CV

Ph.D. in Computer Science

M.S. in Electrical and Computer Engineering, Computer Vision

B.Eng. in Electronic Science and Technology

Applied Scientist Intern — Efficient LLM Reasoning

• Developed step-aware credit assignment for RL-based post-training, redistributing trajectory-level signals according to the contribution of individual reasoning steps.
• Reduced reasoning tokens by 24% on average on Qwen2.5-Math-7B while maintaining or improving accuracy over length-regularization, trace-compression, and fine-grained credit-assignment baselines.

Applied Scientist Intern — Agentic Conversational AI

• Developed workflow-based control mechanisms for tool-using conversational agents, representing procedures as structured memory with prerequisite-aware action specifications.
• Built an offline-to-online pipeline that induces workflows from historical trajectories and uses runtime progress alignment to guide agent execution.
• Improved agent performance and execution reliability on ToolSandbox and tau2-bench.

Software Engineer

Confidence-Orchestrated Self-Evolution against Uncertain LLM Feedback

• Bowen Wei, Nan Wang, Yuqing Zhou, Jinhao Pan, and Ziwei Zhu.

Fighting Spurious Correlations in Text Classification via a Causal Learning Perspective

• Yuqing Zhou and Ziwei Zhu.

Navigating the Shortcut Maze: A Comprehensive Analysis of Shortcut Learning in Text Classification by Language Models

• Yuqing Zhou, Ruixiang Tang, Ziyu Yao, and Ziwei Zhu.

A Generalized Propensity Learning Framework for Unbiased Post-Click Conversion Rate Estimation

• Yuqing Zhou, Tianshu Feng, Mingrui Liu, and Ziwei Zhu.

WISE-Flow: Workflow-Induced Structured Experience for Self-Evolving Conversational Service Agents

• Yuqing Zhou, Zhuoer Wang, Jie Yuan, Hong Wang, Samson Koelle, Ziwei Zhu, and Wei Niu.

Causally Calibrated Robust Classifier (CCR)

• Developed a group-label-free framework for worst-group robustness under distribution shift using causal feature selection and inverse propensity weighting.
• Achieved 70.7% worst-group accuracy on CivilComments, outperforming competing methods that require explicit group labels.

Shortcut Learning Analysis in Language Models

• Developed a taxonomy and controlled benchmarks for occurrence-, style-, and concept-based shortcuts.
• Evaluated BERT and Llama-family models, showing that scale alone does not consistently improve shortcut robustness.

Generalized Propensity Learning for Post-Click Conversion Rate Prediction

• Developed a generalized propensity-learning framework that reduces bias and variance in post-click conversion prediction.
• Improved performance by 7% in DCG@2 and 6% in Recall@2 on the Yahoo dataset.

CS 657 — Mining Massive Datasets

CS 584 — Theory and Applications of Data Mining

CS 455 — Computer Communications and Networking