Bryan Wilder

Michael Poli, Stefano Massaroli, Stefano Ermon et al.

We present a methodology for formulating simplifying abstractions in machine learning systems by identifying and harnessing the utility structure of decisions. Machine learning tasks commonly involve high-dimensional output spaces (e.g., predictions for every pixel in an image or node in a graph), even though a coarser output would often suffice for downstream decision-making (e.g., regions of an image instead of pixels). Developers often hand-engineer abstractions of the output space, but numerous abstractions are possible and it is unclear how the choice of output space for a model impacts its usefulness in downstream decision-making. We propose a method that configures the output space automatically in order to minimize the loss of decision-relevant information. Taking a geometric perspective, we formulate a step of the algorithm as a projection of the probability simplex, termed fold, that minimizes the total loss of decision-related information in the H-entropy sense. Crucially, learning in the abstracted outcome space requires less data, leading to a net improvement in decision quality. We demonstrate the method in two domains: data acquisition for deep neural network training and a closed-loop wildfire management task.

Ananya Joshi, Kathryn Mazaitis, Roni Rosenfeld et al.

Irregularities in public health data streams (like COVID-19 Cases) hamper data-driven decision-making for public health stakeholders. A real-time, computer-generated list of the most important, outlying data points from thousands of daily-updated public health data streams could assist an expert reviewer in identifying these irregularities. However, existing outlier detection frameworks perform poorly on this task because they do not account for the data volume or for the statistical properties of public health streams. Accordingly, we developed FlaSH (Flagging Streams in public Health), a practical outlier detection framework for public health data users that uses simple, scalable models to capture these statistical properties explicitly. In an experiment where human experts evaluate FlaSH and existing methods (including deep learning approaches), FlaSH scales to the data volume of this task, matches or exceeds these other methods in mean accuracy, and identifies the outlier points that users empirically rate as more helpful. Based on these results, FlaSH has been deployed on data streams used by public health stakeholders.

Ruiqi Lyu, Roni Rosenfeld, Bryan Wilder

Epidemic surveillance is a challenging task, especially when crucial data is fragmented across institutions and data custodians are unable or unwilling to share it. This study aims to explore the feasibility of a simple federated surveillance approach. The idea is to conduct hypothesis tests for a rise in counts behind each custodian's firewall and then combine p-values from these tests using techniques from meta-analysis. We propose a hypothesis testing framework to identify surges in epidemic-related data streams and conduct experiments on real and semi-synthetic data to assess the power of different p-value combination methods to detect surges without needing to combine the underlying counts. Our findings show that relatively simple combination methods achieve a high degree of fidelity and suggest that infectious disease outbreaks can be detected without needing to share even aggregate data across institutions.

Aditya Mate, Bryan Wilder, Aparna Taneja et al.

We consider the task of evaluating policies of algorithmic resource allocation through randomized controlled trials (RCTs). Such policies are tasked with optimizing the utilization of limited intervention resources, with the goal of maximizing the benefits derived. Evaluation of such allocation policies through RCTs proves difficult, notwithstanding the scale of the trial, because the individuals' outcomes are inextricably interlinked through resource constraints controlling the policy decisions. Our key contribution is to present a new estimator leveraging our proposed novel concept, that involves retrospective reshuffling of participants across experimental arms at the end of an RCT. We identify conditions under which such reassignments are permissible and can be leveraged to construct counterfactual trials, whose outcomes can be accurately ascertained, for free. We prove theoretically that such an estimator is more accurate than common estimators based on sample means -- we show that it returns an unbiased estimate and simultaneously reduces variance. We demonstrate the value of our approach through empirical experiments on synthetic, semi-synthetic as well as real case study data and show improved estimation accuracy across the board.

JungHo Lee, Johnna Sundberg, Pim Welle et al.

Public service programs often allocate limited resources under uncertainty about their benefits, creating a need for randomization to support credible evaluation. In practice, however, applicants commonly enter waitlists where resources are prioritized toward individuals judged to have higher need through tiered priority queues, making direct randomization difficult. Motivated by this, we develop an experimental design framework for learning treatment effects while treating those most in need where incoming applicants are randomized into priority queues based on their assessed risk scores. Treatments are then provided across queues in priority order and first-in-first-out within queue as budget becomes available. Our contributions are two-fold. First, we characterize what causal effects are identified under this priority-queue allocation. When arrivals are exogenous, treatments are conditionally randomized, and hence standard estimands are identified; when arrivals are endogenous, queue randomization instead provides an instrument for treatment, identifying local treatment effects induced by the queuing process. Second, we develop optimized queue-assignment designs that trade off statistical efficiency against prioritizing higher-need applicants. We show in the process that, despite dependence in treatment assignments induced by the design, usual iid efficiency bounds remain well-justified design objectives. We illustrate the proposed designs using data from a housing allocation program in a large U.S. county.

Khurram Yamin, Jingjing Tang, Eric Horvitz et al.

