I spent three days walking the poster halls at ICML 2026 in Seoul, South Korea, and these are the 114 posters that made me stop, read, and take a photo. They lean heavily toward LLM agents, reasoning, memory, RL, and AI-for-science — a decent snapshot of what the community was excited about this year.
Click any poster to zoom in and read it in full.
Neuro-anatomical Inductive Priors for Sphere-based Brain Decoding
Multimodal Scaling Laws for Task & Data Optimized Models of Visual Cortex
Image-to-Brain Signal Generation for Visual Prosthesis with CLIP-Guided Multimodal Diffusion Models
Identifying Connectivity Distributions from Neural Dynamics using Flows
WorldCache: Accelerating World Models for Free via Heterogeneous Token Caching
CSG: Cognitive Structure Generation for Intelligent Education
MVR-cache: Optimizing Semantic Caching via Multi-Vector Retrieval and Learned Prompt Segmentation
Pianist Transformer: Expressive Piano Performance Rendering via Scalable Self-Supervised Pre-Training
Simultaneous Speech-to-Speech Translation Without Aligned Data
RELO: Reinforcement Learning to Localize for Visual Object Tracking
SpatialReward: Bridging the Perception Gap in Online RL for Image Editing via Explicit Spatial Reasoning
Active Exploring like a Pigeon: Reinforcing Spatial Reasoning via Agentic Vision-Language Models
DR Tulu: Reinforcement Learning with Evolving Rubrics for Deep Research
Retaining by Doing: The Role of On-Policy Data in Mitigating Forgetting
SERA: Soft-Verified Efficient Repository Agents
Synthesizing Multimodal Geometry Datasets from Scratch and Enabling Visual Alignment via Plotting Code
On Information Self-Locking in Reinforcement Learning for Active Reasoning of LLM Agents
Not All Rollouts are Useful: Down-Sampling Rollouts in LLM Reinforcement Learning
Stop Anthropomorphizing Intermediate Tokens as "Reasoning/Thinking" Traces!
REAL: Resolving Knowledge Conflicts in Knowledge-Intensive VQA via Reasoning-Pivot Alignment
The Appeal and Reality of Recycling LoRAs with Adaptive Merging
Intrinsic Credit Assignment for Long Horizon Interaction
AdaHC: Accelerating Multi-Token Prediction with Adaptive Head Chunking with Pipeline Parallelism
Agentic Model Predictive Questioning Control in Visual Design
Demystifying Scientific Problem-Solving in LLMs by Probing Knowledge and Reasoning
CausalArmor: Efficient Indirect Prompt Injection Defense via Causal Attribution
Stop Automating Peer Review Without Rigorous Evaluation
Hugging Carbon: Quantifying the Training Carbon Emissions of AI Models at Scale
Position: Peer Review in ML/AI Conferences Should…
Agentic Proposing: Enhancing LLM Reasoning via Compositional Skill Synthesis
Your Wasserstein Ball is too Round! Loss-Aware DRO via Trainable Optimal Transport Ambiguity Sets
From Statics to Dynamics: Physics-Aware Image Editing with Latent Transition Priors
MOOSE-Star: Unlocking Tractable Training for Scientific Discovery by Breaking the Complexity Barrier
Scalable Option Learning in High-Throughput Environments
UltraHorizon: Benchmarking Agent Capabilities in Ultra Long-Horizon Scenarios
Thinking with Video: Video Generation as a Promising Multimodal Reasoning Paradigm
AI Can Learn Scientific Taste
What Makes Value Learning Efficient in Residual RL?
Conversation for Non-verifiable Learning: Self-Evolving LLMs through Meta-Evaluation
Cycle-of-Science: Reliable Reasoning through Counterfactual Verification for Agent Decision Making
EDCO: Dynamic Curriculum Orchestration for Domain-specific LLM Fine-tuning
Learning to Share: Selective Memory for Efficient Parallel Agentic Systems
Breaking the Exploration Bottleneck: Rubric-Scaffolded RL for Open-Ended LLM Reasoning
Emergence of Exploration in Policy Gradient RL via Retrying
PerceptionRubrics: Calibrating Multimodal Evaluation to Human Perception
Zero-shot Active Mapping via Fused 360-BEV Representations and Vision-Language Models
MetaStreet: Semi-Supervised Multimodal Learning for Street-Level Socioeconomic Prediction
Code2Video: A Code-centric Paradigm for Educational Video Creation
Think in Latent, Explain in Language: Self-Explainable Latent Reasoning
Think Deep, Not Just Long: Measuring LLM Reasoning Effort via Deep-Thinking Tokens
ParaTool: Shifting Tool Representations from Context to Parameters
Prompt Optimization with Minimal Unlabeled Input via Meta-Reasoning
ReJump: A Tree-Jump Representation for Analyzing and Improving LLM Reasoning
MemoryBench: A Benchmark for Memory and Continual Learning in LLM Systems
Diversity over Frequency: Rethinking Tool Use in Visual Chain-of-Thought Agents
