In many applications of AI for Social Impact (e.g., when allocating spots in support programs for underserved communities), resources are scarce and an allocation policy is needed to decide who receives a resource. Before being deployed at scale, a rigorous evaluation of an AI-powered allocation policy is vital. In this paper, we introduce the methods necessary to evaluate index-based allocation policies, which allocate a limited number of resources to those who need them the most. Such policies create dependencies between agents, rendering standard statistical tests invalid and ineffective. Addressing the arising practical and technical challenges, we describe an efficient estimator and methods for drawing valid statistical conclusions. Our extensive experiments validate our methodology in practical settings while also showcasing its statistical power. We conclude by proposing and empirically verifying extensions of our methodology that enable us to reevaluate a past randomized control trial conducted with 10000 beneficiaries for a mHealth program for pregnant women. Our new methodology allows us to draw previously invisible conclusions when comparing two different ML allocation policies.

Predict-then-Optimize is a framework for using machine learning to perform decision-making under uncertainty. The central research question it asks is, "How can we use the structure of a decision-making task to tailor ML models for that specific task?" To this end, recent work has proposed learning task-specific loss functions that capture this underlying structure. However, current approaches make restrictive assumptions about the form of these losses and their impact on ML model behavior. These assumptions both lead to approaches with high computational cost, and when they are violated in practice, poor performance. In this paper, we propose solutions to these issues, avoiding the aforementioned assumptions and utilizing the ML model's features to increase the sample efficiency of learning loss functions. We empirically show that our method achieves state-of-the-art results in four domains from the literature, often requiring an order of magnitude fewer samples than comparable methods from past work. Moreover, our approach outperforms the best existing method by nearly 200% when the localness assumption is broken.

Restless multi-arm bandits (RMABs) are a model for sequentially allocating a limited number of resources to agents modeled as Markov Decision Processes. RMABs have applications in cellular networks, anti-poaching, and in particular, healthcare. For such high-stakes use cases, allocations are often required to treat different groups of agents (e.g., defined by sensitive attributes) fairly. In addition to the fairness challenge, agents’ transition probabilities are often unknown and need to be learned in real-world problems. Thus, group fairness in RMABs requires us to simultaneously learn transition probabilities and how much budget we allocate to each group. Overcoming this key challenge ignored by previous work, we develop a decision-focused-learning pipeline to solve equitable RMABs, using a novel budget allocation algorithm to prevent disparity between groups. Our results on both synthetic and real-world large-scale datasets demonstrate that incorporating fair planning into the learning step greatly improves equity with little sacrifice in utility.

This paper studies restless multi-armed bandit (RMAB) problems with unknown arm transition dynamics but with known correlated arm features. The goal is to learn a model to predict transition dynamics given features, where the Whittle index policy solves the RMAB problems using predicted transitions. However, prior works often learn the model by maximizing the predictive accuracy instead of final RMAB solution quality, causing a mismatch between training and evaluation objectives. To address this shortcoming, we propose a novel approach for decision-focused learning in RMAB that directly trains the predictive model to maximize the Whittle index solution quality. We present three key contributions: (i) we establish differentiability of the Whittle index policy to support decision-focused learning; (ii) we significantly improve the scalability of decision-focused learning approaches in sequential problems, specifically RMAB problems; (iii) we apply our algorithm to a previously collected dataset of maternal and child health to demonstrate its performance. Indeed, our algorithm is the first for decision-focused learning in RMAB that scales to real-world problem sizes.

Decision-Focused Learning (DFL) is a paradigm for tailoring a predictive model to a downstream optimization task that uses its predictions in order to perform better \textit{on that specific task}. The main technical challenge associated with DFL is that it requires being able to differentiate through the optimization problem, which is difficult due to discontinuous solutions and other challenges. Past work has largely gotten around this this issue by \textit{handcrafting} task-specific surrogates to the original optimization problem that provide informative gradients when differentiated through. However, the need to handcraft surrogates for each new task limits the usability of DFL. In addition, there are often no guarantees about the convexity of the resulting surrogates and, as a result, training a predictive model using them can lead to inferior local optima. In this paper, we do away with surrogates altogether and instead \textit{learn} loss functions that capture task-specific information. To the best of our knowledge, ours is the first approach that entirely replaces the optimization component of decision-focused learning with a loss that is automatically learned. Our approach (a) only requires access to a black-box oracle that can solve the optimization problem and is thus \textit{generalizable}, and (b) can be \textit{convex by construction} and so can be easily optimized over. We evaluate our approach on three resource allocation problems from the literature and find that our approach outperforms learning without taking into account task-structure in all three domains, and even hand-crafted surrogates from the literature.