LLMs are increasingly used to make or support high-stakes decisions under uncertainty, where alignment depends not only on factual accuracy but on how models weigh tradeoffs between different outcomes. We present an empirical pipeline for estimating the implied preferences that an LLM's observed choices optimize: we elicit the model's probability distribution over unknowns along with the choice it would make for the decision task and then fit a discrete choice model to recover the cost function that best rationalizes the model's decisions. We show how this revealed-preference description allows rigorous evaluation of whether models behave in a consistently goal-directed way, whether they can verbalize a description of their objectives which matches their revealed decision policy, and whether prompting can reliably steer those policies to implement a user-specified cost function. We apply this evaluation across four medical diagnosis domains and multiple frontier and open-source models. We find that while many models have a nontrivial degree of internal coherence, they also have significant weaknesses in faithfully reporting or adopting preferences in response to user direction.

Ruiqi Lyu, Alistair Turcan, Bryan Wilder

Accurate epidemic forecasting is crucial for public health response, resource allocation, and outbreak intervention, but remains difficult with sparse, noisy, and highly non-stationary data. Because epidemics unfold across interacting regions, spatiotemporal methods are natural candidates for improving forecasts. Despite growing interest in spatial information, no standardized benchmark exists, and current evaluations often use simple chronological train-test splits that do not reflect real-time forecasting practice. We address this gap with SpatialEpiBench, a challenging benchmark for spatiotemporal epidemic forecasting in realistic public-health settings. SpatialEpiBench includes 11 epidemic datasets with standardized rolling evaluations and outbreak-specific metrics. We evaluate adjacency-informed forecasting models with widely used epidemic priors that adapt general models to epidemiology, but find that most methods underperform a simple last-value baseline from 1 day to 1 month ahead, even during outbreaks and with these priors. We identify three major failure modes: (1) poor outbreak anticipation, (2) difficulty handling sparsity and noise, and (3) limited utility of common geographic adjacency for epidemiological spatial information. We release benchmark data, code, and instructions at https://github.com/Rachel-Lyu/SpatialEpiBench to support development of operationally useful epidemic forecasting models.

Naveen Raman, Santiago Cortes-Gomez, Mateo Dulce Rubio et al.

Benchmarks are necessary for healthcare evaluation, but are not sufficient for predicting deployment performance. Our position is that the evaluation--deployment gap arises not because of poorly designed benchmarks, but from implicit assumptions about how users interact with models that cannot be surfaced from benchmarks alone. To make this precise, we propose a classification of assumptions into two categories: task, which can be tested from conversation data alone, and outcome, which requires outcome data and behavioral studies for testing. Critically, outcome assumptions depend on human behavior, something that even well-designed benchmarks cannot directly observe. To demonstrate the operationality of this framework, we retrospectively analyze a healthcare RCT as a case study and find that the gap naturally separates into task and outcome gaps of roughly equal size. To address this, we make two contributions: first, we propose BenchmarkCards, an artifact that documents assumptions, and second, we propose staged evaluation, a procedure that systematically tests assumptions and evaluates performance.

Ivaxi Sheth, Zhijing Jin, Bryan Wilder et al.

In the presence of confounding between an endogenous variable and the outcome, instrumental variables (IVs) are used to isolate the causal effect of the endogenous variable. Identifying valid instruments requires interdisciplinary knowledge, creativity, and contextual understanding, making it a non-trivial task. In this paper, we investigate whether large language models (LLMs) can aid in this task. We perform a two-stage evaluation framework. First, we test whether LLMs can recover well-established instruments from the literature, assessing their ability to replicate standard reasoning. Second, we evaluate whether LLMs can identify and avoid instruments that have been empirically or theoretically discredited. Building on these results, we introduce IV Co-Scientist, a multi-agent system that proposes, critiques, and refines IVs for a given treatment-outcome pair. We also introduce a statistical test to contextualize consistency in the absence of ground truth. Our results show the potential of LLMs to discover valid instrumental variables from a large observational database.

Smriti Jha, Vidhi Jain, Jianyu Xu et al.

The ability to provide trustworthy maternal health information using phone-based chatbots can have a significant impact, particularly in low-resource settings where users have low health literacy and limited access to care. However, deploying such systems is technically challenging: user queries are short, underspecified, and code-mixed across languages, answers require regional context-specific grounding, and partial or missing symptom context makes safe routing decisions difficult. We present a chatbot for maternal health in India developed through a partnership between academic researchers, a health tech company, a public health nonprofit, and a hospital. The system combines (1) stage-aware triage, routing high-risk queries to expert templates, (2) hybrid retrieval over curated maternal/newborn guidelines, and (3) evidence-conditioned generation from an LLM. Our core contribution is an evaluation workflow for high-stakes deployment under limited expert supervision. Targeting both component-level and end-to-end testing, we introduce: (i) a labeled triage benchmark (N=150) achieving 86.7% emergency recall, explicitly reporting the missed-emergency vs. over-escalation trade-off; (ii) a synthetic multi-evidence retrieval benchmark (N=100) with chunk-level evidence labels; (iii) LLM-as-judge comparison on real queries (N=781) using clinician-codesigned criteria; and (iv) expert validation. Our findings show that trustworthy medical assistants in multilingual, noisy settings require defense-in-depth design paired with multi-method evaluation, rather than any single model and evaluation method choice.

