arXiv subjects

Preetam Nandy

Publications and source records attributed to Preetam Nandy.

17 recordsLinked to original sources

arXiv2023

Generalized Causal Tree for Uplift Modeling

Uplift modeling is crucial in various applications ranging from marketing and policy-making to personalized recommendations. The main objective is to learn optimal treatment allocations for a heterogeneous population. A primary line of existing work modifies the loss function of the decision tree algorithm to identify cohorts with heterogeneous treatment effects. Another line of work estimates the individual treatment effects separately for the treatment group and the control group using off-the-shelf supervised learning algorithms. The former approach that directly models the heterogeneous treatment effect is known to outperform the latter in practice. However, the existing tree-based methods are mostly limited to a single treatment and a single control use case, except for a handful of extensions to multiple discrete treatments. In this paper, we propose a generalization of tree-based approaches to tackle multiple discrete and continuous-valued treatments. We focus on a generalization of the well-known causal tree algorithm due to its desirable statistical properties, but our generalization technique can be applied to other tree-based approaches as well. The efficacy of our proposed method is demonstrated using experiments and real data examples.

stat.ME↗

arXiv2023

Detection and Mitigation of Algorithmic Bias via Predictive Rate Parity

Predictive parity (PP), also known as sufficiency, is a core definition of algorithmic fairness essentially stating that model outputs must have the same interpretation of expected outcomes regardless of group. Testing and satisfying PP is especially important in many settings where model scores are interpreted by humans or directly provide access to opportunity, such as healthcare or banking. Solutions for PP violations have primarily been studied through the lens of model calibration. However, we find that existing calibration-based tests and mitigation methods are designed for independent data, which is often not assumable in large-scale applications such as social media or medical testing. In this work, we address this issue by developing a statistically rigorous non-parametric regression based test for PP with dependent observations. We then apply our test to illustrate that PP testing can significantly vary under the two assumptions. Lastly, we provide a mitigation solution to provide a minimally-biased post-processing transformation function to achieve PP.

stat.ME↗

arXiv2022

Long-term Dynamics of Fairness Intervention in Connection Recommender Systems

Recommender system fairness has been studied from the perspectives of a variety of stakeholders including content producers, the content itself and recipients of recommendations. Regardless of which type of stakeholders are considered, most works in this area assess the efficacy of fairness intervention by evaluating a single fixed fairness criterion through the lens of a one-shot, static setting. Yet recommender systems constitute dynamical systems with feedback loops from the recommendations to the underlying population distributions which could lead to unforeseen and adverse consequences if not taken into account. In this paper, we study a connection recommender system patterned after the systems employed by web-scale social networks and analyze the long-term effects of intervening on fairness in the recommendations. We find that, although seemingly fair in aggregate, common exposure and utility parity interventions fail to mitigate amplification of biases in the long term. We theoretically characterize how certain fairness interventions impact the bias amplification dynamics in a stylized Pólya urn model.

cs.CY↗

arXiv2022

Pushing the limits of fairness impossibility: Who's the fairest of them all?

The impossibility theorem of fairness is a foundational result in the algorithmic fairness literature. It states that outside of special cases, one cannot exactly and simultaneously satisfy all three common and intuitive definitions of fairness - demographic parity, equalized odds, and predictive rate parity. This result has driven most works to focus on solutions for one or two of the metrics. Rather than follow suit, in this paper we present a framework that pushes the limits of the impossibility theorem in order to satisfy all three metrics to the best extent possible. We develop an integer-programming based approach that can yield a certifiably optimal post-processing method for simultaneously satisfying multiple fairness criteria under small violations. We show experiments demonstrating that our post-processor can improve fairness across the different definitions simultaneously with minimal model performance reduction. We also discuss applications of our framework for model selection and fairness explainability, thereby attempting to answer the question: who's the fairest of them all?