Rideshare and ride-pooling platforms use artificial intelligence-based matching algorithms to pair riders and drivers. However, these platforms can induce unfairness either through an unequal income distribution or disparate treatment of riders. We investigate two methods to reduce forms of inequality in ride-pooling platforms: by incorporating fairness constraints into the objective function and redistributing income to drivers who deserve more. To test these out, we use New York City taxi data to evaluate their performance on both the rider and driver side. For the first method, we find that optimizing for driver fairness out-performs state-of-the-art models in terms of the number of riders serviced, showing that optimizing for fairness can assist profitability in certain circumstances. For the second method, we explore income redistribution as a method to combat income inequality by having drivers keep an $r$ fraction of their income, and contribute the rest to a redistribution pool. For certain values of $r$, most drivers earn near their Shapley value, while still incentivizing drivers to maximize income, thereby avoiding the free-rider problem and reducing income variability. While the first method is useful because it improves both rider and driver-side fairness, the second method is useful because it improves fairness without affecting profitability, and both methods can be combined to improve rider and driver-side fairness.

Recently, there is increasing interest in multilingual automatic speech recognition (ASR) where a speech recognition system caters to multiple low resource languages by taking advantage of low amounts of labeled corpora in multiple languages. With multilingualism becoming common in today's world, there has been increasing interest in code-switching ASR as well. In code-switching, multiple languages are freely interchanged within a single sentence or between sentences. The success of low-resource multilingual and code-switching ASR often depends on the variety of languages in terms of their acoustics, linguistic characteristics as well as the amount of data available and how these are carefully considered in building the ASR system. In this challenge, we would like to focus on building multilingual and code-switching ASR systems through two different subtasks related to a total of seven Indian languages, namely Hindi, Marathi, Odia, Tamil, Telugu, Gujarati and Bengali. For this purpose, we provide a total of ~600 hours of transcribed speech data, comprising train and test sets, in these languages including two code-switched language pairs, Hindi-English and Bengali-English. We also provide a baseline recipe for both the tasks with a WER of 30.73% and 32.45% on the test sets of multilingual and code-switching subtasks, respectively.

Multi-action restless multi-armed bandits (RMABs) are a powerful framework for constrained resource allocation in which N independent processes are managed. However, previous work only study the offline setting where problem dynamics are known. We address this restrictive assumption, designing the first algorithms for learning good policies for Multi-action RMABs online using combinations of Lagrangian relaxation and Q-learning. Our first approach, MAIQL, extends a method for Q-learning the Whittle index in binary-action RMABs to the multi-action setting. We derive a generalized update rule and convergence proof and establish that, under standard assumptions, MAIQL converges to the asymptotically optimal multi-action RMAB policy as t → ∞. However, MAIQL relies on learning Q-functions and indexes on two timescales which leads to slow convergence and requires problem structure to perform well. Thus, we design a second algorithm, LPQL, which learns the well-performing and more general Lagrange policy for multi-action RMABs by learning to minimize the Lagrange bound through a variant of Q-learning. To ensure fast convergence, we take an approximation strategy that enables learning on a single timescale, then give a guarantee relating the approximation's precision to an upper bound of LPQL's return as t → ∞. Finally, we show that our approaches always outperform baselines across multiple settings, including one derived from real-world medication adherence data.

In the predict-then-optimize framework, the objective is to train a predictive model, mapping from environment features to parameters of an optimization problem, which maximizes decision quality when the optimization is subsequently solved. Recent work on decision-focused learning shows that embedding the optimization problem in the training pipeline can improve decision quality and help generalize better to unseen tasks compared to relying on an intermediate loss function for evaluating prediction quality. We study the predict-then-optimize framework in the context of sequential decision problems (formulated as MDPs) that are solved via reinforcement learning. In particular, we are given environment features and a set of trajectories from training MDPs, which we use to train a predictive model that generalizes to unseen test MDPs without trajectories. Two significant computational challenges arise in applying decision-focused learning to MDPs: (i) large state and action spaces make it infeasible for existing techniques to differentiate through MDP problems, and (ii) the high-dimensional policy space, as parameterized by a neural network, makes differentiating through a policy expensive. We resolve the first challenge by sampling provably unbiased derivatives to approximate and differentiate through optimality conditions, and the second challenge by using a low-rank approximation to the high-dimensional sample-based derivatives. We implement both Bellman-based and policy gradient-based decision-focused learning on three different MDP problems with missing parameters, and show that decision-focused learning performs better in generalization to unseen tasks.