Kevin Ren, Yewon Byun, Bryan Wilder

Distribution shift is a key challenge for predictive models in practice, creating the need to identify potentially harmful shifts in advance of deployment. Existing work typically defines these worst-case shifts as ones that most degrade the individual-level accuracy of the model. However, when models are used to make a downstream population-level decision like the allocation of a scarce resource, individual-level accuracy may be a poor proxy for performance on the task at hand. We introduce a novel framework that employs a hierarchical model structure to identify worst-case distribution shifts in predictive resource allocation settings by capturing shifts both within and across instances of the decision problem. This task is more difficult than in standard distribution shift settings due to combinatorial interactions, where decisions depend on the joint presence of individuals in the allocation task. We show that the problem can be reformulated as a submodular optimization problem, enabling efficient approximations of worst-case loss. Applying our framework to real data, we find empirical evidence that worst-case shifts identified by one metric often significantly diverge from worst-case distributions identified by other metrics.

Ananya Joshi, Tina Townes, Nolan Gormley et al.

Disease control experts inspect public health data streams daily for outliers worth investigating, like those corresponding to data quality issues or disease outbreaks. However, they can only examine a few of the thousands of maximally-tied outliers returned by univariate outlier detection methods applied to large-scale public health data streams. To help experts distinguish the most important outliers from these thousands of tied outliers, we propose a new task for algorithms to rank the outputs of any univariate method applied to each of many streams. Our novel algorithm for this task, which leverages hierarchical networks and extreme value analysis, performed the best across traditional outlier detection metrics in a human-expert evaluation using public health data streams. Most importantly, experts have used our open-source Python implementation since April 2023 and report identifying outliers worth investigating 9.1x faster than their prior baseline. Other organizations can readily adapt this implementation to create rankings from the outputs of their tailored univariate methods across large-scale streams.

Khurram Yamin, Jingjing Tang, Santiago Cortes-Gomez et al.

Large language models (LLMs) are increasingly deployed as agents in high-stakes domains where optimal actions depend on both uncertainty about the world and consideration of utilities of different outcomes, yet their decision logic remains difficult to interpret. We study whether LLMs are rational utility maximizers with coherent beliefs and stable preferences. We consider behaviors of models for diagnosis challenge problems. The results provide insights about the relationship of LLM inferences to ideal Bayesian utility maximization for elicited probabilities and observed actions. Our approach provides falsifiable conditions under which the reported probabilities \emph{cannot} correspond to the true beliefs of any rational agent. We apply this methodology to multiple medical diagnostic domains with evaluations across several LLMs. We discuss implications of the results and directions forward for uses of LLMs in guiding high-stakes decisions.

Bryan Wilder, Pim Welle

Many social programs attempt to allocate scarce resources to people with the greatest need. Indeed, public services increasingly use algorithmic risk assessments motivated by this goal. However, targeting the highest-need recipients often conflicts with attempting to evaluate the causal effect of the program as a whole, as the best evaluations would be obtained by randomizing the allocation. We propose a framework to design randomized allocation rules which optimally balance targeting high-need individuals with learning treatment effects, presenting policymakers with a Pareto frontier between the two goals. We give sample complexity guarantees for the policy learning problem and provide a computationally efficient strategy to implement it. We then collaborate with the human services department of Allegheny County, Pennsylvania to evaluate our methods on data from real service delivery settings. Optimized policies can substantially mitigate the tradeoff between learning and targeting. For example, it is often possible to obtain 90% of the optimal utility in targeting high-need individuals while ensuring that the average treatment effect can be estimated with less than 2 times the samples that a randomized controlled trial would require. Mechanisms for targeting public services often focus on measuring need as accurately as possible. However, our results suggest that algorithmic systems in public services can be most impactful if they incorporate program evaluation as an explicit goal alongside targeting.

Bryan Wilder, Eric Ewing, Bistra Dilkina et al.

Real-world applications often combine learning and optimization problems on graphs. For instance, our objective may be to cluster the graph in order to detect meaningful communities (or solve other common graph optimization problems such as facility location, maxcut, and so on). However, graphs or related attributes are often only partially observed, introducing learning problems such as link prediction which must be solved prior to optimization. Standard approaches treat learning and optimization entirely separately, while recent machine learning work aims to predict the optimal solution directly from the inputs. Here, we propose an alternative decision-focused learning approach that integrates a differentiable proxy for common graph optimization problems as a layer in learned systems. The main idea is to learn a representation that maps the original optimization problem onto a simpler proxy problem that can be efficiently differentiated through. Experimental results show that our ClusterNet system outperforms both pure end-to-end approaches (that directly predict the optimal solution) and standard approaches that entirely separate learning and optimization. Code for our system is available at https://github.com/bwilder0/clusternet.