cs.CY↗

arXiv2022

Achieving Fairness via Post-Processing in Web-Scale Recommender Systems

Building fair recommender systems is a challenging and crucial area of study due to its immense impact on society. We extended the definitions of two commonly accepted notions of fairness to recommender systems, namely equality of opportunity and equalized odds. These fairness measures ensure that equally "qualified" (or "unqualified") candidates are treated equally regardless of their protected attribute status (such as gender or race). We propose scalable methods for achieving equality of opportunity and equalized odds in rankings in the presence of position bias, which commonly plagues data generated from recommender systems. Our algorithms are model agnostic in the sense that they depend only on the final scores provided by a model, making them easily applicable to virtually all web-scale recommender systems. We conduct extensive simulations as well as real-world experiments to show the efficacy of our approach.

stat.ML↗

arXiv2022

Offline Reinforcement Learning for Mobile Notifications

Mobile notification systems have taken a major role in driving and maintaining user engagement for online platforms. They are interesting recommender systems to machine learning practitioners with more sequential and long-term feedback considerations. Most machine learning applications in notification systems are built around response-prediction models, trying to attribute both short-term impact and long-term impact to a notification decision. However, a user's experience depends on a sequence of notifications and attributing impact to a single notification is not always accurate, if not impossible. In this paper, we argue that reinforcement learning is a better framework for notification systems in terms of performance and iteration speed. We propose an offline reinforcement learning framework to optimize sequential notification decisions for driving user engagement. We describe a state-marginalized importance sampling policy evaluation approach, which can be used to evaluate the policy offline and tune learning hyperparameters. Through simulations that approximate the notifications ecosystem, we demonstrate the performance and benefits of the offline evaluation approach as a part of the reinforcement learning modeling approach. Finally, we collect data through online exploration in the production system, train an offline Double Deep Q-Network and launch a successful policy online. We also discuss the practical considerations and results obtained by deploying these policies for a large-scale recommendation system use-case.

cs.LG↗

arXiv2021

A/B Testing for Recommender Systems in a Two-sided Marketplace

Two-sided marketplaces are standard business models of many online platforms (e.g., Amazon, Facebook, LinkedIn), wherein the platforms have consumers, buyers or content viewers on one side and producers, sellers or content-creators on the other. Consumer side measurement of the impact of a treatment variant can be done via simple online A/B testing. Producer side measurement is more challenging because the producer experience depends on the treatment assignment of the consumers. Existing approaches for producer side measurement are either based on graph cluster-based randomization or on certain treatment propagation assumptions. The former approach results in low-powered experiments as the producer-consumer network density increases and the latter approach lacks a strict notion of error control. In this paper, we propose (i) a quantification of the quality of a producer side experiment design, and (ii) a new experiment design mechanism that generates high-quality experiments based on this quantification. Our approach, called UniCoRn (Unifying Counterfactual Rankings), provides explicit control over the quality of the experiment and its computation cost. Further, we prove that our experiment design is optimal to the proposed design quality measure. Our approach is agnostic to the density of the producer-consumer network and does not rely on any treatment propagation assumption. Moreover, unlike the existing approaches, we do not need to know the underlying network in advance, making this widely applicable to the industrial setting where the underlying network is unknown and challenging to predict a priori due to its dynamic nature. We use simulations to validate our approach and compare it against existing methods. We also deployed UniCoRn in an edge recommendation application that serves tens of millions of members and billions of edge recommendations daily.

cs.SI↗

arXiv2021

Inference for Individual Mediation Effects and Interventional Effects in Sparse High-Dimensional Causal Graphical Models

