Short Bio: I’m currently a Ph.D. candidate at Momentum Lab, Dept. of Computer Science at The University of Tokyo, working under the supervision of Prof. Lei Ma. My Ph.D. studies are supported by the IST-RA from the department and Research Fellowship for Young Scientists from JSPS. I chose to “master out” and have graduated from University of Alberta. My studies at UofA were supported by AMII. During my graduate studies, I was grateful to learn from Prof. Tianyi Zhang and Dr. Felix Juefei Xu. Before joining Momentum Lab, I was fortunate to work with Prof. Haoyu Wang, Prof. Dan Pei and Prof. Yuanchun Li.
Research Interests: The uncertainty and complexity inherent in AI-driven systems make it difficult to fully understand them from first principles, posing significant challenges for their trustworthy and efficient deployment. My long-term research goal is to advance the scientific understanding of AI systems from a software engineering perspective and, ultimately, to enable the systematic development of more trustworthy AI.
My past research is guided by the philosophy of decoding complexity through systematic interaction. While the global internals of modern AI systems often remain opaque, actionable insights and local structural knowledge can be uncovered through disciplined observational interaction, such as fuzzing, internal probing, and boundary exploration. By systematically investigating how these systems behave under diverse conditions, we extract informative clues about their reliability, limitations, and vulnerabilities. These insights enable rigorous characterization, targeted improvement, and more dependable deployment of AI systems, even without requiring complete interpretability of their underlying mechanisms.
Research Experience: I have experience on the trustworthiness and reliability assurance of complex AI systems, with a particular focus on foundation models and techniques for testing, analysis, monitoring, and verification. I also have experience developing human-computer interaction (HCI) solutions for machine learning development, as well as experience spanning the AI stack and blockchain technologies. During my Master’s and Ph.D. studies, I actively collaborated with industry partners on the design and deployment of AI applications, including foundation model-based systems.
Summary: We propose characterizing LLM behavior unsupervisedly by jointly analyzing internal neuron-level representations and external uncertainty signals. By capturing complementary behavioral cues, this approach enables more effective and sample-efficient testing.
Summary: The internal hidden states of a code model contain critical signals regarding the trustworthiness of its outputs. However, effectively leveraging these signals from modern, massive LLMs requires an approach that matches their sheer complexity. Because a model's generative power emerges from massive scale, we argue that its risk assessment framework should evolve to match this by applying scalable pre-training approaches to these highly informative internal layers.
Summary: This work represents an early exploratory work that leverages external uncertainty analysis to evaluate model trustworthiness, highlighting that the open-ended, generative mechanisms of LLMs require fundamentally different uncertainty measurements than the fixed-class probability outputs of classical DNNs.
Summary: This work demonstrates that for DNN models (more specifically, Transformer), systematic and rigorous exhaustive testing can move beyond empirical evaluations to provide verifiable, certified results regarding a model's robustness against patch attacks.
Summary: This work performed a large-scale measurement study of the EOSIO ecosystem through graph analysis. It reveals EOSIO's superficial prosperity and pervasive malicious activities.