Santiago Cortes-Gomez, Mateo Dulce Rubio, Carlos Patino et al.

The rise of machine learning has shifted targeted resource allocation in policy and humanitarian settings toward algorithmic targeting based on predicted risk scores. This approach is typically cheaper and faster than traditional screening procedures that directly observe the latent vulnerability status through physical verification. Yet, even access to the true conditional vulnerability probability cannot eliminate misallocation: aleatoric uncertainty over individual vulnerability status is irreducible, and probabilistic targeting inevitably misallocates some resources. In this work we study how screening and algorithmic targeting should be optimally combined in a two-stage allocation framework where a screening stage observes true outcomes for a subset of units before a final allocation stage assigns the resource under a fixed coverage budget. We show that the optimal strategy screens units at the margin of algorithmic allocation, while directly targeting the highest-risk units. Furthermore, we empirically characterize when screening and algorithmic targeting act as complements or substitutes: efficiency gains from screening grow as the aleatoric uncertainty in the population increases. We illustrate our framework with applications in income-based social protection programs and humanitarian demining in Colombia, where the tension between screening costs and allocation efficiency is operationally consequential.

Yewon Byun, Dylan Sam, Michael Oberst et al.

Many definitions of fairness or inequity involve unobservable causal quantities that cannot be directly estimated without strong assumptions. For instance, it is particularly difficult to estimate notions of fairness that rely on hard-to-measure concepts such as risk (e.g., quantifying whether patients at the same risk level have equal probability of treatment, regardless of group membership). Such measurements of risk can be accurately obtained when no unobserved confounders have jointly influenced past decisions and outcomes. However, in the real world, this assumption rarely holds. In this paper, we show that, surprisingly, one can still compute meaningful bounds on treatment rates for high-risk individuals (i.e., conditional on their true, \textit{unobserved} negative outcome), even when entirely eliminating or relaxing the assumption that we observe all relevant risk factors used by decision makers. We use the fact that in many real-world settings (e.g., the release of a new treatment) we have data from prior to any allocation to derive unbiased estimates of risk. This result enables us to audit unfair outcomes of existing decision-making systems in a principled manner. We demonstrate the effectiveness of our framework with a real-world study of Paxlovid allocation, provably identifying that observed racial inequity cannot be explained by unobserved confounders of the same strength as important observed covariates.

Ruiqi Lyu, Alistair Turcan, Bryan Wilder

Responding to disease outbreaks requires close surveillance of their trajectories, but outbreak detection is hindered by the high noise in epidemic time series. Aggregating information across data sources has shown great denoising ability in other fields, but remains underexplored in epidemiology. Here, we present LRTrend, an interpretable machine learning framework to identify outbreaks in real time. LRTrend effectively aggregates diverse health and behavioral data streams within one region and learns disease-specific epidemic networks to aggregate information across regions. We reveal diverse epidemic clusters and connections across the United States that are not well explained by commonly used human mobility networks and may be informative for future public health coordination. We apply LRTrend to 2 years of COVID-19 data in 305 hospital referral regions and frequently detect regional Delta and Omicron waves within 2 weeks of the outbreak's start, when case counts are a small fraction of the wave's resulting peak.

Arpan Dasgupta, Niclas Boehmer, Neha Madhiwalla et al.

Automated voice calls are an effective method of delivering maternal and child health information to mothers in underserved communities. One method to fight dwindling listenership is through an intervention in which health workers make live service calls. Previous work has shown that we can use AI to identify beneficiaries whose listenership gets the greatest boost from an intervention. It has also been demonstrated that listening to the automated voice calls consistently leads to improved health outcomes for the beneficiaries of the program. These two observations combined suggest the positive effect of AI-based intervention scheduling on behavioral and health outcomes. This study analyzes the relationship between the two. Specifically, we are interested in mothers' health knowledge in the post-natal period, measured through survey questions. We present evidence that improved listenership through AI-scheduled interventions leads to a better understanding of key health issues during pregnancy and infancy. This improved understanding has the potential to benefit the health outcomes of mothers and their babies.

Lily Xu, Bryan Wilder, Elias B. Khalil et al.

Reinforcement learning (RL) has increasingly been applied to solve real-world planning problems, with progress in handling large state spaces and time horizons. However, a key bottleneck in many domains is that RL methods cannot accommodate large, combinatorially structured action spaces. In such settings, even representing the set of feasible actions at a single step may require a complex discrete optimization formulation. We leverage recent advances in embedding trained neural networks into optimization problems to propose SEQUOIA, an RL algorithm that directly optimizes for long-term reward over the feasible action space. Our approach embeds a Q-network into a mixed-integer program to select a combinatorial action in each timestep. Here, we focus on planning over restless bandits, a class of planning problems which capture many real-world examples of sequential decision making. We introduce coRMAB, a broader class of restless bandits with combinatorial actions that cannot be decoupled across the arms of the restless bandit, requiring direct solving over the joint, exponentially large action space. We empirically validate SEQUOIA on four novel restless bandit problems with combinatorial constraints: multiple interventions, path constraints, bipartite matching, and capacity constraints. Our approach significantly outperforms existing methods -- which cannot address sequential planning and combinatorial selection simultaneously -- by an average of 24.8\% on these difficult instances.

