Anish Agarwal
Assistant Professor, Columbia University
My current research interests are in: (1) causal machine learning; (2) reinforcement learning; (3) high-dimensional statistics. I am particularly interested in data-driven decision-making in engineering and social systems using tools from econometrics and machine learning.
Brief Bio
Most recently, I was a research scientist at Amazon, Core AI, before which I was a postdoctoral research fellow at the Simons Institute at UC Berkeley. I did my PhD at MIT EECS, where I was fortunate to be advised by Alberto Abadie, Munther Dahleh, and Devavrat Shah. I've spent time at Microsoft Research. I've also served as a technical consultant to TauRx Therapeutics and Uber Technologies on questions related to experiment design and causal inference. Before the PhD, I was a management consultant at Boston Consulting Group. I got my undergrad and master's degrees from Caltech under the supervision of Mani Chandy and Adam Wierman.
Selected Awards
Best Paper Award, USENIX NSDI, 2023
Best Student Paper Award (Government Statistics, Survey Research Methods, & Social Statistics), American Statistical Association, 2023
ACM Sigmetrics Outstanding Doctoral Dissertation Prize (2nd Place), 2022
INFORMS George B. Dantzig Dissertation Award (Honorable Mention), 2022
Simons-UC Berkeley Postdoctoral Fellowship, 2022
Papers
Causal Inference, Online Learning, & Combinatorial Interventions
A. Agarwal, A. Agarwal, L. Masoero, J. Whitehouse. "Multi-Armed Bandits with Network Interference".
A. Agarwal, A. Agarwal, S. Vijaykumar. "Synthetic Combinations: A Causal Inference Framework for Combinatorial Interventions".
Conference: NeurIPS, 2023.
A. Agarwal, S. Cen, D. Shah, C. Yu. "Network Synthetic Interventions: A Causal Inference Framework for Panel Data with Network Interference".
Causal Inference, Reinforcement Learning, & Building Simulators from Data
A. Agarwal, V. Syrgkanis. "Synthetic Blip Effects: A Causal Inference Framework for Linear Dynamical Systems".
A. Alomar, P. Hamadanian, A. Esfahany, A. Agarwal, M. Alizadeh, D. Shah, “CausalSim: A Causal Framework for Unbiased Trace-Driven Simulation".
Conference: NSDI, 2023. Best Paper Award.
A. Agarwal, A. Alomar, V. Alumootil, D. Shah, D. Shen, Z. Xu, C. Yang. "PerSim: Offline Reinforcement Learning via Personalized Simulators".
Conference: NeurIPS, 2021. (Slides)
Causal Inference & Medicine
D. Shen, A. Agarwal, V. Misra, B. Schelter, D. Shah, H. Shiells, C. Wischik. "Personalized Predictions w. Population Level Data: Alzheimer's Disease".
C. Squires, D. Shen, A. Agarwal, D. Shah, C. Uhler. "Causal Imputation via Synthetic Interventions".
Conference: Causal Learning and Reasoning (CLeaR), 2022.
Causal Inference & Matrix/Tensor Completion
A. Abadie, A. Agarwal, R. Dwivedi, A. Shah. "Doubly Robust Inference in Causal Latent Factor Models".
A. Agarwal, M. Dahleh, D. Shah, D. Shen. "Causal Matrix Completion". (Video, Slides, Open-source)
Conference: Conference on Learning Theory (COLT), 2023.
A. Agarwal, D. Shah, D. Shen. "Synthetic Interventions". (Video, Slides, Open-source)
Causal Inference & Mechanism Design
D. Ngo, K. Harris, A. Agarwal, V. Syrgkanis, S. Wu. "Incentive-Aware Synthetic Controls: Counterfactual Estimation via Incentivized Exploration".
K. Harris, A. Agarwal, C. Podimata, S. Wu. "Strategyproof Decision-Making in Panel Data Settings and Beyond".
Conference: ACM Sigmetrics, 2023.
High-Dimensional Statistics, Principal Component Regression, & Error-in-Variables Regression
A. Agarwal, K. Harris, J. Whitehouse, S. Wu. "Adaptive Principal Component Regression with Applications to Panel Data".
Conference: NeurIPS, 2023.
A. Agarwal, R. Singh. "Causal Inference with Corrupted Data: Measurement Error, Missing Values, Discretization, and Differential Privacy".
American Statistical Association. Best Student Paper Award.
A. Agarwal, D. Shah, D. Shen. "On Model Identification and Out-of-Sample Prediction of PCR: Applications to Synthetic Controls".
A. Agarwal, D. Shah, D. Shen, D. Song. "On Robustness of Principal Component Regression".
Journal: Journal of the American Statistical Association, 2021.
Conference: NeurIPS, 2019. Oral Presentation. (Video, Slides)
High-Dimensional Statistics &Time Series Analysis
A. Agarwal, A. Alomar, D. Shah. "On Multivariate Singular Spectrum Analysis and its Variants".
A. Agarwal, A. Alomar, D. Shah. "tspDB: Time Series Predict Database".
Conference: NeurIPS Demo Track, 2021. (Video, Slides, Open-source library)
A. Agarwal, M. Amjad, D. Shah, D. Shen. "Model Agnostic Time Series Analysis via Matrix Estimation".
Journal: Proceedings of the ACM on Measurement and Analysis of Computing Systems, 2019.
Conference: ACM Sigmetrics, 2019. (Slides)
Data Markets & Value of Data
A. Abadie, A. Agarwal, G. Imbens, S. Jia, J. McQueen, S. Stepaniants. "Estimating the Value of Evidence-Based Decision Making".
A. Agarwal, M. Dahleh, T. Horel, M. Rui. "Towards Data Auctions with Externalities".
A. Agarwal, M. Dahleh, D. Shah, D. Sleeper, A. Tsai, M. Wong. "Zorro: A Model Agnostic System to Price Consumer Data."
A. Agarwal, M. Dahleh, T. Sarkar. "A Marketplace for Data: An Algorithmic Solution".
Selected Talks
CARE Reading Group '24
Synthetic Combinations:
A Causal Inference Framework for Combinational Interventions
Stanford Online Causal Inference Seminar '24
Double Robustness Inference in Causal Latent Factor Models
(Talk given by Raaz Dwivedi)
Stanford Online Causal Inference Seminar '23
On Causal Inference with Temporal and Spatial Spillovers with Panel Data
Stanford Online Causal Inference Seminar '22
Causal Inference with Corrupted Data (Talk given by Rahul Singh)
Simons Institute-UC Berkeley '22
(also at Stanford Online Causal Seminar'21, Chamberlain Seminar '21 )
Synthetic Interventions
Simons Institute-UC Berkeley '22
Causal Matrix Completion: Application to Offline Causal RL
Sigmetrics '22
On Multivariate Singular Spectrum Analysis and its Variants
NeurIPS Demo '21
tspDB: Time Series Predict Database
NeurIPS Oral Presentation '19
On Principal Component Regression
ACM Economics and Computation '19
A Marketplace for Data: An Algorithmic Solution