10 papers · 1 filter
Scaling Agentic Data for Long-Horizon Terminal Intelligence
Shun Zou, Ziyu Ma, Yi Zou +6
Terminal agents are moving beyond local operations toward completing real-world professional tasks. Short trajectories can demonstrate commands and tool calls, but they do not full…
Towards Unbiased On-Policy Distillation for Block Diffusion Language Models
Zaiquan Yang, Fei Wei, Yong Wang +6
On-policy distillation (OPD) has emerged as an effective post-training paradigm for language models, with recent efforts extending it to block diffusion language models (BDLMs). Ho…
Beyond Entropy: Self-Diagnostic Multi-Role Token Optimization for Video Reasoning
Yudong Han, Yong Wang, Zaiquan Yang +4
Reinforcement learning with verifiable rewards has substantially advanced multimodal reasoning, yet it remains fundamentally limited by ambiguous token-level credit assignment. Whi…
The Right Future for Action: Learning Action-Relevant Predictive States in World Action Models
Qiwen Gu, Jifan Li, Bingjie Gao +4
Generation-free world action models (WAMs) retain future-video prediction during training but act from internal video features at inference, leaving unclear what these features sho…
IntBMoE: Integrating Block-Level Conditioning into Expert Composition for Full-Participation Mixture-of-Experts
Ran Cheng, Longfei Xu, Zheng Liu +2
Mixture-of-Experts (MoE) scales capacity, but existing designs cannot set three quantities independently. For a single token, participation is how many experts contribute knowledge…
Do Video Generators Track the World Across Segments? A Benchmark and Method for World-State Reasoning in Video Continuation
Yingmao Miao, Pengfei Zhang, Chaoran Xu +5
Video generators build long videos by composing shorter parts, either by generating segments one after another or by autoregressively extending chunks. Each new part usually depend…
SkillForge: Evolving Verifiable Skills for Reinforcement Learning Agents
Shidong Yang, Ziyu Ma, Tongwen Huang +5
Large language model (LLM) agents are trained with reinforcement learning (RL) for complex decision-making tasks. However, most RL-trained agents remain episodic and cannot accumul…
BlockPilot: Instance-Adaptive Policy Learning for Diffusion-based Speculative Decoding
Hao Zhang, Yiming Hu, Yong Wang +3
Speculative decoding accelerates inference by using a lightweight draft model to generate candidate tokens in parallel, and are then verified by the target model, enabling lossless…
DEvo: Dual Difficulty-Aware Self-Evolution for Data-Efficient Reinforcement Learning
Ru Zhang, Renda Li, Ziyu Ma +4
Reinforcement learning (RL) has demonstrated potential for enhancing reasoning in large language models (LLMs). However, effective RL training, which requires medium-difficulty tra…
Visual Enhanced Depth Scaling for Multimodal Latent Reasoning
Yudong Han, Yong Wang, Zaiquan Yang +3
Multimodal latent reasoning has emerged as a promising paradigm that replaces explicit Chain-of-Thought (CoT) decoding with implicit feature propagation, simultaneously enhancing r…