Our alumnus Dr. Pei Kexin currently serves as a Neubauer Family Assistant Professor of Computer Science at the University of Chicago. He discovered his passion for research during his studies at HKBU. After graduating from HKBU, he completed his PhD at Columbia University in the City of New York, the United States.
Today, his research sits at the exciting intersection of security, software engineering, and machine learning. His work focuses on data-driven programme analysis and developing machine learning models that can reason about programme structure and behaviour, enabling precise and efficient detection and fixing of software bugs and vulnerabilities.
His works have earned him numerous accolades, including the Best Paper Award in Symposium on Operating Systems Principles (SOSP), a Distinguished Artifact Award, recognition as a Communications of the Association for Computing Machinery (CACM) Research Highlight, and Cyber Security Awareness Week (CSAW) Applied Research Competition Runner-Up. He has also collaborated with the Learning for Code team at Google DeepMind, building cutting-edge programme analysis tools powered by large language models.
Discovering a Passion for Research at HKBU
During his time at HKBU, he found inspiration in both experiences and mentors that ultimately shaped his pursuit of research in computer science. He was fortunate to join the summer research programme under the guidance of Professor Haibo Hu, Professor Jianliang Xu, and Professor Xiaowen Chu. While working in the database group led by Professor Hu and Professor Xu, he collaborated closely with PhD and MPhil students and received close supervision from the faculty, particularly Professor Haibo Hu.
The defining moment came when he independently developed a new algorithm that improved the performance of the system they implemented by orders of magnitude. That achievement was not only technically rewarding but also a turning point, as it revealed to him a genuine passion for research.
For him, the research resources in HKBU are abundant, and transdisciplinary learning is more than a trend. In fact, research opportunities nowadays often arise at the intersection of multiple disciplines. Fresh perspectives from different disciplines can often drag the mindset that is too used to a single field, so it becomes an “echo chamber”. He finds it extremely effective in bringing insights from one field to another, where one discipline often becomes a “disruptive technology” to the other. In particular, his research has always been on the intersection of two camps in AI: neuro and symbolic, where both have complementary attributes critical to building a trustworthy system. These two camps were historically flourishing in different fields. His research bridges these two philosophies in different disciplines and leads to impactful findings and results.
Reflecting on this journey, he advises students to keep exploring research opportunities with various faculty members, remain curious about solving hard problems, stay open-minded about broad disciplines, and explore opportunities at their intersections.
Inspiration for Bridging Disciplines
His research bridges security, software engineering, and machine learning, and the inspiration to focus on this intersection came from the growing complexity of modern software systems. As these systems scale, it becomes increasingly difficult to fully understand and ensure their security and reliability. Traditionally, computer scientists have used strict, rule-based methods to analyse software, but these approaches can struggle with things like natural language (for example, comments written by developers) and often don’t handle large, complicated programmes well.
Machine learning offers a different way to tackle these challenges. While it can be less precise or harder to explain compared to traditional methods, it can handle ambiguity and scale much better. His research is about finding ways to combine both approaches—using the strengths of machine learning and traditional reasoning methods—to solve important problems in areas like autonomous vehicles, medical diagnosis, robotics, and systems that connect computers with the real world.
To put it simply, machine learning helps us “reason” about software in a way that’s similar to how humans do. For instance, when people read a piece of code, they don’t always follow every detail; instead, they use intuition, look at key operations, and read comments to get a general sense of what the code does. Machine learning can imitate this process, analysing both the structure of the programme and any natural language hints to make guesses about its behaviour.
The real challenge is finding the right balance between precise, “rule-based reasoning” and the more flexible, “intuitive reasoning” that machine learning provides. By integrating both reasoning approaches, he hopes to make software safer and more reliable, ultimately helping solve real-world problems where security and accuracy are critical.
With strong code reasoning capabilities, machine learning models can quickly identify vulnerabilities in large-scale software, analyse malicious behaviours of the malware, and even fix them without expensive human expert intervention. This can greatly improve the security and reliability of the software, which has taken over our daily lives.
Building Trustworthy AI for a Human-Centred Future
He envisions the “AI of the future” as a force that will significantly increase human productivity in problem-solving and in creating new worlds. The difficulty of many tasks will be redefined, and the demand for many services will increase because of the increasing accessibility, instead of being replaced and diminished. Research and coding will be significantly accelerated, but the capabilities to vet and validate AI-generated outputs will become even more critical.
What excites him most is the opportunity to build trustworthy AI systems capable of making safe decisions, ensuring that the entire digital system empowering our world is reliable and secure. This is also his greatest concern – how can we ensure AI systems are trustworthy, responsible, secure, and reliable?
In the future, he believes that researchers and developers will be most valuable not for writing every line of code themselves, but for knowing how to think creatively with AI, clearly explain what they want systems to do, and make good decisions about design, safety, and real-world impact. As AI takes over routine tasks, human strengths like problem-solving, communication, critical thinking, and understanding how complex systems fit together become even more important.
He believes students can prepare by learning the basics of coding and AI tools, practising how to collaborate with AI rather than rely on it, and building strong foundations in reasoning, ethics, and domain knowledge. In short, the future belongs to people who can work with AI thoughtfully, not just use it.
[Source:
The Buddy Post]
