PUBLICATIONS

Work from the TraceLite team on agent evaluation, multimodal reasoning, and dialogue agents.

  1. Trial by Trace: Benchmarking AI Agent Harnesses for Reliability and Failure Attribution

    TraceLite AI

    TraceTrial-Bench · 95 tasks across seven harness reliability dimensions, with separate task-outcome and harness assessments.

    Technical reportSeptember 2026

  2. MM-VeriRubric: Verifiable Evidence-Grounded Rubrics for Multimodal Reward Models

    Zhiyong Cao, Bo Wang, Jintao Li, Shuxia Liu, Kairan Zhang, Aijia Wang, Yi-Liang Wei, Sicheng Zhou

    EMNLP 2026

  3. Disentangled Privileged States for Simulator-Assisted Multi-Turn Dialogue Policy Training

    Zhiyong Cao, Dunqiang Liu, Sicheng Zhou, Lin Cong

    EMNLP 2026 Findings

  4. Absorbing Gradient Conflicts: Modeling Semantic Variance via Kent Distributions for Cross-Modal Hashing

    Hengjie Zhu, Dayan Wu, Zihao Zhang, Xinze Liu, Jingxuan Yu, Peng Fu, Zheng Lin, Weiping Wang

    IJCAI 2026

  5. SimRPD: Optimizing Recruitment Proactive Dialogue Agents through Simulator-Based Data Evaluation and Selection

    Zhiyong Cao, Dunqiang Liu, Qi Dai, Haojun Xu, Huai Yuen Khor, Hao Wang, Huan He, Yafei Liu, Ke Ma, Ruqian Shi, Sicheng Zhou, Sijia Yao

    OralACL 2026

  6. Look Less, Reason More: Rollout-Guided Adaptive Pixel-Space Reasoning

    Xuchen Li, Xuzhao Li, Jiahui Gao, Renjie Pi, Shiyu Hu, Wentao Zhang

    ACL 2026

  7. Select Less, Reason More: Prioritizing Evidence Purity for Video Reasoning

    Xuchen Li, Xuzhao Li, Shiyu Hu, Kaiqi Huang

    CVPR 2026

  8. CausalStep: A Benchmark for Explicit Stepwise Causal Reasoning in Videos

    Xuchen Li, Xuzhao Li, Shiyu Hu, Kaiqi Huang, Wentao Zhang

    OralAAAI 2026

  9. Judging by Evidence: Typed Evidence Acquisition for Agent-as-a-Judge

    Anonymous submission

    Under Review

  10. ChameleonBench: Evaluating the Cross-Environment Robustness Across Heterogeneous Systems

    Anonymous submission

    Under Review

  11. STRATA: Stratified Trajectory-to-Knowledge Augmentation for Terminal Agents

    Anonymous submission

    Under Review

  12. EviSeek: An Evidence-Centric Agentic Framework for Embodied Active Information Seeking

    Anonymous submission

    Under Review