Jianmin Guo

Jianmin Guo

Senior Research Engineer

2012 Lab, Huawei Technologies Co., Ltd.

Biography

My name is Jianmin Guo (郭建敏). I am a senior research engineer of 2012 Lab, Huawei in Beijing, focusing on AI/LLM security research and its applications, such as adversarial robustness, safety assessment and content moderation, etc. I obtained my Ph.D. degree at School of Software, Tsinghua University in 2022, during which my research topics are deep learning testing and adversarial attacks. I received my BSc degree in software engineering from Beijing University of Posts and Telecommunications in 2017, ranking first and graduate with honors. In my future work, I will devote myself to high-value research projects and make impactful contributions.

Interests

  • AI & LLM security
  • Adversarial robustness
  • LLM safety assessment
  • Unsafe data generation & moderation

Education

  • PhD in Software Engineering, 2017-2022

    Tsinghua University

  • BSc in Software Engineering, 2013-2017

    Beijing University of Posts and Telecommunications

Experience

 
 
 
 
 

Senior research engineer

Huawei

Aug 2022 – Present Beijing

Core works:

  • Train a specialized LLM to assess response safety, evaluating whether outputs from tested LLMs meet predefined safety criteria when handling unsafe queries.
  • Detect deepfake and AIGC images & videos with classifiers of multiple features, as well as reconstruction-based methods utilizing diffusion models.
  • Construct unsafe multimodal data generation pipeline and design a semantics-aware moderation framework with MLLMs.
  • Defend black-box query attack on face recognition with probabilistic fingerprints.
 
 
 
 
 

Research intern

Tencent

Apr 2021 – Dec 2021 Beijing
Train in-vehicle speech enhancement model, where audios with noise (particularly in-vehicle situation) will sound clear after speech enhancement.
 
 
 
 
 

Ph.D. student

Tsinghua University

Sep 2017 – Jul 2022 Beijing

Representative works:

  • RNN-Test: a general-purpose adversarial testing framework for seq2seq tasks in RNN systems. It produces adversarial examples by maximizing RNN state inconsistency against their inner dependencies. We also design two state-based coverage metrics for RNNs, enabling RNN-Test to generate adversarial examples and improve the coverage.
  • DLFuzz: the first DL testing framework combined with fuzz testing. DLFuzz mutates seed inputs to maximize neuron coverage and the prediction difference between original and mutated inputs, with multiple neuron selection strategies. Its paper published on ESEC/FSE 2018 is widely recognized with >300 citations.
 
 
 
 
 

Research assistant

Nanyang Technological University

Jan 2017 – May 2017 Singapore
Static program analysis: extract and assess program metrics of tested software programs to identify vulnerability-prone functions and CVEs, assisted with fuzzing tools.
 
 
 
 
 

Undergraduate student

Beijing University of Posts and Telecommunications

Sep 2013 – Jul 2017 Beijing

Main achievements:

  • GPA: 3.9/4.0, Rank: 1/123
  • National scholarship
  • Meritorious Winner of MCM/ICM
  • Second Prize in Intel Cup National Collegiate Software Innovation Contest
  • First Prize in Student Innovation Contest

Services

Reviewer

Journal of Software (软件学报,CCF-A)

Teaching assistant

In Tsinghua University:

  • Fundamentals of Computer Culture (Fall 2019, Fall 2020)
  • Operating System (Spring 2018, Fall 2018, Spring 2019)
  • Intelligent Engineering Systems (Fall 2017)

Contact

  • guojm17@tsinghua.org.cn
  • Beijing, 100095