AI & interpretability researcher · Staff Research Scientist at Voyage AI, training large embedding models
_{Previously ~4 years at Microsoft Research (Deep Learning group) · PhD in AI at UC Berkeley with Bin Yu}

🔍 interpretability  ·  💊 clinical decision rules  ·  🧠 semantic brain mapping  ·  🤖 LLMs

• imodels Interpretable, accurate, sklearn-compatible predictive modeling, built on decision trees and simple things that work JOSS 2021 FIGS·PNAS 2022 HSTree·ICML 2022
• imodelsX Interpretability for text and LLMs Aug‑imodels·Nat. Comms 2023 Tree‑Prompt·EMNLP 2023 iPrompt·ICLR wkshp 2023 SASC·NeurIPS wkshp 2023 QA‑Embs·NeurIPS 2024
• agentic-imodels Evolving agent-facing interpretability tools via autoresearch NeurIPS 2026
• adaptive-wavelets Replace neural nets with adaptive, interpretable wavelet models NeurIPS 2021
• deep-explanation-penalization Penalize neural-network explanations to align models with prior knowledge ICML 2020
• hierarchical-dnn-interpretations Hierarchical interpretations for neural network predictions ICLR 2019
• transformation-importance Feature importance for transformations ICLR wkshp 2020

• clinical-rule-development Building and vetting clinical decision rules with data and LLMs Intra‑abdominal rule·PLOS DH 2022 Patient perspectives·Nature SR 2025 CDI bias·npj Digit. Med. 2025
• HACHI Human-AI co-design of clinical models from EHR, building on BC-LLM (Bayesian concept bottlenecks with LLM priors) HACHI·npj Digit. Med. 2026 BC‑LLM·NeurIPS 2025
• clinical-self-verification Improving LLM-based clinical information extraction via self-verification
• clinical-rule-vetting General PECARN data preprocessing for rule development
• covid19-severity-prediction Extensive COVID-19 data + forecasting for counties and hospitals HDSR 2021

• automated-brain-explanations Building natural-language explanations of fMRI language responses GCT·Nature Neuroscience 2026 QA models·bioRxiv 2025
• interpretable-embeddings QA encoding models: model fMRI responses to verbal theories using LLM annotations NeurIPS 2024
• molecular-partner-prediction Predicting successful CME events using only clathrin markers

• token-economizer Experience-driven black-box cost control for coding agents (led by Aaron Li) arXiv 2026
• sim-dino Simplifying DINO via coding rate regularization (led by Ziyang Wu) ICML 2025
• pasta Post-hoc attention steering for LLMs (led by Qingru Zhang) ICLR 2024
• meta-tree Learning a decision tree algorithm with transformers (led by Yufan Zhuang) TMLR 2024
• mixture-of-inputs Text generation beyond discrete token sampling (led by Yufan Zhuang) NeurIPS 2025
• induction-gram Interpretable language modeling via induction-head ngram models (led by Eunji Kim & Sriya Mantena) NeurIPS 2025

• ml-ai notes Notes and resources on AI
• vflow Utilities for trustworthy data science JOSS 2021