Xiangjun Fan

Think Then Embed: Generative Context Improves Multimodal Embedding

Beyond Reward Hacking: Causal Rewards for Large Language Model Alignment

Exploring System-1 and System-2 Communication for Latent Reasoning in LLMs

ReMix: Reinforcement Routing for Mixtures of LoRAs in LLM Finetuning

DAG-MoE: From Simple Mixture to Structural Aggregation in Mixture-of-Experts

Self-Supervised LLM-based Hard Negative Sampling for Two-Tower Model

S'MoRE: Structural Mixture of Residual Experts for Parameter-Efficient LLM Fine-tuning

Quantifying Generalization Complexity for Large Language Models