Publications

The tension between rational and irrational behaviors in human decision-making has been acknowledged across a wide range of disciplines, from philosophy to psychology, neuroscience to behavioral economics. Models of multi-agent interactions, such as von Neumann and Morgenstern’s expected utility theory and Nash’s game theory, provide rigorous mathematical frameworks for how agents should behave when rationality is sought. However, the rationality assumption has been extensively challenged, as human decision-making is often irrational, influenced by biases, emotions, and uncertainty, which may even have a positive effect in certain cases. Behavioral economics, for example, attempts to explain such irrational behaviors, including Kahneman’s dual-process theory and Thaler’s nudging concept, and accounts for deviations from rationality. In this paper, we analyze this tension through a causal lens and develop a framework that accounts for rational and irrational decision-making, which we term Causal Game Theory. We then introduce a novel notion called counterfactual rationality, which allows agents to make choices leveraging their irrational tendencies. We extend the notion of Nash Equilibrium to counterfactual actions and Pearl Causal Hierarchy (PCH), and show that strategies following counterfactual rationality dominate strategies based on standard game theory. We further develop an algorithm to learn such strategies when not all information about other agents is available.

Estimating causal effects from observational data remains a fundamental challenge in causal inference, especially in the presence of latent confounders. This paper focuses on estimating causal effects in Gaussian Linear Structural Causal Models (GL-SCMs), which are widely used due to their analytical tractability. However, parameter estimation in GL-SCMs is often infeasible with finite data, primarily due to overparameterization. To address this, we introduce the class of Centralized Gaussian Linear SCMs (CGL-SCMs), a simplified yet expressive subclass where exogenous variables follow standardized distributions. We show that CGL-SCMs are equally expressive in terms of causal effect identifiability from observational distributions and present a novel EM-based estimation algorithm that can learn CGL-SCM parameters and estimate identifiable causal effects from finite observational samples. Our theoretical analysis is validated through experiments on synthetic data and benchmark causal graphs, demonstrating that the learned models accurately recover causal distributions.

Reasoning with counterfactuals is one of the hallmarks of human cognition, involved in various tasks such as explanation, credit assignment, blame, and responsibility. Counterfactual quantities that are not identifiable in the general non-parametric case may be identified under shape constraints on the functional mechanisms, such as monotonicity. One prominent example of such an approach is the celebrated result by Angrist and Imbens on identifying the Local Average Treatment Effect (LATE) in the instrumental variable setting. In this paper, we study the identification problem of more general settings under monotonicity constraints. We begin by proving the monotonicity reduction lemma, which simplifies counterfactual queries using monotonicity assumptions and facilitates the reduction of a larger class of these queries to interventional quantities. We then extend the existing identification results on Probabilities of Causation (PoCs) and LATE to a broader set of queries and graphs. Finally, we develop an algorithm, M-ID, for identifying arbitrary counterfactual queries from combinations of observational and experimental data, which takes as input a causal diagram with monotonicity constraints. We show that M-ID subsumes the previously known identification results in the literature. We demonstrate the applicability of our results using synthetic and real data.

We study a variant of causal contextual bandits where the context is chosen based on an initial intervention chosen by the learner. At the beginning of each round, the learner selects an initial action, depending on which a stochastic context is revealed by the environment. Following this, the learner then selects a final action and receives a reward. Given $T$ rounds of interactions with the environment, the objective of the learner is to learn a policy (of selecting the initial and the final action) with maximum expected reward. In this paper we study the specific situation where every action corresponds to intervening on a node in some known causal graph. We extend prior work from the deterministic context setting to obtain simple regret minimization guarantees. This is achieved through an instance-dependent causal parameter, $\lambda$, which characterizes our upper bound. Furthermore, we prove that our simple regret is essentially tight for a large class of instances. A key feature of our work is that we use convex optimization to address the bandit exploration problem. We also conduct experiments to validate our theoretical results, and release our code at the project GitHub Repository.

