11 papers
UI-MOPD: Multi-Platform On-Policy Distillation for Unified GUI Agents
Niu Lian, Tongbo Chen, Zhehao Yu +8
Recent advances in multimodal foundation models and agent systems have driven GUI agents from single-platform task execution toward cross-platform interaction. However, unified mul…
Balancing Efficiency and Efficacy: Training-Free Attention-Guided Switching Between Explicit and Latent Thoughts for MLLMs
Haoqian Kang, Liupeng Li, Kuofeng Gao +5
Reasoning in Multimodal Large Language Models (MLLMs) requires both fine-grained visual perception and rigorous logical deduction. Explicit text-based Chain-of-Thought (CoT) is com…
Unlocking Proactivity in Task-Oriented Dialogue
Azure Zhang, Ning Gao, Yuqin Dai +7
Proactive task-oriented dialogue (TOD), such as outbound sales, demands a persuasive agent that actively probes the user's concerns and steers the conversation toward acceptance wi…
CVSearch: Empowering Multimodal LLMs with Cognitive Visual Search for High-Resolution Image Perception
Liupeng Li, Haoqian Kang, Zhenyu Lu +4
High-resolution (HR) image perception presents a key bottleneck for multimodal large language models (MLLMs). While visual search offers a promising solution, existing methods stru…
SegCompass: Exploring Interpretable Alignment with Sparse Autoencoders for Enhanced Reasoning Segmentation
Zhenyu Lu, Liupeng Li, Jinpeng Wang +4
While large language models provide strong compositional reasoning, existing reasoning segmentation pipelines fail to transparently connect this reasoning to visual perception. Cur…
Tailoring Teaching to Aptitude: Direction-Adaptive Self-Distillation for LLM Reasoning
Hongbin Zhang, Chaozheng Wang, Kehai Chen +4
On-policy self-distillation (OPSD) is an emerging LLM post-training paradigm in which the model serves as its own teacher: conditioned on privileged information such as a reference…
SAGE: A Service Agent Graph-guided Evaluation Benchmark
Ling Shi, Yuqin Dai, Ziyin Wang +7
The development of Large Language Models (LLMs) has catalyzed automation in customer service, yet benchmarking their performance remains challenging. Existing benchmarks predominan…
Reinforcing Real-world Service Agents: Balancing Utility and Cost in Task-oriented Dialogue
Ning Gao, Wei Zhang, Yuqin Dai +6
The rapid evolution of Large Language Models (LLMs) has accelerated the transition from conversational chatbots to general agents. However, effectively balancing empathetic communi…
ContextRL: Enhancing MLLM's Knowledge Discovery Efficiency with Context-Augmented RL
Xingyu Lu, Jinpeng Wang, YiFan Zhang +12
We propose ContextRL, a novel framework that leverages context augmentation to overcome these bottlenecks. Specifically, to enhance Identifiability, we provide the reward model wit…
VLM as Policy: Common-Law Content Moderation Framework for Short Video Platform
Xingyu Lu, Tianke Zhang, Chang Meng +14
Exponentially growing short video platforms (SVPs) face significant challenges in moderating content detrimental to users' mental health, particularly for minors. The dissemination…
InstructEngine: Instruction-driven Text-to-Image Alignment
Xingyu Lu, Yuhang Hu, YiFan Zhang +9
Reinforcement Learning from Human/AI Feedback (RLHF/RLAIF) has been extensively utilized for preference alignment of text-to-image models. Existing methods face certain limitations…