Research Publications

Title: Stress-Testing Neural Network Verifiers with Provably Robust Instances

Submitted to NeurIPS 2026

Description: Neural network verifiers aim to provide formal guarantees on model behavior, but existing verification benchmarks are fundamentally limited by their lack of ground-truth labels. As a result, verifier evaluation relies on indirect heuristics, which prevents exact scoring and systematic study of verifier failure modes. We address this gap by introducing a reusable framework for generating verification instances whose ground-truth robustness labels are known a priori through analytic construction. 

Subject Areas:   AI/ML     Neural Network Verification

Title: Differentiable Optimization Layers for Guaranteed Fairness in Deep Learning

Journal: International Conference of Machine Learning (2026)

Year Published: 2026

Description: We introduce the concept of a "fairness layer": a differentiable optimization layer appended to a model's output layer that guarantees a chosen notion of output parity is satisfied when integrated into a neural network. Additionally, we introduce an online primal-dual inference algorithm that provides provable aggregate fairness guarantees for streaming predictions with arbitrarily small batch sizes, where traditional per-batch constraints become overly restrictive. Numerical experiments demonstrate the effectiveness of the fairness layer and associated algorithm, and theoretical analysis characterizes the layer's differentiability and stability properties during model training and backpropagation.

Subject Areas:   AI/ML    Interpretability and Explainability

Title: Public Health Factors Help Explain Cross Country Heterogeneity in Excess Death During the COVID19 Pandemic

Journal: Nature Scientific Reports

Year Published: 2023

Description: We gather and analyze data relating to excess death rates across countries for the COVID-19 pandemic. All covariates are split into "intrinsic" or "actionable" groups based on a country's ability to change them over the course of the pandemic. We implement a de-correlation process and build a gradient boosting model after a robust repeated cross validation framework to determine which covariates and interaction of covariates are important when predicting excess death. Our bootstrapped hypothesis test reveals actionable public health variables aid the prediction process beyond simply adding random variables. We investigate the performance of specific countries before and after considering actionable features.

Subject Areas:   AI/ML     Interpretability and Explainability

Title: A Cardinality Minimization Approach to Security-Constrained Economic Dispatch

Journal: IEEE Transactions on Power Systems

Year Published: 2022

Description: A new mathematical optimization problem formulation is constructed to simultaneously minimize the cost of running a power grid system while also minimizing the number of power lines that fail during contingency events. The new formulation is based on a cardinality optimization problem. This original problem is approximated using a continuous, convex function, and a difference-of-convex functions algorithm (DCA) is employed. The performance of this new model is thoroughly compared to various baseline models.

Subject Areas:   Mathematical Optimization

Title: Tractable Continuous Approximations for Constraint Selection via Cardinality Minimization

Journal: INFORMS Journal on Optimization

Year Published: 2025

Description: Multiple novel continuous models are constructed to solve cardinality minimization problems (CMP) where the goal is to optimize some objective function while simultaneously minimizing the number of soft constraints violated. Theory is developed to guarantee optimality conditions of the new approximations. Numerical experiments are developed which display the efficacy of the continuous approximation models in applications such as intensity-modulated radiation therapy (IMRT) and portfolio optimization.

Subject Areas:   Mathematical Optimization