Khurram Yamin, Shantanu Gupta, Gaurav R. Ghosal et al.

The ability to robustly identify causal relationships is essential for autonomous decision-making and adaptation to novel scenarios. However, accurately inferring causal structure requires integrating both world knowledge and abstract logical reasoning. In this work, we investigate the interaction between these two capabilities through the representative task of causal reasoning over narratives. Through controlled synthetic, semi-synthetic, and real-world experiments, we find that state-of-the-art large language models (LLMs) often rely on superficial heuristics -- for example, inferring causality from event order or recalling memorized world knowledge without attending to context. Furthermore, we show that simple reformulations of the task can elicit more robust reasoning behavior. Our evaluation spans a range of causal structures, from linear chains to complex graphs involving colliders and forks. These findings uncover systematic patterns in how LLMs perform causal reasoning and lay the groundwork for developing methods that better align LLM behavior with principled causal inference.

Lydia T. Liu, Inioluwa Deborah Raji, Angela Zhou et al.

Many automated decision systems (ADS) are designed to solve prediction problems -- where the goal is to learn patterns from a sample of the population and apply them to individuals from the same population. In reality, these prediction systems operationalize holistic policy interventions in deployment. Once deployed, ADS can shape impacted population outcomes through an effective policy change in how decision-makers operate, while also being defined by past and present interactions between stakeholders and the limitations of existing organizational, as well as societal, infrastructure and context. In this work, we consider the ways in which we must shift from a prediction-focused paradigm to an interventionist paradigm when considering the impact of ADS within social systems. We argue this requires a new default problem setup for ADS beyond prediction, to instead consider predictions as decision support, final decisions, and outcomes. We highlight how this perspective unifies modern statistical frameworks and other tools to study the design, implementation, and evaluation of ADS systems, and point to the research directions necessary to operationalize this paradigm shift. Using these tools, we characterize the limitations of focusing on isolated prediction tasks, and lay the foundation for a more intervention-oriented approach to developing and deploying ADS.

Khurram Yamin, Gaurav Ghosal, Bryan Wilder

Large Language Models have been shown to contain extensive world knowledge in their parameters, enabling impressive performance on many knowledge intensive tasks. However, when deployed in novel settings, LLMs often encounter situations where they must integrate parametric knowledge with new or unfamiliar information. In this work, we explore whether LLMs can combine knowledge in-context with their parametric knowledge through the lens of counterfactual reasoning. Through synthetic and real experiments in multi-hop reasoning problems, we show that LLMs generally struggle with counterfactual reasoning, often resorting to exclusively using their parametric knowledge. Moreover, we show that simple post-hoc finetuning can struggle to instill counterfactual reasoning ability -- often leading to degradation in stored parametric knowledge. Ultimately, our work reveals important limitations of current LLM's abilities to re-purpose parametric knowledge in novel settings.

Vibhhu Sharma, Bryan Wilder

Machine learning is increasingly used to select which individuals receive limited-resource interventions in domains such as human services, education, development, and more. However, it is often not apparent what the right quantity is for models to predict. Policymakers rarely have access to data from a randomized controlled trial (RCT) that would enable accurate estimates of which individuals would benefit more from the intervention, while observational data creates a substantial risk of bias in treatment effect estimates. Practitioners instead commonly use a technique termed ``risk-based targeting" where the model is just used to predict each individual's status quo outcome (an easier, non-causal task). Those with higher predicted risk are offered treatment. There is currently almost no empirical evidence to inform which choices lead to the most effective machine learning-informed targeting strategies in social domains. In this work, we use data from 5 real-world RCTs in a variety of domains to empirically assess such choices. We find that when treatment effects can be estimated with high accuracy (which we simulate by allowing the model to partially observe outcomes in advance), treatment effect based targeting substantially outperforms risk-based targeting, even when treatment effect estimates are biased. Moreover, these results hold even when the policymaker has strong normative preferences for assisting higher-risk individuals. However, the features and data actually available in most RCTs we examine do not suffice for accurate estimates of heterogeneous treatment effects. Our results suggest treatment effect targeting has significant potential benefits, but realizing these benefits requires improvements to data collection and model training beyond what is currently common in practice.

Kanad Pardeshi, Bryan Wilder, Aarti Singh

Online learning algorithms continually update their models as data arrive, making it essential to accurately estimate the expected loss at the current time step. The prequential method is an effective estimation approach which can be practically deployed in various ways. However, theoretical guarantees have previously been established under strong conditions on the algorithm, and practical algorithms have hyperparameters which require careful tuning. We introduce OEUVRE, an estimator that evaluates each incoming sample on the function learned at the current and previous time steps, recursively updating the loss estimate in constant time and memory. We use algorithmic stability, a property satisfied by many popular online learners, for optimal updates and prove consistency, convergence rates, and concentration bounds for our estimator. We design a method to adaptively tune OEUVRE's hyperparameters and test it across diverse online and stochastic tasks. We observe that OEUVRE matches or outperforms other estimators even when their hyperparameters are tuned with oracle access to ground truth.