Online browsing on firms’ sites generates user behavior logs (or, logs). These logs are mainstays that drive several user modeling tasks. The logs that inform user modeling are the ones that are attributed to each user, termed Attributed Behaviors (AB). But, a lot more logs are anonymous, upwards of 85%. For example, many users do not sign in while browsing. These logs are not attributed to users, termed Unattributed Behaviors (UB), and are not recognized in user modeling. We examine whether and how UB can benefit user modeling. We focus on a common task, that of user segmentation, for which the prior art uses only AB. We demonstrate that information from UBs, although unattributed to any individual, when used along with ABs, enriches performance of machine learning model for user segmentation. We perform predictive segmentation, whereby predicted outcomes for each segment are evaluated against actual outcomes. Multiple evaluations on two datasets, one of which is public, relative to state of the art baseline, show strong performance of our model in predicting outcomes and in reducing user segmentation error.

Users’ behavioral footprints online enable firms to discover behavior-based user segments (or, segments) and deliver segment specific messages to users. Following the discovery of segments, delivery of messages to users through preferred media channels like Facebook and Google can be challenging, as only a portion of users in a behavior segment find match in a medium, and only a fraction of those matched actually see the message (exposure). Even high quality discovery becomes futile when delivery fails. Many sophisticated algorithms exist for discovering behavioral segments; however, these ignore the delivery component. The problem is compounded because (i) the discovery is performed on the behavior data space in firms’ data (e.g., user clicks), while the delivery is predicated on the static data space (e.g., geo, age) as defined by media; and (ii) firms work under budget constraint. We introduce a stochastic optimization based algorithm for delivery optimized discovery of behavioral user segments and offer new metrics to address the joint optimization. We leverage optimization under a budget constraint for delivery combined with a learning-based component for discovery. Extensive experiments on a public dataset from Google and a proprietary dataset show the effectiveness of our approach by simultaneously improving delivery metrics, reducing budget spend and achieving strong predictive performance in discovery.

We study the problem of determining the best atomic intervention in a Causal Bayesian Network (CBN) specified only by its causal graph. We model this as a stochastic multi-armed bandit (MAB) problem with side-information, where interventions on CBN correspond to arms of the bandit instance. First, we propose a simple regret minimization algorithm that takes as input a causal graph with observable and unobservable nodes and in $T$ exploration rounds achieves $O(m(C)/T)$ expected simple regret. Here $m(C)$ is a parameter dependent on the input CBN $C$ and could be much smaller than the number of arms. We also show that this is almost optimal for CBNs whose causal graphs have an n-ary tree structure. Next, we propose a cumulative regret minimization algorithm that takes as input a causal graph with observable nodes and performs better than the optimal MAB algorithms that do not use causal side-information. We experimentally compare both our algorithms with the best known algorithms in the literature.

This paper investigates the dynamics of competition among organizations with unequal expertise. Multi-agent reinforcement learning has been used to simulate and understand the impact of various incentive schemes designed to offset such inequality. We design Touch-Mark, a game based on well-known multi-agent-particle-environment, where two teams (weak, strong) with unequal but changing skill levels compete against each other. For training such a game, we propose a novel controller assisted multi-agent reinforcement learning algorithm, which empowers each agent with an ensemble of policies along with a supervised controller that by selectively partitioning the sample space, triggers intelligent role division among the teammates. Using C-MADDPG as an underlying framework, we propose an incentive scheme for the weak team such that the final rewards of both teams become the same. We find that in spite of the incentive, the final reward of the weak team falls short of the strong team. On inspecting, we realize that an overall incentive scheme for the weak team does not incentivize the weaker agents within that team to learn and improve. To offset this, we now specially incentivize the weaker player to learn and as a result, observe that the weak team beyond an initial phase performs at par with the stronger team. The final goal of the paper has been to formulate a dynamic incentive scheme that continuously balances the reward of the two teams. This is achieved by devising an incentive scheme enriched with an RL agent which takes minimum information from the environment.

We study the problem of separating a mixture of distributions, all of which come from interventions on a known causal bayesian network. Given oracle access to marginals of all distributions resulting from interventions on the network, and estimates of marginals from the mixture distribution, we want to recover the mixing proportions of different mixture components. We show that in the worst case, mixing proportions cannot be identified using marginals only. If exact marginals of the mixture distribution were known, under a simple assumption of excluding a few distributions from the mixture, we show that the mixing proportions become identifiable. Our identifiability proof is constructive and gives an efficient algorithm recovering the mixing proportions exactly. When exact marginals are not available, we design an optimization framework to estimate the mixing proportions. Our problem is motivated from a real-world scenario of an e-commerce business, where multiple interventions occur at a given time, leading to deviations in expected metrics. We conduct experiments on the well known publicly available ALARM network and on a proprietary dataset from a large e-commerce company validating the performance of our method.