We consider the problem of identifying intermediate variables (or mediators) that regulate the effect of a treatment on a response variable. While there has been significant research on this classical topic, little work has been done when the set of potential mediators is high-dimensional (HD). A further complication arises when these mediators are interrelated (with unknown dependencies). In particular, we assume that the causal structure of the treatment, the confounders, the potential mediators and the response is a (possibly unknown) directed acyclic graph (DAG). HD DAG models have previously been used for the estimation of causal effects from observational data. In particular, methods called IDA and joint-IDA have been developed for estimating the effects of single and multiple simultaneous interventions, respectively. In this paper, we propose an IDA-type method called MIDA for estimating so-called individual mediation effects from HD observational data. Although IDA and joint-IDA estimators have been shown to be consistent in certain sparse HD settings, their asymptotic properties such as convergence in distribution and inferential tools in such settings have remained unknown. In this paper, we prove HD consistency of MIDA for linear structural equation models with sub-Gaussian errors. More importantly, we derive distributional convergence results for MIDA in similar HD settings, which are applicable to IDA and joint-IDA estimators as well. To our knowledge, these are the first such distributional convergence results facilitating inference for IDA-type estimators. These are built on our novel theoretical results regarding uniform bounds for linear regression estimators over varying subsets of HD covariates which may be of independent interest. Finally, we empirically validate our asymptotic theory for MIDA and demonstrate its usefulness via simulations and a real data application.

stat.ME↗

arXiv2020

Personalized Treatment Selection using Causal Heterogeneity

Randomized experimentation (also known as A/B testing or bucket testing) is widely used in the internet industry to measure the metric impact obtained by different treatment variants. A/B tests identify the treatment variant showing the best performance, which then becomes the chosen or selected treatment for the entire population. However, the effect of a given treatment can differ across experimental units and a personalized approach for treatment selection can greatly improve upon the usual global selection strategy. In this work, we develop a framework for personalization through (i) estimation of heterogeneous treatment effect at either a cohort or member-level, followed by (ii) selection of optimal treatment variants for cohorts (or members) obtained through (deterministic or stochastic) constrained optimization. We perform a two-fold evaluation of our proposed methods. First, a simulation analysis is conducted to study the effect of personalized treatment selection under carefully controlled settings. This simulation illustrates the differences between the proposed methods and the suitability of each with increasing uncertainty. We also demonstrate the effectiveness of the method through a real-life example related to serving notifications at Linkedin. The solution significantly outperformed both heuristic solutions and the global treatment selection baseline leading to a sizable win on top-line metrics like member visits.

stat.ME↗

arXiv2020

A/B Testing in Dense Large-Scale Networks: Design and Inference

Design of experiments and estimation of treatment effects in large-scale networks, in the presence of strong interference, is a challenging and important problem. Most existing methods' performance deteriorates as the density of the network increases. In this paper, we present a novel strategy for accurately estimating the causal effects of a class of treatments in a dense large-scale network. First, we design an approximate randomized controlled experiment by solving an optimization problem to allocate treatments in the presence of competition among neighboring nodes. Then we apply an importance sampling adjustment to correct for any leftover bias (from the approximation) in estimating average treatment effects. We provide theoretical guarantees, verify robustness in a simulation study, and validate the scalability and usefulness of our procedure in a real-world experiment on a large social network.

stat.ME↗

arXiv2019

Optimal Convergence for Stochastic Optimization with Multiple Expectation Constraints

In this paper, we focus on the problem of stochastic optimization where the objective function can be written as an expectation function over a closed convex set. We also consider multiple expectation constraints which restrict the domain of the problem. We extend the cooperative stochastic approximation algorithm from Lan and Zhou [2016] to solve the particular problem. We close the gaps in the previous analysis and provide a novel proof technique to show that our algorithm attains the optimal rate of convergence for both optimality gap and constraint violation when the functions are generally convex. We also compare our algorithm empirically to the state-of-the-art and show improved convergence in many situations.

math.ST↗

arXiv2018

Robust causal structure learning with some hidden variables

We introduce a new method to estimate the Markov equivalence class of a directed acyclic graph (DAG) in the presence of hidden variables, in settings where the underlying DAG among the observed variables is sparse, and there are a few hidden variables that have a direct effect on many of the observed ones. Building on the so-called low rank plus sparse framework, we suggest a two-stage approach which first removes the effect of the hidden variables, and then estimates the Markov equivalence class of the underlying DAG under the assumption that there are no remaining hidden variables. This approach is consistent in certain high-dimensional regimes and performs favourably when compared to the state of the art, both in terms of graphical structure recovery and total causal effect estimation.

