Anish Agarwal
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Anish Agarwal has a total of 20 co-authors. Here are the top co-authors by number of shared publications:
- Chara Podimata
Massachusetts Institute of Technology
2 shared publications - Devavrat Shah
Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139
2 shared publications - Keegan Harris
Carnegie Mellon University, Pittsburgh, USA
2 shared publications - Zhiwei Steven Wu
Carnegie Mellon University
2 shared publications - Abdullah Alomar
Department of Electrical Engineering and Computer Science, MIT, Cambridge, Massachusetts 02139
1 shared publication - Bowen Zheng
Department of Electrical and Computer Engineering, University of Massachusetts at Lowell, Lowell, MA, USA
1 shared publication
Latest Publications
- 1 November 2026
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Operations Research
- Vol. 74 (6) , 3019-3037
On Multivariate Singular Spectrum Analysis: Tensor and Matrix Variants
Tensor and Matrix Methods for Multivariate Time Series, with Provable Guarantees Real-world time series data—from financial markets to electricity demand—are often multivariate, noisy, and riddled with missing observations. In “On multivariate singular spectrum analysis: Tensor and matrix variants,” Anish Agarwal, Abdullah Alomar, and Devavrat Shah introduce and rigorously analyze two extensions of singular spectrum analysis (SSA) for this challenging setting. Their matrix variant (mSSA) achieves imputation and forecasting error scaling as [Formula: see text] improving over both univariate SSA and standard matrix estimation methods that ignore temporal structure. A novel tensor variant (tSSA) further improves sample complexity in certain regimes, strengthening the link between time series analysis and tensor estimation. These results are made possible by a spatio-temporal factor model that accommodates a rich class of dynamics—harmonics, polynomials, and smooth periodic functions—which the authors extend further by introducing a “Hankel calculus” establishing closure under component-wise addition and multiplication. Empirically, mSSA matches state-of-the-art deep learning methods while significantly outperforming classical approaches such as vector autoregression.
- 1 November 2026
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Operations Research
- Vol. 74 (6) , 3019-3037
On Multivariate Singular Spectrum Analysis: Tensor and Matrix Variants
Tensor and Matrix Methods for Multivariate Time Series, with Provable Guarantees Real-world time series data—from financial markets to electricity demand—are often multivariate, noisy, and riddled with missing observations. In “On multivariate singular spectrum analysis: Tensor and matrix variants,” Anish Agarwal, Abdullah Alomar, and Devavrat Shah introduce and rigorously analyze two extensions of singular spectrum analysis (SSA) for this challenging setting. Their matrix variant (mSSA) achieves imputation and forecasting error scaling as [Formula: see text] improving over both univariate SSA and standard matrix estimation methods that ignore temporal structure. A novel tensor variant (tSSA) further improves sample complexity in certain regimes, strengthening the link between time series analysis and tensor estimation. These results are made possible by a spatio-temporal factor model that accommodates a rich class of dynamics—harmonics, polynomials, and smooth periodic functions—which the authors extend further by introducing a “Hankel calculus” establishing closure under component-wise addition and multiplication. Empirically, mSSA matches state-of-the-art deep learning methods while significantly outperforming classical approaches such as vector autoregression.
- Top Cited Papers
- 1 March 2026
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Operations Research
- Vol. 74 (2) , 840-859
Synthetic Interventions: Extending Synthetic Controls to Multiple Treatments
Beyond One Policy: A New Framework for Comparing Multiple Interventions Traditional methods for policy evaluation typically focus on a single intervention. Yet many real-world settings feature multiple, often concurrent, interventions—making it crucial to understand their comparative effects. In “Synthetic Interventions: Extending Synthetic Controls to Multiple Treatments,” Agarwal, Shah, and Shen introduce the synthetic interventions framework, a generalization of the synthetic controls method that accommodates multiple treatments within a unified model. By representing outcomes as a low-rank tensor capturing relationships across time, units, and interventions, their approach enables researchers to estimate counterfactual outcomes under all interventions for each unit. The authors establish consistency of their estimator and show that a bias-corrected version achieves asymptotic normality, permitting valid statistical inference. This work offers a new perspective for evaluating complex, multipolicy environments where traditional causal inference tools fall short.
- 1 March 2026
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Operations Research
- Vol. 74 (2) , 840-859
Top Cited PapersSynthetic Interventions: Extending Synthetic Controls to Multiple Treatments
Beyond One Policy: A New Framework for Comparing Multiple Interventions Traditional methods for policy evaluation typically focus on a single intervention. Yet many real-world settings feature multiple, often concurrent, interventions—making it crucial to understand their comparative effects. In “Synthetic Interventions: Extending Synthetic Controls to Multiple Treatments,” Agarwal, Shah, and Shen introduce the synthetic interventions framework, a generalization of the synthetic controls method that accommodates multiple treatments within a unified model. By representing outcomes as a low-rank tensor capturing relationships across time, units, and interventions, their approach enables researchers to estimate counterfactual outcomes under all interventions for each unit. The authors establish consistency of their estimator and show that a bias-corrected version achieves asymptotic normality, permitting valid statistical inference. This work offers a new perspective for evaluating complex, multipolicy environments where traditional causal inference tools fall short.
