Research StatementMy long-term research goal is to develop human-centric AI that continuously understands an evolving world and supports human decision-making with timely, interpretable, and actionable intelligence. My earlier Ph.D. and postdoctoral work established the methodological foundation: continuous-time sequential models for irregular event streams, including temporal point processes, latent-state models, and sequential decision methods. My recent work at CUHK-Shenzhen builds on this foundation, shifting from modeling when and what events occur to understanding what event sequences reveal about human behavior. I view sequences as behavioral traces of actions, preferences, decisions, and interactions, and develop interpretable models that predict behavior, uncover mechanisms, and support decisions aligned with human needs. This work advances along three connected directions. 1. Interpretable temporal mechanisms from behavioral tracesMost sequence models focus on predicting future events without explaining the mechanisms that generate them. Building on temporal point processes, I develop models that combine statistical learning with symbolic rules, structured domain knowledge, and forward-chaining inference. These models capture irregular event dynamics, infer hidden states, and recover temporal logic that can be inspected, verified, and revised. I further study cognitively inspired models of behavior, including how people form internal world models, reuse learned habits, and switch between fast intuitive responses and slower deliberative reasoning. This direction moves beyond next-event prediction toward mechanism discovery in evolving systems such as clinical trajectories, treatment policies, and human action patterns. 2. Preference mechanisms behind human decisionsWhile temporal models explain how behavior evolves, this direction studies how context shapes what people choose and value. Choices depend on the available alternatives, prior interactions, framing, and individual heterogeneity. I develop interpretable neural and symbolic choice models that capture context effects, interactions among options, behavioral heterogeneity, and evolving preferences. I also study the statistical and optimization questions that arise when these models inform assortments, recommendations, and related decision-support interventions. This direction connects sequential behavior modeling with choice and preference learning, enabling AI systems to anticipate not only what people will do, but also what outcomes they value. 3. Collaborative AI for human-centered decision supportRather than replacing human expertise, I study how AI can serve as a complementary decision partner. Building on temporal mechanisms and preference models, this line investigates how AI should recommend, ask questions, trigger deeper deliberation, coordinate with humans, or identify the appropriate expert when decisions unfold over time. I develop algorithms that align predictions, explanations, and interventions with human intentions, preferences, and reasoning while accounting for uncertainty, bias, and cognitive limitations. In clinical decision support and other high-stakes settings, this work aims to create collaborative AI systems that know when to recommend, when to deliberate, and whom to consult. Together, these directions form a coherent research agenda that moves from event prediction to human understanding and, ultimately, to human-aligned AI support that is interpretable, adaptive, and useful for decisions that matter. Representative WorksThe papers below are representative rather than exhaustive. See my CV or Google Scholar for the complete publication list. Methodological FoundationsContinuous-time models for irregular event streams, latent-state dynamics, scalable learning, and diffusion-based generation.
Temporal Reasoning and Cognitive ModelsTemporal and neuro-symbolic mechanisms.
Cognitive and clinical extensions.
Choice and Preference ModelingInterpretable choice and preference models.
Statistical guarantees and decision optimization.
Collaborative AI and High-Stakes Decision Support
|