Heng Ji

Multimodal Policy Internalization for Conversational Agents

Perception-Aware Policy Optimization for Multimodal Reasoning

Energy-Based Transformers are Scalable Learners and Thinkers

Constructive Distortion: Improving MLLMs with Attention-Guided Image Warping

BEAT: Visual Backdoor Attacks on VLM-based Embodied Agents via Contrastive Trigger Learning

SWERank: Software Issue Localization with Code Ranking

mCLM: A Modular Chemical Language Model that Generates Functional and Makeable Molecules

Neural Synchrony Between Socially Interacting Language Models

ProteinZero: Self-Improving Protein Generation via Online Reinforcement Learning

Beyond Log Likelihood: Probability-Based Objectives for Supervised Fine-Tuning across the Model Capability Continuum

Analyzing and Internalizing Complex Policy Documents for LLM Agents

Teaching Language Model to Act Efficiently

RM-R1: Reward Modeling as Reasoning

TrustGen: A Platform of Dynamic Benchmarking on the Trustworthiness of Generative Foundation Models

ReIn: Conversational Error Recovery with Reasoning Inception

UserRL: Training Interactive User-Centric Agent via Reinforcement Learning

MM-PoisonRAG: Disrupting Multimodal RAG with Local and Global Knowledge Poisoning Attacks

Geometri-Disentangelment Unlearning

CURV: Enhancing Chart Understanding through Visual Grounded Reasoning

ERA: Transforming VLMs into Embodied Agents via Embodied Prior Learning and Online Reinforcement Learning

Self-Improving LLM Agents at Test-Time

EmbodiedBench: Comprehensive Benchmarking Multi-modal Large Language Models for Vision-Driven Embodied Agents

PARTONOMY: Large Multimodal Models with Part-Level Visual Understanding

ToolRL: Reward is All Tool Learning Needs

OpenHands: An Open Platform for AI Software Developers as Generalist Agents

Variational Supervised Contrastive Learning

DyMU: Dynamic Merging and Virtual Unmerging for Efficient Variable-Length VLMs

Contrastive Visual Data Augmentation

Eliminating Position Bias of Language Models: A Mechanistic Approach

CoRNStack: High-Quality Contrastive Data for Better Code Retrieval and Reranking

SyncMind: Measuring Agent Out-of-Sync Recovery in Collaborative Software Engineering

Geometry Informed Tokenization of Molecules for Language Model Generation

Geometry Informed Tokenization of Molecules for Language Model Generation

Fragment and Geometry Aware Tokenization of Molecules for Structure-Based Drug Design Using Language Models

Weak-to-strong Generalization via Formative Learning from Student Demonstrations & Teacher Evaluation

Model Extrapolation Expedites Alignment

Establishing Knowledge Preference in Language Models

Visually Descriptive Language Model for Vector Graphics Reasoning

Scaling Laws for Predicting Downstream Performance in LLMs

Training Software Engineering Agents and Verifiers with SWE-Gym

FARM: Functional Group-Aware Representations for Small Molecules

MentalArena: Self-play Training of Language Models for Diagnosis and Treatment of Mental Health Disorders