RESEARCH
Selected from 20+ peer-reviewed publications across AI, statistics, and economics.
AI & MACHINE LEARNING
When evaluators cry wolf
EMNLP • under review • 2026
Systematic prompt enhancement & ensemble selection approach to improve LLM evaluator signal quality.
Generalization in neural networks: a broad survey
Neurocomputing • 2025
Identifies persistent overfitting in vision models’ use of contextual cues, like mistaking a lion on a city street for a dog. Applies SHAP to clarify the factors driving vision & LLM-based outputs.
arXiv • GitHub
Problem-dependent attention & effort in neural networks
Image & Vision Computing • 2023
Confidence-based ensemble routing of easy vs. hard cases at inference time. Cuts inference costs by 80% while maintaining accuracy.
arXiv • GitHub
Effects of FOMC communications on US Treasury rates
EMNLP • 2016
Applies topic-driven NLP to estimate effects of Federal Reserve communications on Treasury rates with 93% accuracy.
STATISTICS, CAUSAL INFERENCE, & COMPUTATION
Forbidden knowledge and specialized training
The American Statistician • 2023
Identifies key sources of overfitting in regression and introduces new metric to predict out-of-sample error. Outperforms existing approach by up to 45% on simulated and brain MRI data.
arXiv • GitHub
Optimal bandwidth for kernel regression
IEEE PDCO • 2017
Efficient cross-validation algorithms for nonparametric bandwidth selection, optimizing and reusing intermediate steps and parallelizing calculation on a GPU for a combined 7x acceleration.
Preprint • GitHub
Airbag regulations and the value of life
American Economic Journal: Economic Policy • 2015
Large-scale study of causal impacts of air bags on fatalities and consumer demand, funded by $120k NIH grant. Extensive panel dataset compiled from disparate industry sources, covering 300 million autos and 0.7 million fatalities.