curse of Depth in LLMs
When Do Graph Foundation Models Transfer? A Data-Centric Theory
When Do Hallucinations Arise? A Graph Perspective on the Evolution of Path Reuse and Path Compression
When Drafts Evolve: Speculative Decoding Meets Online Learning
When Embedding-Based Defenses Fail: Rethinking Safety in LLM-Based Multi-Agent Systems
When Generalized Zero-Shot Learning Meets PU Learning: A Plug-and-Play Framework for Seen-Class Bias Mitigation
When Is Rank-1 Enough? Geometry-Guided Initialization for Parameter-Efficient Fine-Tuning
When Labelers Stay Silent: The Power of Ties in Cost-Effective Preference Learning
When LLMs Develop Languages: Symbolic Communication for Efficient Multi-Agent Reasoning
When LLMs Encounter Open-world Graph Learning: A Fresh View on Unlabeled Data Uncertainty
When Model Merging Breaks Routing: Training-Free Calibration for MoE
When More Data Doesn't Help: Limits of Adaptation in Multitask Learning
When More Experts Hurt: Underfitting in Multi-Expert Learning to Defer
When Planning Fails Despite Correct Execution: On Epistemic Calibration for LLM-Based Multi-Agent Systems
When Preference Labels Fall Short: Aligning Diffusion Models from Real Data
When RAG Hurts: Diagnosing and Mitigating Attention Distraction in Retrieval-Augmented LVLMs
When Random Saliency Looks Trained: Architectural Center Bias in CNN Interpretability
When Replanning Becomes the Bottleneck: Budgeted Replanning for Embodied Agents
When RL Meets Adaptive Speculative Training: A Unified Training-Serving System
When Sample Selection Bias Precipitates Model Collapse
When Search Goes Wrong: Red-Teaming Web-Augmented Large Language Models
When Shared Knowledge Hurts: Spectral Over-Accumulation in Model Merging
When Simple Problems Wear Complex Costumes: Improving Efficiency in LRM's Adaptive Reasoning
When Single Answer Is Not Enough: Rethinking Single-Step Retrosynthesis Benchmarks for LLMs
When Softmax Fails at the Top: Extreme‑Value Corrections for InfoNCE
When Tabular Foundation Models Meet Strategic Tabular Data: A Prior Alignment Approach
When the Prompt Becomes Visual: Vision-Centric Jailbreak Attacks for Large Image Editing Models

« prev Page 128/130 next »