- Open Access
- 11 June 2024
- Published by Association for Computing Machinery (ACM) in ACM SIGMETRICS Performance Evaluation Review
- Vol. 52 (1) , 69-70
Strategyproof Decision-Making in Panel Data Settings and Beyond
We consider the problem of decision-making using panel data, in which a decision-maker gets noisy, repeated measurements of multiple units (or agents). We consider the setup used in synthetic control methods, where there is a pre-intervention period when the principal observes the outcomes of each unit, after which the principal uses these observations to assign a treatment to each unit. Unlike this classical setting, we permit the units generating the panel data to be strategic, i.e. units may modify their pre-intervention outcomes in order to receive a more desirable intervention. The principal's goal is to design a strategyproof intervention policy, i.e. a policy that assigns units their utility-maximizing interventions despite their potential strategizing. We first identify a necessary and sufficient condition under which a strategyproof intervention policy exists, and provide a strategyproof mechanism with a simple closed form when one does exist. Along the way, we prove impossibility results for strategic multiclass classification, a natural extension of the well-studied (binary) strategic classification problem to the multiclass setting, which may be of independent interest. When there are two interventions, we establish that there always exists a strategyproof mechanism, and provide an algorithm for learning such a mechanism. For three or more interventions, we provide an algorithm for learning a strategyproof mechanism if there exists a sufficiently large gap in the principal's rewards between different interventions. Finally, we empirically evaluate our model using real-world panel data collected from product sales over 18 months. We find that our methods compare favorably to baselines which do not take strategic interactions into consideration, even in the presence of model misspecification.
- 11 June 2024
- Published by Association for Computing Machinery (ACM) in ACM SIGMETRICS Performance Evaluation Review
- Vol. 52 (1) , 69-70
Open AccessStrategyproof Decision-Making in Panel Data Settings and Beyond
We consider the problem of decision-making using panel data, in which a decision-maker gets noisy, repeated measurements of multiple units (or agents). We consider the setup used in synthetic control methods, where there is a pre-intervention period when the principal observes the outcomes of each unit, after which the principal uses these observations to assign a treatment to each unit. Unlike this classical setting, we permit the units generating the panel data to be strategic, i.e. units may modify their pre-intervention outcomes in order to receive a more desirable intervention. The principal's goal is to design a strategyproof intervention policy, i.e. a policy that assigns units their utility-maximizing interventions despite their potential strategizing. We first identify a necessary and sufficient condition under which a strategyproof intervention policy exists, and provide a strategyproof mechanism with a simple closed form when one does exist. Along the way, we prove impossibility results for strategic multiclass classification, a natural extension of the well-studied (binary) strategic classification problem to the multiclass setting, which may be of independent interest. When there are two interventions, we establish that there always exists a strategyproof mechanism, and provide an algorithm for learning such a mechanism. For three or more interventions, we provide an algorithm for learning a strategyproof mechanism if there exists a sufficiently large gap in the principal's rewards between different interventions. Finally, we empirically evaluate our model using real-world panel data collected from product sales over 18 months. We find that our methods compare favorably to baselines which do not take strategic interactions into consideration, even in the presence of model misspecification.
- Open Access
- 10 June 2024
- Published by Association for Computing Machinery (ACM)
- p. 69-70
Strategyproof Decision-Making in Panel Data Settings and Beyond
We consider the problem of decision-making using panel data, in which a decision-maker gets noisy, repeated measurements of multiple units (or agents). We consider the setup used in synthetic control methods, where there is a pre-intervention period when the principal observes the outcomes of each unit, after which the principal uses these observations to assign a treatment to each unit. Unlike this classical setting, we permit the units generating the panel data to be strategic, i.e. units may modify their pre-intervention outcomes in order to receive a more desirable intervention. The principal's goal is to design a strategyproof intervention policy, i.e. a policy that assigns units their utility-maximizing interventions despite their potential strategizing. We first identify a necessary and sufficient condition under which a strategyproof intervention policy exists, and provide a strategyproof mechanism with a simple closed form when one does exist. Along the way, we prove impossibility results for strategic multiclass classification, a natural extension of the well-studied (binary) strategic classification problem to the multiclass setting, which may be of independent interest. When there are two interventions, we establish that there always exists a strategyproof mechanism, and provide an algorithm for learning such a mechanism. For three or more interventions, we provide an algorithm for learning a strategyproof mechanism if there exists a sufficiently large gap in the principal's rewards between different interventions. Finally, we empirically evaluate our model using real-world panel data collected from product sales over 18 months. We find that our methods compare favorably to baselines which do not take strategic interactions into consideration, even in the presence of model misspecification.
