Yilun Du

Reasoning with Sampling: Your Base Model is Smarter Than You Think

World-In-World: World Models in a Closed-Loop World

Any-Order Flexible Length Masked Diffusion

Geometry-aware Policy Imitation

Inference-time scaling of diffusion models through classical search

Energy-Based Transformers are Scalable Learners and Thinkers

Product of Experts for Visual Generation

Self-Improving Loops for Visual Robotic Planning

OpenGLA: Topology and Task Adaptive Foundation Model for Power System Graph-Language Answering

Intra-Request Branch Orchestration for Efficient LLM Reasoning

Controllable Video Synthesis via Variational Inference

Implicit State Estimation via Video Replanning

SLM-MUX: Orchestrating Small Language Models for Reasoning

Equilibrium Matching: Generative Modeling with Implicit Energy-Based Models

Model-Based Diffusion Sampling for Predictive Control in Offline Decision Making

TOWARD MEMORY-AIDED WORLD MODELS: BENCHMARKING VIA SPATIAL CONSISTENCY

MoSEL: Modular Self-Reflective Learning for Embodied Decision-Making

Long-Text-to-Image Generation via Compositional Prompt Decomposition

MoTVLA: A Vision-Language-Action Model with Unified Fast-Slow Reasoning

Flow Equivariant World Modeling for Partially Observed Dynamic Environments

4D Latent World Model for Robot Planning

DiZCo: Planning Zero-Shot Coordination in World Models

Understanding the Design Space and Cross-Modality Transfer for Vision-Language Models

Selective Underfitting in Diffusion Models

From Score Distributions to Balance: Plug-and-Play Mixture-of-Experts Routing

Test-Time Graph Search for Goal-Conditioned Reinforcement Learning

Building Scalable Real-World Robot Data Generation via Compositional Simulation

Generalizable Reasoning through Compositional Energy Minimization

Generative Trajectory Stitching through Diffusion Composition

EvoLM: In Search of Lost Language Model Training Dynamics

Follow the Energy, Find the Path: Riemannian Metrics from Energy-Based Models

Grounding Video Models to Actions through Goal Conditioned Exploration

AuroraCap: Efficient, Performant Video Detailed Captioning and a New Benchmark

Multiagent Finetuning: Self Improvement with Diverse Reasoning Chains

COMBO: Compositional World Models for Embodied Multi-Agent Cooperation

History-Guided Video Diffusion

Looped Transformers for Length Generalization

Multi-Agent Verification: Scaling Test-Time Compute with Multiple Verifiers

Learning 3D Persistent Embodied World Models

Compositional Scene Understanding through Inverse Generative Modeling

Win Fast or Lose Slow: Balancing Speed and Accuracy in Latency-Sensitive Decisions of LLMs

MindJourney: Test-Time Scaling with World Models for Spatial Reasoning

Solving New Tasks by Adapting Internet Video Knowledge

AdaWorld: Learning Adaptable World Models with Latent Actions

Learning 4D Embodied World Models

SnapMem: Snapshot-based 3D Scene Memory for Embodied Exploration and Reasoning