Tianxin Wei's Preliminary Exam

Recent advances in large-scale machine learning and generative models have enabled intelligent systems to interact with users in increasingly complex and personalized ways. However, building effective personalized intelligence systems remains challenging due to evolving user preferences, heterogeneous data sources, and the need for models to continuously adapt while maintaining reliability and efficiency at scale. This dissertation studies principled frameworks for developing autonomous personalized intelligence systems that can learn from interactions and evolve over time. The research explores how structured modeling of user interactions can improve the robustness and generalization of personalization, how the capabilities of large foundation models can be leveraged to scale personalized intelligence, and how adaptive mechanisms such as agentic reasoning and test-time learning can enable systems to refine their behavior through ongoing interaction with users and environments. Together, these efforts aim to advance the foundations of personalized AI by developing learning paradigms that integrate interaction modeling, foundation models, and adaptive learning. The resulting frameworks provide both conceptual insights and practical methods toward building next-generation intelligent systems capable of sustained personalization and autonomous improvement.

Questions? Contact Tianxin Wei.