16 papers
Routing Before Looking: Query-Adaptive Evidence Acquisition for Long-form Video Understanding
Tianyue Wang, Xuying Wu, Yuxiang Ma +7
Long-form video understanding remains challenging for video agents due to the mismatch between query demands and evidence acquisition strategies. Although recent planning-before-pe…
Continual Learning in Transition
Zhiyan Hou, Dan Zhang, Tao Feng +11
Classical continual learning (CL) has primarily focused on enabling models to update and retain knowledge through parameter-centric mechanisms, e.g., training strategies, architect…
DASH: Divergence-Adaptive Supervision Horizons for On-Policy Self-Distillation of Reasoning Models
ZhiYan Hou, Xinyu Tang, Hongyan An +9
Reinforcement learning with verifiable rewards (RLVR) improves the reasoning capabilities of large language models using automatically verifiable outcome signals, but these signals…
Don't Mix Rewards, Mix Policies: Policy Decomposition and Optimization for Multi-Reward RL
Ruiming Liang, Yi Zhong, Yizhen Yuan +6
Modern large language models (LLMs) are expected not just to answer correctly, but to adapt their behavior to different human values and use cases. As a result, multi-reward reinfo…
On-Policy Distillation with Curriculum Turn-level Guidance for Multi-turn Agents
Gengsheng Li, Mao Zheng, Mingyang Song +8
Multi-turn agents that plan, invoke tools, and interact with environments offer a promising paradigm for solving complex tasks, yet their capabilities typically rely on very large…
ResMerge: Residual-based Spectral Merging of Large Language Models
Yandu Sun, Zhiyan Hou, Hongyan An +6
Model merging offers a training-free way to combine multiple post-trained expert models, but merging experts obtained through reinforcement learning (RL) remains challenging. Exist…
ST-Prune: Training-Free Spatio-Temporal Token Pruning for Vision-Language Models in Autonomous Driving
Lin Sha, Haiyun Guo, Tao Wang +4
Vision-Language Models (VLMs) have become central to autonomous driving systems, yet their deployment is severely bottlenecked by the massive computational overhead of multi-view c…
PLUME: Latent Reasoning Based Universal Multimodal Embedding
Chenwei He, Xiangzhao Hao, Tianyu Yang +6
Universal multimodal embedding (UME) maps heterogeneous inputs into a shared retrieval space with a single model. Recent approaches improve UME by generating explicit chain-of-thou…
Unifying Group-Relative and Self-Distillation Policy Optimization via Sample Routing
Gengsheng Li, Tianyu Yang, Junfeng Fang +6
Reinforcement learning with verifiable rewards (RLVR) has become a standard paradigm for post-training large language models. While Group Relative Policy Optimization (GRPO) is wid…
ReCALL: Recalibrating Capability Degradation for MLLM-based Composed Image Retrieval
Tianyu Yang, Chenwei He, Xiangzhao Hao +6
Composed Image Retrieval (CIR) aims to retrieve target images based on a hybrid query comprising a reference image and a modification text. Early dual-tower Vision-Language Models…
WISER: Wider Search, Deeper Thinking, and Adaptive Fusion for Training-Free Zero-Shot Composed Image Retrieval
Tianyue Wang, Leigang Qu, Tianyu Yang +4
Zero-Shot Composed Image Retrieval (ZS-CIR) aims to retrieve target images given a multimodal query (comprising a reference image and a modification text), without training on anno…
TRACE: Task-Adaptive Reasoning and Representation Learning for Universal Multimodal Retrieval
Xiangzhao Hao, Shijie Wang, Tianyu Yang +3
Universal Multimodal Retrieval requires unified embedding models capable of interpreting diverse user intents, ranging from simple keywords to complex compositional instructions. W…
R-Diverse: Mitigating Diversity Illusion in Self-Play LLM Training
Gengsheng Li, Jinghan He, Shijie Wang +7
Self-play bootstraps LLM reasoning through an iterative Challenger-Solver loop: the Challenger is trained to generate questions that target the Solver's capabilities, and the Solve…
Active Zero: Self-Evolving Vision-Language Models through Active Environment Exploration
Jinghan He, Junfeng Fang, Feng Xiong +5
Self-play has enabled large language models to autonomously improve through self-generated challenges. However, existing self-play methods for vision-language models rely on passiv…
PASs-MoE: Mitigating Misaligned Co-drift among Router and Experts via Pathway Activation Subspaces for Continual Learning
Zhiyan Hou, Haiyun Guo, Haokai Ma +3
Continual instruction tuning (CIT) requires multimodal large language models (MLLMs) to adapt to a stream of tasks without forgetting prior capabilities. A common strategy is to is…
MLLM-CTBench: A Benchmark for Continual Instruction Tuning with Reasoning Process Diagnosis
Haiyun Guo, Zhiyan Hou, Yandu Sun +6
Continual instruction tuning(CIT) during the post-training phase is crucial for adapting multimodal large language models (MLLMs) to evolving real-world demands. However, the progr…