Peng Kuang | PhD Student, UIUC (DREAM Lab)

Peng Kuang is a PhD student at the University of Illinois Urbana-Champaign, working in Prof. Haohan Wang’s DREAM Lab.

He studies how to develop effective and efficient supervision for evolving AI systems, and how those systems can learn and adapt from feedback. His work aims to make AI more reliable and capable while keeping the cost of supervision practical and scalable as systems take on increasingly complex tasks.

Research interests:

LLM agents
AI supervision
Efficiency

Also at krumperisok.github.io

Publications

Selected papers. Full list on Google Scholar.

Published

KV-PRM: Efficient Process Reward Modeling via KV-Cache Transfer for Multi-Agent Test-Time Scaling

Peng Kuang, Haibo Jin, Xiaoyu Han, Yanli Wang, Xiaopeng Yuan, Ye Yu, Kaidi Xu, Haohan Wang

Conference on Neural Information Processing Systems (NeurIPS), 2026 · project · arXiv

Makes process reward modeling much more efficient in long-context settings, so multi-agent systems scale better at test time.

Agent Primitives: Reusable Latent Building Blocks for Multi-Agent Systems

Haibo Jin, Peng Kuang, Ye Yu, Xiaopeng Yuan, Haohan Wang

International Conference on Machine Learning (ICML), 2026 · arXiv

Optimal Aggregation of LLM and PRM Signals for Efficient Test-Time Scaling

Peng Kuang, Yanli Wang, Xiaoyu Han, Yaowenqi Liu, Kaidi Xu, Haohan Wang

International Conference on Learning Representations (ICLR), 2026 · arXiv

Echo: Reverberation-Based Fast Black-Box Adversarial Attacks on Intelligent Audio Systems

Meng Xue, Kuang Peng, Xueluan Gong, Qian Zhang, Yanjiao Chen, Routing Li

Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies (IMWUT), 2023 · paper

Preprints

Composing Task-specific Agent Harnesses at Test Time with Reusable Primitives

Peng Kuang, Haibo Jin, Dehao Wu, Feiyang Deng, Xiaopeng Yuan, Jerry Wang, Haohan Wang

Preprint, 2026 · arXiv

BabelArena: A Large-Scale Multilingual Benchmark for LLM Agents

Peng Kuang, Yuchun Fan, Jiangnan Li, Minghao Wu, Jialong Tang, Hao-Ran Wei, Weixuan Wang, Jianhong Tu, Baosong Yang, Tong Xiao

Preprint, 2026 · arXiv

Beyond Surface Statistics: Robust Conformal Prediction for LLMs via Internal Representations

Yanli Wang, Peng Kuang, Xiaoyu Han, Kaidi Xu, Haohan Wang

Preprint, 2026 · arXiv

TIM-PRM: Verifying Multimodal Reasoning with Tool-Integrated PRM

Peng Kuang, Xiangxiang Wang, Wentao Liu, Jian Dong, Kaidi Xu

Preprint, 2025 · arXiv

Rethinking Debiasing: Real-World Bias Analysis and Mitigation

Peng Kuang, Zhibo Wang, Zhixuan Chu, Jingyi Wang, Kui Ren

Preprint, 2024

Research Experience

DREAM Lab, University of Illinois Urbana-Champaign

Research Assistant, Prof. Haohan Wang’s group

Mar 2025 – Present

Process reward modeling, test-time scaling, and reusable building blocks for LLM-based multi-agent systems and agent harnesses.

Qwen Team, Alibaba Group

Research Intern

Mar 2026 – Aug 2026

Multilingual LLM agents: adapting executable agent benchmarks to multilingual settings, RL training in multilingual agent environments, and the BabelArena benchmark.