stat.ME↗

arXiv2018

High-dimensional consistency in score-based and hybrid structure learning

Main approaches for learning Bayesian networks can be classified as constraint-based, score-based or hybrid methods. Although high-dimensional consistency results are available for constraint-based methods like the PC algorithm, such results have not been proved for score-based or hybrid methods, and most of the hybrid methods have not even shown to be consistent in the classical setting where the number of variables remains fixed and the sample size tends to infinity. In this paper, we show that consistency of hybrid methods based on greedy equivalence search (GES) can be achieved in the classical setting with adaptive restrictions on the search space that depend on the current state of the algorithm. Moreover, we prove consistency of GES and adaptively restricted GES (ARGES) in several sparse high-dimensional settings. ARGES scales well to sparse graphs with thousands of variables and our simulation study indicates that both GES and ARGES generally outperform the PC algorithm.

math.ST↗

arXiv2017

Structure Learning of Linear Gaussian Structural Equation Models with Weak Edges

We consider structure learning of linear Gaussian structural equation models with weak edges. Since the presence of weak edges can lead to a loss of edge orientations in the true underlying CPDAG, we define a new graphical object that can contain more edge orientations. We show that this object can be recovered from observational data under a type of strong faithfulness assumption. We present a new algorithm for this purpose, called aggregated greedy equivalence search (AGES), that aggregates the solution path of the greedy equivalence search (GES) algorithm for varying values of the penalty parameter. We prove consistency of AGES and demonstrate its performance in a simulation study and on single cell data from Sachs et al. (2005). The algorithm will be made available in the R-package pcalg.

stat.ME↗

arXiv2016

Estimating the effect of joint interventions from observational data in sparse high-dimensional settings

We consider the estimation of joint causal effects from observational data. In particular, we propose new methods to estimate the effect of multiple simultaneous interventions (e.g., multiple gene knockouts), under the assumption that the observational data come from an unknown linear structural equation model with independent errors. We derive asymptotic variances of our estimators when the underlying causal structure is partly known, as well as high-dimensional consistency when the causal structure is fully unknown and the joint distribution is multivariate Gaussian. We also propose a generalization of our methodology to the class of nonparanormal distributions. We evaluate the estimators in simulation studies and also illustrate them on data from the DREAM4 challenge.

stat.ME↗

arXiv2016

Large-Sample Theory for the Bergsma-Dassios Sign Covariance

The Bergsma-Dassios sign covariance is a recently proposed extension of Kendall's tau. In contrast to tau or also Spearman's rho, the new sign covariance $τ^*$ vanishes if and only if the two considered random variables are independent. Specifically, this result has been shown for continuous as well as discrete variables. We develop large-sample distribution theory for the empirical version of $τ^*$. In particular, we use theory for degenerate U-statistics to derive asymptotic null distributions under independence and demonstrate in simulations that the limiting distributions give useful approximations.

math.ST↗

arXiv2015

A review of some recent advances in causal inference

We give a selective review of some recent developments in causal inference, intended for researchers who are not familiar with graphical models and causality, and with a focus on methods that are applicable to large data sets. We mainly address the problem of estimating causal effects from observational data. For example, one can think of estimating the effect of single or multiple gene knockouts from wild-type gene expression data, that is, from gene expression measurements that were obtained without doing any gene knockout experiments. We assume that the observational data are generated from a causal structure that can be represented by a directed acyclic graph (DAG). First, we discuss estimation of causal effects when the underlying causal DAG is known. In large-scale networks, however, the causal DAG is often unknown. Next, we therefore discuss causal structure learning, that is, learning information about the causal structure from observational data. We then combine these two parts and discuss methods to estimate (bounds on) causal effects from observational data when the causal structure is unknown. We also illustrate this method on a yeast gene expression data set. We close by mentioning several extensions of the discussed work.

stat.ME↗