Chengshuai Zhao

Ph.D. Student

School of Computing and Augmented Intelligence

Email: czhao93 [at] asu [dot] edu

Office: BYENG 561

Address: 699 S Mill Ave, Tempe, AZ 85281

I am Chengshuai Zhao (he/him), a second-year Ph.D. student in Computer Science at Arizona State University (ASU). I am currently a member of the Data Mining and Machine Learning Lab (DMML), advised by Prof. Huan Liu.

I completed my undergraduate studies at Huazhong Agricultural University (HZAU) and the University of California, Irvine (UCI). During that time, I conducted research in the BioMedical Big Data Mining Lab, advised by Prof. Wen Zhang. I also worked as a research intern in the AI Research and Innovation Laboratory at Westlake University, collaborating closely with Prof. Jun Xia and Prof. Stan Z. Li.

At the core of my research lies a fascination with the interplay between observation and understanding: how learning enables us to see beyond the apparent and uncover truths that reshape our perspective. My work spans Data Mining, AI for Science, Representation Learning, and Large Language Models, through which I aim to develop systems that are more generalizable, transparent, and capable of revealing knowledge at the frontier of human perception.

News

Aug 05, 2025	One paper for CoT reasoning of LLMs has been released on arXiv.
Aug 04, 2025	One paper for AI4Edu has been accepted by CIKM 2025 demo track.
May 15, 2025	One paper for multi-agent LLMs has been accepted by the ACL 2025 main conference.

Latest Posts

Aug 17, 2025	How to Make Good Figures for Scientific Papers

Selected Publications

arXiv
Is Chain-of-Thought Reasoning of LLMs a Mirage? A Data Distribution Lens

Chengshuai Zhao, Zhen Tan, Pingchuan Ma, and 5 more authors

arXiv preprint arXiv:2508.01191, 2025

Abs arXiv Bib HTML PDF Code

Chain-of-Thought (CoT) prompting has been shown to improve Large Language Model (LLM) performance on various tasks. With this approach, LLMs appear to produce human-like reasoning steps before providing answers (a.k.a., CoT reasoning), which often leads to the perception that they engage in deliberate inferential processes. However, some initial findings suggest that CoT reasoning may be more superficial than it appears, motivating us to explore further. In this paper, we study CoT reasoning via a data distribution lens and investigate if CoT reasoning reflects a structured inductive bias learned from in-distribution data, allowing the model to conditionally generate reasoning paths that approximate those seen during training. Thus, its effectiveness is fundamentally bounded by the degree of distribution discrepancy between the training data and the test queries. With this lens, we dissect CoT reasoning via three dimensions: task, length, and format. To investigate each dimension, we design DataAlchemy, an isolated and controlled environment to train LLMs from scratch and systematically probe them under various distribution conditions. Our results reveal that CoT reasoning is a brittle mirage that vanishes when it is pushed beyond training distributions. This work offers a deeper understanding of why and when CoT reasoning fails, emphasizing the ongoing challenge of achieving genuine and generalizable reasoning.
@article{zhao2025chain, title = {Is Chain-of-Thought Reasoning of LLMs a Mirage? A Data Distribution Lens}, author = {Zhao, Chengshuai and Tan, Zhen and Ma, Pingchuan and Li, Dawei and Jiang, Bohan and Wang, Yancheng and Yang, Yingzhen and Liu, Huan}, journal = {arXiv preprint arXiv:2508.01191}, year = {2025}, }