Maitreyi Swaroop, Tamar Krishnamurti, Bryan Wilder

We study the problem of selecting limited features to observe such that models trained on them can perform well simultaneously across multiple subpopulations. This problem has applications in settings where collecting each feature is costly, e.g. requiring adding survey questions or physical sensors, and we must be able to use the selected features to create high-quality downstream models for different populations. Our method frames the problem as a continuous relaxation of traditional variable selection using a noising mechanism, without requiring backpropagation through model training processes. By optimizing over the variance of a Bayes-optimal predictor, we develop a model-agnostic framework that balances overall performance of downstream prediction across populations. We validate our approach through experiments on both synthetic datasets and real-world data.

Bryan Wilder, Angela Zhou

There has been increasing research interest in AI/ML for social impact, and correspondingly more publication venues have refined review criteria for practice-driven AI/ML research. However, these review guidelines tend to most concretely recognize projects that simultaneously achieve deployment and novel ML methodological innovation. We argue that this introduces incentives for researchers that undermine the sustainability of a broader research ecosystem of social impact, which benefits from projects that make contributions on single front (applied or methodological) that may better meet project partner needs. Our position is that researchers and reviewers in machine learning for social impact must simultaneously adopt: 1) a more expansive conception of social impacts beyond deployment and 2) more rigorous evaluations of the impact of deployed systems.

Jianyu Xu, Vidhi Jain, Bryan Wilder et al.

With advances in generative AI, decision-making agents can now dynamically create new actions during online learning, but action generation typically incurs costs that must be balanced against potential benefits. We study an online learning problem where an agent can generate new actions at any time step by paying a one-time cost, with these actions becoming permanently available for future use. The challenge lies in learning the optimal sequence of two-fold decisions: which action to take and when to generate new ones, further complicated by the triangular tradeoffs among exploitation, exploration and $\textit{creation}$. To solve this problem, we propose a doubly-optimistic algorithm that employs Lower Confidence Bounds (LCB) for action selection and Upper Confidence Bounds (UCB) for action generation. Empirical evaluation on healthcare question-answering datasets demonstrates that our approach achieves favorable generation-quality tradeoffs compared to baseline strategies. From theoretical perspectives, we prove that our algorithm achieves the optimal regret of $O(T^{\frac{d}{d+2}}d^{\frac{d}{d+2}} + d\sqrt{T\log T})$, providing the first sublinear regret bound for online learning with expanding action spaces.

Ruiqi Lyu, Alistair Turcan, Martin Jinye Zhang et al.

Learning causal structure from observational data is central to scientific modeling and decision-making. Constraint-based methods aim to recover conditional independence (CI) relations in a causal directed acyclic graph (DAG). Classical approaches such as PC and subsequent methods orient v-structures first and then propagate edge directions from these seeds, assuming perfect CI tests and exhaustive search of separating subsets -- assumptions often violated in practice, leading to cascading errors in the final graph. Recent work has explored using large language models (LLMs) as experts, prompting sets of nodes for edge directions, and could augment edge orientation when assumptions are not met. However, such methods implicitly assume perfect experts, which is unrealistic for hallucination-prone LLMs. We propose MosaCD, a causal discovery method that propagates edges from a high-confidence set of seeds derived from both CI tests and LLM annotations. To filter hallucinations, we introduce shuffled queries that exploit LLMs' positional bias, retaining only high-confidence seeds. We then apply a novel confidence-down propagation strategy that orients the most reliable edges first, and can be integrated with any skeleton-based discovery method. Across multiple real-world graphs, MosaCD achieves higher accuracy in final graph construction than existing constraint-based methods, largely due to the improved reliability of initial seeds and robust propagation strategies.

Kevin Ren, Santiago Cortes-Gomez, Carlos Miguel Patiño et al.

Recent work has investigated the capabilities of large language models (LLMs) as zero-shot models for generating individual-level characteristics (e.g., to serve as risk models or augment survey datasets). However, when should a user have confidence that an LLM will provide high-quality predictions for their particular task? To address this question, we conduct a large-scale empirical study of LLMs' zero-shot predictive capabilities across a wide range of tabular prediction tasks. We find that LLMs' performance is highly variable, both on tasks within the same dataset and across different datasets. However, when the LLM performs well on the base prediction task, its predicted probabilities become a stronger signal for individual-level accuracy. Then, we construct metrics to predict LLMs' performance at the task level, aiming to distinguish between tasks where LLMs may perform well and where they are likely unsuitable. We find that some of these metrics, each of which are assessed without labeled data, yield strong signals of LLMs' predictive performance on new tasks.

