Purva Pruthi

I am a fifth year MS/Ph.D. student at UMass Amherst, advised by David Jensen. My work lies at the intersection of causal inference, computational systems and machine learning. Specifically, I work on developing methods to learn causal models of the computational systems such as databases, software programs to enable automated causal reasoning in the complex real-world systems.

Many real-world systems have modular structure and can be represented as sets of interacting components. Examples of such systems include computational systems such as query processors, natural systems such as cells, and social systems such as families. Leveraging the modular structure of the world allows us to build data-driven models of the world that are re-usable, data efficient, and easily maintainable. Complex causal mechanisms governing these systems can also be decomposed into simpler causal mechanisms. My work focuses on understanding the benefits of explicitly modeling the compositional structure of the causal mechanisms in models and understanding the systematic generalization characteristics of compositional causal models.

I’ve also spent eight months (Sept 2021 - May 2022) as AI resident at Google X, The Moonshot Factory where I worked on applying causal inference techniques in solving the most challenging real-world problems. I also spent Summer 2018 at Amazon Cambridge, United Kingdom, working with Javier Gonzalez and Xiaoyu Lu on designing a transfer-learning framework for reinforcement learning focusing on environments with visually changing features but same underlying causal dynamics. Previously, I also worked with Madalina Fiterau at the intersection of causal reasoning and healthcare.

Prior to joining UMass as a graduate student, I worked at Goldman Sachs for three years. I worked on building statistical models for systematic investment strategies. I also worked on building statistical tools to analyze the root causes of undesirable behavior in trade settlement procedures.

I graduated from Indian Institute of Technology, Roorkee in 2015. During my undergraduate program, I worked with Durga Toshniwal on spatio-temporal diffusion analysis of online information on Twitter. I also enjoyed doing competitive programming as part of Programming and Algorithms group in college.

Outside work, I enjoy reading books, writing my thoughts about science and everything else (prose-style poetry). I love cooking, playing piano and taking long walks looking at the sunset, stars, (and sometimes the moon!)

news

Feb 1, 2024	I started as part-time applied AI consultant at Alpha Studio. I am helping early-stage startups building innovative AI products.
Feb 1, 2024	I am awarded the CICS Dissertation Writing Fellowship for Spring 2024.
May 30, 2023	This summer, I am working as Data Science for Common Good (DS4CG) Fellow focusing on analysing energy consumption in Commonwealth buildings of Massachusetts.
May 13, 2022	Completed AI residency at Google X, The Moonshot Factory. Returned back to the PhD program.
Sep 20, 2021	Started AI residency at Google X, The Moonshot Factory.
May 30, 2021	I passed Ph.D. Candidacy at UMass Amherst.
May 7, 2021	Our paper titled “How and Why to Use Experimental Data to Evaluate Methods for Observational Causal Inference” has been accepted for presentation at the International Conference on Machine Learning 2021. [Paper]
Jun 20, 2020	Our paper titled “Structure Mapping for Transferability of Causal Models” has been accepted for presentation at the Inductive Biases, Invariances and Generalization in Reinforcement Learning Workshop, ICML 2020. [Talk]
May 20, 2019	Started summer internship at Amazon, Cambridge, UK working with causal inference team.

selected publications

How and Why to Use Experimental Data to Evaluate Methods for Observational Causal Inference

Amanda M Gentzel, Purva Pruthi, and David Jensen

2021

Abs PDF

Methods that infer causal dependence from observational data are central to many areas of science, including medicine, economics, and the social sciences. A variety of theoretical properties of these methods have been proven, but empirical evaluation remains a challenge, largely due to the lack of observational data sets for which treatment effect is known. We describe and analyze observational sampling from randomized controlled trials (OSRCT), a method for evaluating causal inference methods using data from randomized controlled trials (RCTs). This method can be used to create constructed observational data sets with corresponding unbiased estimates of treatment effect, substantially increasing the number of data sets available for evaluating causal inference methods. We show that, in expectation, OSRCT creates data sets that are equivalent to those produced by randomly sampling from empirical data sets in which all potential outcomes are available. We then perform a large-scale evaluation of seven causal inference methods over 37 data sets, drawn from RCTs, as well as simulators, real-world computational systems, and observational data sets augmented with a synthetic response variable. We find notable performance differences when comparing across data from different sources, demonstrating the importance of using data from a variety of sources when evaluating any causal inference method.
Structure Mapping for Transferability of Causal Models

Purva Pruthi, Javier González, Xiaoyu Lu, and Madalina Fiterau

2020

Abs PDF

Human beings learn causal models and constantly use them to transfer knowledge between similar environments. We use this intuition to design a transfer-learning framework using object-oriented representations to learn the causal relationships between objects. A learned causal dynamics model can be used to transfer between variants of an environment with exchangeable perceptual features among objects but with the same underlying causal dynamics. We adapt continuous optimization for structure learning techniques to explicitly learn the cause and effects of the actions in an interactive environment and transfer to the target domain by categorization of the objects based on causal knowledge. We demonstrate the advantages of our approach in a gridworld setting by combining causal model-based approach with model-free approach in reinforcement learning.