On-demand ride-pooling (e.g., UberPool, LyftLine, GrabShare) has recently become popular because of its ability to lower costs for passengers while simultaneously increasing revenue for drivers and aggregation companies (e.g., Uber). Unlike in Taxi on Demand (ToD) services – where a vehicle is assigned one passenger at a time – in on-demand ride-pooling, each vehicle must simultaneously serve multiple passengers with heterogeneous origin and destination pairs without violating any quality constraints. To ensure near real-time response, existing solutions to the real-time ride-pooling problem are myopic in that they optimise the objective (e.g., maximise the number of passengers served) for the current time step without considering the effect such an assignment could have on assignments in future time steps. However, considering the future effects of an assignment that also has to consider what combinations of passenger requests can be assigned to vehicles adds a layer of combinatorial complexity to the already challenging problem of considering future effects in the ToD case.A popular approach that addresses the limitations of myopic assignments in ToD problems is Approximate Dynamic Programming (ADP). Existing ADP methods for ToD can only handle Linear Program (LP) based assignments, however, as the value update relies on dual values from the LP. The assignment problem in ride pooling requires an Integer Linear Program (ILP) that has bad LP relaxations. Therefore, our key technical contribution is in providing a general ADP method that can learn from the ILP based assignment found in ride-pooling. Additionally, we handle the extra combinatorial complexity from combinations of passenger requests by using a Neural Network based approximate value function and show a connection to Deep Reinforcement Learning that allows us to learn this value-function with increased stability and sample-efficiency. We show that our approach easily outperforms leading approaches for on-demand ride-pooling on a real-world dataset by up to 16%, a significant improvement in city-scale transportation problems.

Large-scale screening for potential threats with limited resources and capacity for screening is a problem of interest at airports, seaports, and other ports of entry. Adversaries can observe screening procedures and arrive at a time when there will be gaps in screening due to limited resource capacities. To capture this game between ports and adversaries, this problem has been previously represented as a Stackelberg game, referred to as a Threat Screening Game (TSG). Given the significant complexity associated with solving TSGs and uncertainty in arrivals of customers, existing work has assumed that screenees arrive and are allocated security resources at the beginning of the time-window. In practice, screenees such as airport passengers arrive in bursts correlated with flight time and are not bound by fixed time-windows. To address this, we propose an online threat screening model in which the screening strategy is determined adaptively as a passenger arrives while satisfying a hard bound on acceptable risk of not screening a threat. To solve the online problem, we first reformulate it as a Markov Decision Process (MDP) in which the hard bound on risk translates to a constraint on the action space and then solve the resultant MDP using Deep Reinforcement Learning (DRL). To this end, we provide a novel way to efficiently enforce linear inequality constraints on the action output in DRL. We show that our solution allows us to significantly reduce screenee wait time without compromising on the risk.

Visual Question Answering (VQA) has emerged as an important problem spanning Computer Vision, Natural Language Processing and Artificial Intelligence (AI). In conventional VQA, one may ask questions about an image which can be answered purely based on its content. For example, given an image with people in it, a typical VQA question may inquire about the number of people in the image. More recently, there is growing interest in answering questions which require commonsense knowledge involving common nouns (e.g., cats, dogs, microphones) present in the image. In spite of this progress, the important problem of answering questions requiring world knowledge about named entities (e.g., Barack Obama, White House, United Nations) in the image has not been addressed in prior research. We address this gap in this paper, and introduce KVQA – the first dataset for the task of (world) knowledge-aware VQA. KVQA consists of 183K question-answer pairs involving more than 18K named entities and 24K images. Questions in this dataset require multi-entity, multi-relation, and multi-hop reasoning over large Knowledge Graphs (KG) to arrive at an answer. To the best of our knowledge, KVQA is the largest dataset for exploring VQA over KG. Further, we also provide baseline performances using state-of-the-art methods on KVQA.