7. "Interpreting Predictive Models for Human-in-the-Loop Analytics" (Joint work with O. Bastani and C. Kim)
Preliminary version appeared in FATML 2017.
*Previously circulated as "Interpreting Blackbox Models via Model Extraction."

Interpretability has become an important issue as machine learning is increasingly used to inform consequential decisions. We propose an approach for interpreting a blackbox model by extracting a decision tree that approximates the model. The algorithm avoids overfitting by actively sampling new training points using the blackbox model. We use this technique to interpret a random forest classifier for predicting diabetes risk. Physicians successfully used our interpretation to discover an unexpected causal issue in the diabetes classifer..

6. "Mostly Exploration-Free Algorithms for Contextual Bandits" (Joint work with M. Bayati and K. Khosravi)
Working Paper.

The contextual bandit literature has traditionally focused on the exploration-exploitation tradeoff. Exploration-free greedy policies are often desirable in practice when experimentation is prohibitively costly or unethical (e.g., clinical trials), but may be sub-optimal in general. We prove that, under some assumptions on the distribution of the contexts, the greedy policy is asymptotically rate-optimal for the two-armed contextual bandit. Next, we introduce a novel algorithm that monitors observed data while following a greedy policy to determine whether or not exploration is required.

5. "Online Decision-Making with High-Dimensional Covariates" (Joint work with M. Bayati)
Major Revision in Operations Research.
Winner of Pierskalla Award for Best Paper in Healthcare (2016), George Nicholson Student Paper Competition (2016), MSOM Student Paper Competition (2016) and IBM Service Science Best Student Paper Award (2016).

Big data has enabled decision-makers to personalize decisions at the individual-level. We present an efficient method to solve this problem in an online setting and a corresponding regret analysis. We provide both theoretical guarantees and empirical evidence that our algorithm significantly outperforms existing methods. Furthermore, we illustrate the practical relevance of our algorithm on a real clinical problem of warfarin dosing. Our algorithm correctly doses a majority of patients compared to both existing bandit methods as well as current physician practice.

4. "Analysis of Medicare Pay-for-Performance Contracts" (Joint work with M. Bayati, M. Braverman, R. Gummadi, and R. Johari)
Working Paper.

Medicare's pay-for-performance (P4P) programs are subject to a variety of practical and institutional constraints, such as the use of "small" performance-based incentives. We develop a framework based on a stylized principal-agent model to characterize the optimal P4P mechanism within any set of feasible mechanisms (this set can be flexibly modified to include institutional constraints). We apply our results to examine debated design choices in existing Medicare P4P programs, and offer several insights and policy recommendations.

3. "Evidence of Upcoding in Pay-for-Performance Programs," (Joint work with J. Goh and M. Bayati)
Accepted to Management Science.
Winner, 2015 Health Applications Society Best Student Paper Award.
News coverage: Science Daily.
*Previously circulated as "Evidence of Strategic Behavior in Medicare Claims Reporting."

Upcoding is the practice where medical providers alter claims to receive increased reimbursement. Prior studies on upcoding have been limited by unobserved confounders (e.g. provider quality and patient risk). We present a novel approach that exploits state-level variations in adverse event regulation and instrumental variables to detect upcoding at a national scale. We also make several policy recommendations for reducing upcoding.

2. "Creating Optical Vortex Modes with a Single Cylinder Lens," (Joint work with M. Cohen and J. Noe)
Proc. SPIE 7613 (2010).

1. "Multiplex coherent anti-Stokes Raman scattering (MCARS) for chemically sensitive, label-free flow cytometry," (Joint work with C. Camp, S. Yegnanarayanan, A. Eftekhar, and A. Adibi)
Opt. Express 17, 22879-22889 (2009).