Systems and methods for delivery aware audience segmentation and subsequent delivery of content are described. Embodiments are configured to obtain activity data for a user, assign the user to a user segment based on the activity data using a machine learning model, generate a reach prediction for the user segment, select a media channel for communicating with the user based on the user segment and the reach prediction, and provide targeted content to the user via the selected media channel. According to some aspects, the machine learning model is trained based on content reach data.

Systems and methods for content customization are provided. One aspect of the systems and methods includes receiving dynamic characteristics for a plurality of users, wherein the dynamic characteristics include interactions between the plurality of users and a digital content channel; clustering the plurality of users in a plurality of segments based on the dynamic characteristics using a machine learning model; assigning a user to a segment of the plurality of segments based on static characteristics of the user; and providing customized digital content for the user based on the segment.

In implementations of systems for estimating terminal event likelihood, a computing device implements a termination system to receive observed data describing values of a treatment metric and indications of a terminal event. Values of the treatment metric are grouped into groups using a mixture model that represents the treatment metric as a mixture of distributions. Parameters of a distribution are estimated for each of the groups and mixing proportions are also estimated for each of the groups. In response to receiving a user input requesting an estimate of a likelihood of the terminal event for a particular value of the treatment metric, the termination system generates an indication of the estimate of the likelihood of the terminal event for the particular value based on a distribution density at the particular value for each of the groups and a probability of including the particular value in each of the groups.

One aspect of a method for data processing includes identifying target time series data for a target metric and candidate time series data for a plurality of indicators predictive of the target metric; training a machine learning model to predict the target time series data based on the candidate time series data; computing first through third predictivity values based on the machine learning model, wherein the first predictivity value indicates that a source indicator from the plurality of indicators is predictive of the target metric, the second predictivity value indicates that an intermediate indicator from the plurality of indicators is predictive of the target metric, and the third predictivity value indicates that the source indicator is predictive of the intermediate indicator; and displaying a portion of the candidate time series data corresponding to the intermediate indicator and the source indicator based on the first through third predictivity values.

The present disclosure relates to methods, systems, and non-transitory computer-readable media for determining causal contributions of dimension values to anomalous data based on causal effects of such dimension values on the occurrence of other dimension values from interventions performed in a causal graph. For example, the disclosed systems can identify an anomalous dimension value that reflects a threshold change in value between an anomalous time period and a reference time period. The disclosed systems can determine causal effects by traversing a causal network representing dependencies between different dimensions associated with the dimension values. Based on the causal effects, the disclosed systems can determine causal contributions of particular dimension values on the anomalous dimension value. Further, the disclosed systems can generate a causal-contribution ranking of the particular dimension values based on the determined causal contributions.

The present disclosure relates to systems, non-transitory computer-readable media, and methods for incorporating unobserved behaviors when generating user segments or predictions of future user actions. In particular, in one or more embodiments, the disclosed systems utilize a deep learning-based clustering algorithm that segments the behavioral history of users based on a future outcome. Further, the disclosed systems recognize that users may exhibit behaviors that represent two or more segments and allow for targeted marketing to users based on the user’s inclusion in multiple segments.

Systems and methods for data analytics are described. One or more embodiments of the present disclosure receive target time series data and candidate time series data, where the candidate time series data includes data corresponding to each of a plurality of candidate metrics, train a prediction network to predict the target time series data based on the candidate time series data by setting temporal attention weights corresponding to a plurality of rolling time windows and by setting candidate attention weights corresponding to the plurality of candidate metrics, identify a leading indicator metric for the target time series data from the plurality of candidate metrics based on the temporal attention weights and the candidate attention weights, and signal the leading indicator metric for the target time series data.

A treatment effect system estimates treatment effects by trading off between observational samples and interventional samples to maintain within a budget while providing high confidence. The treatment effect system determines whether to perform interventions by comparing the cost of interventional samples with metrics regarding the joint probability distribution of treatments and their parents in a first set of observational samples. If it is determined to not perform interventions, the treatment effect for each treatment is determined using an estimator that uses the first set of observational samples independent of a second set of observational samples. If it is determined to perform interventions, each treatment is identified as a reliable or unreliable treatment. The treatment effect for reliable treatments is estimated using an estimator that uses the first set of observational samples split into two portions. The treatment effect for unreliable treatments is estimated using interventional samples generated from interventions.