regsearch
Hybrid search over genomics literature — full-text and embedding retrieval fused with reciprocal rank fusion, plus a fine-tuned reranker for +41% MRR.
Patrick Yang
CS undergrad at Cornell working on numerical stability of deep RL training. Also interested in computational biology; currently building deep learning models for gene expression at Yale.
Selected work
Rational Activations for Genomics
A matched-capacity controlled study testing whether learned rational activation functions beat GELU on sequence-to-expression regression, across 9 activation arms and 315 training runs.
Heterogeneity Score
Attention-pooled CNN predicting chromatin variability from DNA sequence alone across 5.6M genomic regions, with an analysis of why the task resists learning.
Experience
-
Cornell University — Bindel Group
August 2026 – PresentUndergraduate Researcher
- Leading a project with Prof. David Bindel on the numerical stability of deep RL training — studying target-network divergence via fixed-point iteration and spectral stability analysis, and deriving the target update rate instead of hand-tuning it.
-
Yale University — Garg Lab
January 2026 – PresentResearch Software Engineer
- Built the training and validation infrastructure behind the lab's activation-function studies: fault-tolerant SLURM grids on preemptible A100 GPUs, and data-quality checks that caught two silent corruption bugs in a 5.6M-region dataset.
Education
-
Cornell University, College of Engineering
Expected December 2028B.S. Computer Science
- Coursework: Reinforcement Learning, Machine Learning, Numerical Analysis, Probability & Statistics, Object-Oriented Programming & Data Structures.
Awards
-
Adjutant General's Coin for Excellence
June 2025Connecticut National Guard
-
Army ROTC Scholarship
March 2025U.S. Army Cadet Command
-
CT Boys State — Governor
June 2024American Legion