Yewon Byun, Shantanu Gupta, Zachary C. Lipton et al.

Predictions and generations from large language models are increasingly being explored as an aid in limited data regimes, such as in computational social science and human subjects research. While prior technical work has mainly explored the potential to use model-predicted labels for unlabeled data in a principled manner, there is increasing interest in using large language models to generate entirely new synthetic samples (e.g., synthetic simulations), such as in responses to surveys. However, it remains unclear by what means practitioners can combine such data with real data and yet produce statistically valid conclusions upon them. In this paper, we introduce a new estimator based on generalized method of moments, providing a hyperparameter-free solution with strong theoretical guarantees to address this challenge. Intriguingly, we find that interactions between the moment residuals of synthetic data and those of real data (i.e., when they are predictive of each other) can greatly improve estimates of the target parameter. We validate the finite-sample performance of our estimator across different tasks in computational social science applications, demonstrating large empirical gains.

Khurram Yamin, Edward Kennedy, Bryan Wilder

Policymakers in resource-constrained settings require experimental designs that satisfy strict budget limits while ensuring precise estimation of treatment effects. We propose a framework that applies a dependent randomized rounding procedure to convert assignment probabilities into binary treatment decisions. Our proposed solution preserves the marginal treatment probabilities while inducing negative correlations among assignments, leading to improved estimator precision through variance reduction. We establish theoretical guarantees for the inverse propensity weighted and general linear estimators, and demonstrate through empirical studies that our approach yields efficient and accurate inference under fixed budget constraints.

Ananya Joshi, Nolan Gormley, Richa Gadgil et al.

Public health experts need scalable approaches to monitor large volumes of health data (e.g., cases, hospitalizations, deaths) for outbreaks or data quality issues. Traditional alert-based monitoring systems struggle with modern public health data monitoring systems for several reasons, including that alerting thresholds need to be constantly reset and the data volumes may cause application lag. Instead, we propose a ranking-based monitoring paradigm that leverages new AI anomaly detection methods. Through a multi-year interdisciplinary collaboration, the resulting system has been deployed at a national organization to monitor up to 5,000,000 data points daily. A three-month longitudinal deployed evaluation revealed a significant improvement in monitoring objectives, with a 54x increase in reviewer speed efficiency compared to traditional alert-based methods. This work highlights the potential of human-centered AI to transform public health decision-making.

Ruiqi Lyu, Alistair Turcan, Bryan Wilder

Regardless the amount of data a machine learning (ML) model is trained on, there will inevitably be data that differs from their training set, lowering model performance. Concept shift occurs when the distribution of labels conditioned on the features changes, making even a well-tuned ML model to have learned a fundamentally incorrect representation. Identifying these shifted features provides unique insight into how one dataset differs from another, considering the difference may be across a scientifically relevant dimension, such as time, disease status, population, etc. In this paper, we propose SGShift, a model for detecting concept shift in tabular data and attributing reduced model performance to a sparse set of shifted features. SGShift models concept shift with a Generalized Additive Model (GAM) and performs subsequent feature selection to identify shifted features. We propose further extensions of SGShift by incorporating knockoffs to control false discoveries and an absorption term to account for models with poor fit to the data. We conduct extensive experiments in synthetic and real data across various ML models and find SGShift can identify shifted features with AUC $>0.9$ and recall $>90\%$, often 2 or 3 times as high as baseline methods.

Yewon Byun, Sanket Vaibhav Mehta, Saurabh Garg et al. · deepmind

In many real-world settings, regulations and economic incentives permit the sharing of models but not data across institutional boundaries. In such scenarios, practitioners might hope to adapt models to new domains, without losing performance on previous domains (so-called catastrophic forgetting). While any single model may struggle to achieve this goal, learning an ensemble of domain-specific experts offers the potential to adapt more closely to each individual institution. However, a core challenge in this context is determining which expert to deploy at test time. In this paper, we propose Generate to Discriminate (G2D), a domain-incremental continual learning method that leverages synthetic data to train a domain-discriminator that routes samples at inference time to the appropriate expert. Surprisingly, we find that leveraging synthetic data in this capacity is more effective than using the samples to \textit{directly} train the downstream classifier (the more common approach to leveraging synthetic data in the lifelong learning literature). We observe that G2D outperforms competitive domain-incremental learning methods on tasks in both vision and language modalities, providing a new perspective on the use of synthetic data in the lifelong learning literature.

Khurram Yamin, Vibhhu Sharma, Ed Kennedy et al.

Many applications of causal inference require using treatment effects estimated on a study population to make decisions in a separate target population. We consider the challenging setting where there are covariates that are observed in the target population that were not seen in the original study. Our goal is to estimate the tightest possible bounds on heterogeneous treatment effects conditioned on such newly observed covariates. We introduce a novel partial identification strategy based on ideas from ecological inference; the main idea is that estimates of conditional treatment effects for the full covariate set must marginalize correctly when restricted to only the covariates observed in both populations. Furthermore, we introduce a bias-corrected estimator for these bounds and prove that it enjoys fast convergence rates and statistical guarantees (e.g., asymptotic normality). Experimental results on both real and synthetic data demonstrate that our framework can produce bounds that are much tighter than would otherwise be possible.

