memory representations · long-term agents · continual learning · reliable NLP

Google Scholar  ·  Email  ·  UNC-Chapel Hill

I study how AI systems represent, retrieve, and use memory over long interactions. I am an undergraduate at the University of North Carolina at Chapel Hill studying Computer Science and Economics, with a minor in Mathematics.

• Long-term memory representations: how bounded memory systems should preserve source-grounded evidence as interaction histories grow.
• Answerability space: how retrieval, contextual proposition encoding, and closure can construct a sufficient answer witness before Reader use.
• Retrieval signals from frozen models: whether internal state changes can reveal answer-relevant evidence without a learned retrieval head.

• Bridge Starvation — identifies a reproducible failure mode of single-pass multi-hop retrieval and introduces a bounded retrieve-bridge-retrieve repair; presented at the ICML 2026 FAGEN Workshop.
• Graph Tokens before Graph Propagation — uses proposition-level retrieval keys with deterministic source-chunk payloads; presented at the ICML 2026 GFM Workshop.
• SIFT — uses LLM-imagined scene differences to reweight features for low-resource tabular learning; oral presentation at the IEEE ICDM 2025 VISTA Workshop.

Six semesters as an Undergraduate Learning Assistant in econometrics, time series, and calculus, including review sessions and office hours.