Linxin Song

Linxin Song is a second-year Computer Science Ph.D. student at the University of Southern California, co-advised by Prof. Jieyu Zhao and Prof. Yue Wang. Before that, he was a M.Eng student at the Graduate School of Creative Science and Engineering, Waseda University (早稲田大学) in Tokyo, supervised by Prof. Masayuki Goto. He also collaborates closely with Jieyu Zhang, whose work centers on interactive and data-centric AI/ML.

Research Interests

I work on multimodal language models and the digital agentic systems they power, spanning data curation, post-training methods, and end-to-end agent design. My current focus is building robust, efficient computer-use and coding agents. I’m also deeply interested in embodied AI, and I’m looking for opportunities to transfer paradigms between physical agents and digital agents.

Selected Publications

(* denotes equal contribution)

Reinforcement Finetuning

• Experiential Reinforcement Learning
  Taiwei Shi, Sihao Chen, Bowen Jiang, Linxin Song, Longqi Yang, Jieyu Zhao
• The Hallucination Tax of Reinforcement Finetuning
  Linxin Song*, Taiwei Shi*, Jieyu Zhao
  EMNLP 2025 (findings)
  Covered by: MarkTechPost
• Efficient Reinforcement Finetuning via Adaptive Curriculum Learning
  Taiwei Shi, Yiyang Wu, Linxin Song, Tianyi Zhou, Jieyu Zhao
  Preprint

Agentic AI

• CoAct-1: Computer-using Multi-Agent System with Coding Actions
  Linxin Song, Yutong Dai, Viraj Prabhu, Jieyu Zhang, Taiwei Shi, Li Li, Junnan Li, Silvio Savarese, Zeyuan Chen, Jieyu Zhao, Ran Xu, Caiming Xiong
  ICLR 2026.
  Covered by: MarkTechPost | VentureBeat

• Adaptive In-conversation Team Building for Language Model Agents
  Linxin Song*, Jiale Liu*, Jieyu Zhang, Shaokun Zhang, Ao Luo, Shijian Wang, Qingyun Wu, Chi Wang
  AIA @COLM 2025.

Language Model Evaluation

• NEW! The Blind Spot of Agent Safety: How Benign User Instructions Expose Critical Vulnerabilities in Computer-Use Agents
  Xuwei Ding*, Skylar Zhai*, Linxin Song*, Jiate Li, Taiwei Shi, Nicholas Meade, Siva Reddy, Jian Kang, Jieyu Zhao
  Preprint
• NEW! Video-Based Reward Modeling for Computer-Use Agent
  Linxin Song, Jieyu Zhang, Huanxin Sheng, Taiwei Shi, Gupta Rahul, Yang Liu, Ranjay Krishna, Jian Kang, Jieyu Zhao
  Preprint

• Discovering Knowledge Deficiencies of Language Models on Massive Knowledge Base
  Linxin Song, Xuwei Ding, Jieyu Zhang, Taiwei Shi, Ryotaro Shimizu, Rahul Gupta, Yang Liu, Jian Kang, Jieyu Zhao
  COLM 2025 | Webpage
  
• Explaining Length Bias in LLM-Based Preference Evaluations
  Zhengyu Hu, Linxin Song, Jieyu Zhang, Zheyuan Xiao, Jingang Wang, Zhenyu Chen, Hui Xiong
  EMNLP 2025 (findings)

Before PhD

• Attributed Synthetic Data Generation for Zero-shot Image Classification
  Shijian Wang, Linxin Song, Ryotaro Shimizu, Masayuki Goto, Hanqian Wu
  ICME 2025 (Oral)
• Offline Training of Language Model Agents with Functions as Learnable Weights
  Shaokun Zhang*, Jieyu Zhang*, Jiale Liu, Linxin Song, Chi Wang, Ranjay Krishna, Qingyun Wu
  ICML 2024.
  Covered by: The Forbes Article
• ProVision: Programmatically Scaling Vision-centric Instruction Data for Multimodal Language Models
  Jieyu Zhang, Le Xue, Linxin Song, Jun Wang, Weikai Huang, Manli Shu, An Yan, Zixian Ma, Juan Carlos Niebles, silvio savarese, Caiming Xiong, Zeyuan Chen, Ranjay Krishna, Ran Xu
  Covered by: VentureBeat | MarkTechPost
• Better Explain Transformers by Illuminating Important Information
  Linxin Song, Yan Cui, Ao Luo, Freddy Lecue, Irene Li
  EACL 2024 (findings).
• Investigating the Scaling Effect of Instruction Templates for Training Multimodal Language Model
  Shijian Wang*, Linxin Song*, Jieyu Zhang, Ryotaro Shimizu, Ao Luo, Li Yao, Cunjian Chen, Julian McAuley, Haiqian Wu
• NLPBench: Evaluating Large Language Models on Solving NLP Problems
  Linxin Song, Jieyu Zhang, Lechao Cheng, Pengyuan Zhou, Tianyi Zhou, Irene Li
  ITIF @ NeurIPS 2023.
• Leveraging Instance Features for Label Aggregation in Programmatic Weak Supervision
  Jieyu Zhang*, Linxin Song*, Alexander Ratner
  AISTATS 2023.
• Adaptive Ranking-based Sample Selection for Weakly Supervised Class-imbalanced Text Classification
  Linxin Song, Jieyu Zhang, Tianxiang Yang, Masayuki Goto
  EMNLP 2022 (findings).

Teaching

• (TA) DSCI-250: Introduction to Data Science, 2024 Fall
• (TA) DSCI-566: Deep Learning and its Applications, 2025 Spring

Internships

• Microsoft Research
  Incoming
• Google
  2026.02-now
  Mentor: Jiefeng Chen, Jinsung Yoon
• Salesforce Research - Research Intern
  2025.05-2026.01
  Mentor: Yutong Dai, Zeyuan Chen, Ran Xu

Professional Services

• Maintainer of AG2 (Autogen).
• Reviewer (for multiple years): WACV, KDD, NeurIPS, DMLR, ICLR, AISTATS, ACL, EMNLP, etc