Ph.D. student in the Machine Learning Department at Carnegie Mellon University, advised by Prof. Artur Dubrawski.
Research
Democratizing expertisein hospitals, and foundation models.
Detecting hospital outbreaks with multimodal models trained on routinely available multimodal data: MALDI-TOF mass spectra, antimicrobial resistance profiles, and electronic health records. The aim is to reduce dependency on the whole-genome sequencing gold standard, so that specialist-grade detection stays affordable for every hospital.
Improving LLM reasoning through better extraction and use of experiential abstractions, via in-context learning or post-training. The aim is to distill expert tips and past trajectories in the form of abstractions into open-source models and allow self-improvement.
Education
Carnegie Mellon University, School of Computer Science
Tsinghua University — Yao Class, Institute for Interdisciplinary Information Sciences · GPA 3.91/4.00
Industry
TikTok
Modeling user preference over professionally generated content (PGC) to improve recommendation performance.
Publications
Notes to Self: Can LLMs Benefit from Experiential Abstractions?
Towards Practical Multimodal Outbreak Detection
Exploring the Utility of MALDI-TOF Mass Spectrometry and Antimicrobial Resistance in Hospital Outbreak Detection
Depth as a Scaling Vector: Simple Pruning and Evaluation of Emergent Abilities in Pruned LLMs
LayerMerge: Modality-Agnostic Depth Pruning for Efficient Foundation Model Deployment
From Aggregation to Guidance: Strategies for Personalized Federated Fine-Tuning of Foundation Models
Bridging the Utility Gap Between MALDI-TOF and WGS for Affordable Outbreak Cluster Detection
Multimodal Structure Preservation Learning
A Probabilistic Knowledge Graph Approach for Target Identification
Improving Target-disease Association Prediction through a Graph Neural Network with Credibility Information
† equal contribution · * corresponding author