CV

Basics

Name Chengshuai Zhao
Label Ph.D. Student in CS @ ASU | DMML Lab | LLM Reasoning, Safety & Privacy | Data Mining | AI4Science
Email czhao93 [at] asu [dot] edu
Phone +1 (949) 227-7633
Url https://chengshuaizhao0.github.io/

Education

  • 2024.01 - Present

    Tempe, AZ, USA

    Ph.D.
    Arizona State University
    Computer Science
    • Advisor: Prof. Huan Liu
    • Expected graduation: May 2028
  • 2022.09 - 2023.06

    Irvine, CA, USA

    B.S.
    University of California, Irvine
    Data Science
  • 2019.09 - 2023.06

    Wuhan, China

    B.S.
    Huazhong Agricultural University
    Data Science
    • GPA: 3.97/4.0, Ranking: 1/63
  • 2018.09 - 2023.06

    Wuhan, China

    B.S.
    Huazhong Agricultural University
    Life Science
    • GPA: 3.87/4.0, Ranking: 2/364

Work

  • 2024.01 - Present

    Tempe, AZ, USA

    Graduate Research Assistant
    Data Mining and Machine Learning Lab (DMML), Arizona State University
    Advised by Prof. Huan Liu.
    • LLM post-training & reasoning: reasoning distillation via explanatory data synthesis and RL with structured rewards; an RL-optimized knowledge-graph retriever for reasoning-augmented generation; a controlled synthetic environment for pre-training LLMs from scratch and probing CoT reasoning.
    • LLM/MLLM security, privacy & safety: bilevel-optimized knowledge-poisoning attacks on RAG systems and covert malicious-agent injection into MLLM societies; adversarial data synthesis that protects data from unauthorized fine-tuning of LLMs and VLMs; RL-driven defenses that shield proprietary LLMs from unauthorized API-based knowledge distillation.
    • RAG & LLM agent applications: knowledge-graph-grounded response verification in RAG for domain-specific question answering; a multi-agent RL framework in Graph-RAG for multi-hop reasoning; LLM multi-agent frameworks with human-in-the-loop collaboration for large-scale content analysis.
    • LLM evaluation: LLM-as-a-judge and the detectability of LLM-generated judgments; detection and evaluation of LLM-generated content; evaluation of CoT reasoning under distribution shift.
    • Causality, interpretability, efficiency & recommendation: causality-guided representation learning for robust generalization across domains, styles, and modalities; concept bottleneck models for interpretable open-ended grading; low-rank node-level graph diffusion for few-shot node classification; diffusion language models for bundle recommendation.
  • 2022.07 - 2022.09

    Hangzhou, China

    Research Intern
    Center for AI Research and Innovation (CAIRI), Westlake University
    Advised by Prof. Stan Z. Li.
    • Molecular foundation models: self-supervised pre-training of GNNs for molecules with a context-aware atom-level tokenizer and masked atom modeling with contrastive learning.
  • 2019.10 - 2022.09

    Wuhan, China

    Undergraduate Researcher
    BioMedical Big Data Mining Lab (BBDM), Huazhong Agricultural University
    Advised by Prof. Wen Zhang.
    • Graph representation learning & link prediction: contrastive self-supervised GNNs, equivariant positional-encoding GNNs, and graph-embedding ensembles for predicting molecular interactions.

Awards

Skills

Research
LLM Post-Training (SFT, RL, Knowledge Distillation)
LLM Reasoning
LLM/MLLM System Security, Privacy & Safety
LLM Agents & Retrieval-Augmented Generation
LLM Evaluation
Explainable AI & Model Interpretability
Graph Representation Learning (Link Prediction, Node Classification, Graph Generation)
Causal Representation Learning
Programming Languages
C
C++
Java
Python
R
Go
Scala
SQL
Verilog HDL
Frameworks
PyTorch
HF Transformers
DeepSpeed
vLLM
TensorFlow
Scikit-learn
Hadoop
Spark
Databases
Oracle
MySQL
SQLite
MongoDB
Tools
Git
Docker
Conda
Overleaf
Operating Systems
Ubuntu
CentOS
macOS
MS Windows

Interests

Large Language Models
Post-Training & Reasoning
Agents & RAG
Evaluation & LLM-as-a-Judge
Trustworthy AI
LLM/MLLM Security, Privacy & Safety
Explainability & Interpretability
Causal Representation Learning
Graph Learning & AI for Science
Graph Representation Learning
Molecular Foundation Models
Biomedical Data Mining

Languages

English
Professional working proficiency
Chinese
Native proficiency