cv

PhD Candidate in Causal Machine Learning

• Major
  • Epidemiology and Biostatistics
• Secondary Field
  • Computer Science
• Advisors
  • Dr. Andrew Beam
  • Dr. James Robins

Machine Learning (Graduate level)

Quantitative Methods for NLP (Graduate level)

Geometric Methods for Machine Learning (Graduate level)

Stochastic Methods for Data Analysis, Inference and Optimization (Graduate level)

Algorithms for Data Science (Graduate level)

Linear Algebra and Learning from Data

Introduction to Functional Analysis

Probability (Graduate level)

Statistical Inference I, II (Graduate level)

Advanced Regression and Statistical Learning (Graduate level)

System Development for Computational Science

High Performance Computing for Science and Engineering (Graduate level)

Causal Inference (Graduate level)

MARS Fellowship

• Completed competitive 3-month research fellowship focused on AI safety and alignment. Conducted independent research on detecting pathological reasoning behaviors in large language models under mentorship from University of Cambridge with collaborators from UCL and Mila. Developed novel evaluation methodologies for Chain-of-Thought pathologies with implications for AI system reliability and trustworthiness.

Technical AI safety Fellowship

• 8-week reading group on AI safety, covering topics such as neural network interpretability, robustness, and alignment.

PhD Researcher

• Aim 1
  • Engineered a noval DAG-aware transformer model to precisely estimate causal effects, addressing foundational challenges in unifying causal effect estimation under various scenarios.
• Aim 2
  • Integrated doubly robust estimators into Monte Carlo Tree Search (MCTS), enabling large language models to perform complex, multi-step reasoning and planning with higher accuracy in real-world scenarios.
• Aim 3
  • Developed and implemented comprehensive evaluation metrics to identify and monitor pathologies in Chain-of-Thought (CoT) reasoning across large language models, including post-hoc, internalized, and encoded reasoning patterns. Collaborated on fine-tuning open-weight LLM models to elicit internalized and encoded reasoning capabilities in model organisms.

Biostatistician

• Collaborated with clinical researchers, biostatisticians, and data scientists to harness electronic health record (EHR) data for machine learning applications.
• Developed and deployed predictive models to forecast patient outcomes, enabling data-driven decision-making in healthcare settings.
• Led cross-functional efforts to integrate machine learning workflows into clinical practice, optimizing efficiency and enhancing patient care outcomes.

DAG-aware-transformer

• Our codes for the paper "DAG-aware Transformer for Causal Inference" published in NeurIPS 2024, CRL workshop.

COT Health Metrics

• A library for evaluating reasoning pathologies in large language models (LLMs), including post-hoc, internalized, and encoded reasoning flaws. Provides health metrics such as reliance, paraphrasability, and substitutability to assess and analyze these pathologies.