- 10 June 2024
- Published by Association for Computing Machinery (ACM)
- p. 69-70
Open AccessStrategyproof Decision-Making in Panel Data Settings and Beyond
We consider the problem of decision-making using panel data, in which a decision-maker gets noisy, repeated measurements of multiple units (or agents). We consider the setup used in synthetic control methods, where there is a pre-intervention period when the principal observes the outcomes of each unit, after which the principal uses these observations to assign a treatment to each unit. Unlike this classical setting, we permit the units generating the panel data to be strategic, i.e. units may modify their pre-intervention outcomes in order to receive a more desirable intervention. The principal's goal is to design a strategyproof intervention policy, i.e. a policy that assigns units their utility-maximizing interventions despite their potential strategizing. We first identify a necessary and sufficient condition under which a strategyproof intervention policy exists, and provide a strategyproof mechanism with a simple closed form when one does exist. Along the way, we prove impossibility results for strategic multiclass classification, a natural extension of the well-studied (binary) strategic classification problem to the multiclass setting, which may be of independent interest. When there are two interventions, we establish that there always exists a strategyproof mechanism, and provide an algorithm for learning such a mechanism. For three or more interventions, we provide an algorithm for learning a strategyproof mechanism if there exists a sufficiently large gap in the principal's rewards between different interventions. Finally, we empirically evaluate our model using real-world panel data collected from product sales over 18 months. We find that our methods compare favorably to baselines which do not take strategic interactions into consideration, even in the presence of model misspecification.
- 21 November 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 2583-2586
A Deep Learning Approach to Predict the Mutual Coupling Effect in Metasurfaces
In recent years, metasurfaces have become a novel and promising substitute to their counterparts — bulky 3D geometrical optics. Meta-surfaces are usually enabled by arrays of meta-atoms that act as scattering particles, which provide independent phase and amplitude control at the subwavelength level. Conventional meta-atom design approach involves the parametric sweep and careful selection of meta-atoms with desired performance. During this process, a periodic boundary condition assumption is usually applied to each meta-atom in numerical simulations, which assumes all meta-atoms under consideration are surrounded by identical neighbors in an infinite-large array. However, in practical metasurface devices such as lenses, beam deflectors and holograms, meta-atoms are usually surrounded by different-shaped neighbors, which leads to different electromagnetic (EM) performances due to the near field coupling effect. This deviation causes varying degrees of performance degradation in metasurface devices, depending on the strength of the mutual coupling. In this manuscript, we use a deep learning approach to predict a meta-atom’s actual EM performance when it’s placed in a large array surrounded by different neighbors, with the mutual coupling effects taken into account. Specifically, we constructed a deep convolutional neural network that takes the physical specifications of the target meta-atom and all its neighbors as input, then predicts the actual phase and amplitude responses of the target meta-atom with mutual coupling taken into consideration. After trained with ample amount of data, this neural network is able to generate accurate phase and amplitude in millisecond timescale. This approach not only explains the Meta-surfaces’ performance deterioration caused by mutual coupling, but also can be used for the inverse optimization of various meta-devices. We demonstrate the efficacy of this methodology by inversely optimizing the efficiency of different metasurface devices including beam deflectors and lenses.
- 21 November 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 2583-2586
A Deep Learning Approach to Predict the Mutual Coupling Effect in Metasurfaces
In recent years, metasurfaces have become a novel and promising substitute to their counterparts — bulky 3D geometrical optics. Meta-surfaces are usually enabled by arrays of meta-atoms that act as scattering particles, which provide independent phase and amplitude control at the subwavelength level. Conventional meta-atom design approach involves the parametric sweep and careful selection of meta-atoms with desired performance. During this process, a periodic boundary condition assumption is usually applied to each meta-atom in numerical simulations, which assumes all meta-atoms under consideration are surrounded by identical neighbors in an infinite-large array. However, in practical metasurface devices such as lenses, beam deflectors and holograms, meta-atoms are usually surrounded by different-shaped neighbors, which leads to different electromagnetic (EM) performances due to the near field coupling effect. This deviation causes varying degrees of performance degradation in metasurface devices, depending on the strength of the mutual coupling. In this manuscript, we use a deep learning approach to predict a meta-atom’s actual EM performance when it’s placed in a large array surrounded by different neighbors, with the mutual coupling effects taken into account. Specifically, we constructed a deep convolutional neural network that takes the physical specifications of the target meta-atom and all its neighbors as input, then predicts the actual phase and amplitude responses of the target meta-atom with mutual coupling taken into consideration. After trained with ample amount of data, this neural network is able to generate accurate phase and amplitude in millisecond timescale. This approach not only explains the Meta-surfaces’ performance deterioration caused by mutual coupling, but also can be used for the inverse optimization of various meta-devices. We demonstrate the efficacy of this methodology by inversely optimizing the efficiency of different metasurface devices including beam deflectors and lenses.