Justin Whitehouse, Christopher Jung, Vasilis Syrgkanis et al.

Estimates of heterogeneous treatment effects such as conditional average treatment effects (CATEs) and conditional quantile treatment effects (CQTEs) play an important role in real-world decision making. Given this importance, one should ensure these estimates are calibrated. While there is a rich literature on calibrating estimators of non-causal parameters, very few methods have been derived for calibrating estimators of causal parameters, or more generally estimators of quantities involving nuisance parameters. In this work, we develop general algorithms for reducing the task of causal calibration to that of calibrating a standard (non-causal) predictive model. Throughout, we study a notion of calibration defined with respect to an arbitrary, nuisance-dependent loss $\ell$, under which we say an estimator $θ$ is calibrated if its predictions cannot be changed on any level set to decrease loss. For losses $\ell$ satisfying a condition called universal orthogonality, we present a simple algorithm that transforms partially-observed data into generalized pseudo-outcomes and applies any off-the-shelf calibration procedure. For losses $\ell$ satisfying a weaker assumption called conditional orthogonality, we provide a similar sample splitting algorithm the performs empirical risk minimization over an appropriately defined class of functions. Convergence of both algorithms follows from a generic, two term upper bound of the calibration error of any model. We demonstrate the practical applicability of our results in experiments on both observational and synthetic data. Our results are exceedingly general, showing that essentially any existing calibration algorithm can be used in causal settings, with additional loss only arising from errors in nuisance estimation.

Ben Chugg, Santiago Cortes-Gomez, Bryan Wilder et al.

We provide practical, efficient, and nonparametric methods for auditing the fairness of deployed classification and regression models. Whereas previous work relies on a fixed-sample size, our methods are sequential and allow for the continuous monitoring of incoming data, making them highly amenable to tracking the fairness of real-world systems. We also allow the data to be collected by a probabilistic policy as opposed to sampled uniformly from the population. This enables auditing to be conducted on data gathered for another purpose. Moreover, this policy may change over time and different policies may be used on different subpopulations. Finally, our methods can handle distribution shift resulting from either changes to the model or changes in the underlying population. Our approach is based on recent progress in anytime-valid inference and game-theoretic statistics-the "testing by betting" framework in particular. These connections ensure that our methods are interpretable, fast, and easy to implement. We demonstrate the efficacy of our approach on three benchmark fairness datasets.

Sanket Shah, Andrew Perrault, Bryan Wilder et al.

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 the structure of a decision-making task be used 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.

Sanket Shah, Kai Wang, Bryan Wilder et al.

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 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 issue by 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 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 generalizable, and (b) can be 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.

Kai Wang, Bryan Wilder, Sze-chuan Suen et al.

There is significant interest in learning and optimizing a complex system composed of multiple sub-components, where these components may be agents or autonomous sensors. Among the rich literature on this topic, agent-based and domain-specific simulations can capture complex dynamics and subgroup interaction, but optimizing over such simulations can be computationally and algorithmically challenging. Bayesian approaches, such as Gaussian processes (GPs), can be used to learn a computationally tractable approximation to the underlying dynamics but typically neglect the detailed information about subgroups in the complicated system. We attempt to find the best of both worlds by proposing the idea of decomposed feedback, which captures group-based heterogeneity and dynamics. We introduce a novel decomposed GP regression to incorporate the subgroup decomposed feedback. Our modified regression has provably lower variance -- and thus a more accurate posterior -- compared to previous approaches; it also allows us to introduce a decomposed GP-UCB optimization algorithm that leverages subgroup feedback. The Bayesian nature of our method makes the optimization algorithm trackable with a theoretical guarantee on convergence and no-regret property. To demonstrate the wide applicability of this work, we execute our algorithm on two disparate social problems: infectious disease control in a heterogeneous population and allocation of distributed weather sensors. Experimental results show that our new method provides significant improvement compared to the state-of-the-art.

James Kotary, Ferdinando Fioretto, Pascal Van Hentenryck et al.

This paper surveys the recent attempts at leveraging machine learning to solve constrained optimization problems. It focuses on surveying the work on integrating combinatorial solvers and optimization methods with machine learning architectures. These approaches hold the promise to develop new hybrid machine learning and optimization methods to predict fast, approximate, solutions to combinatorial problems and to enable structural logical inference. This paper presents a conceptual review of the recent advancements in this emerging area.

Bryan Wilder, Michael J. Mina, Milind Tambe

The COVID-19 pandemic provides new motivation for a classic problem in epidemiology: estimating the empirical rate of transmission during an outbreak (formally, the time-varying reproduction number) from case counts. While standard methods exist, they work best at coarse-grained national or state scales with abundant data, and struggle to accommodate the partial observability and sparse data common at finer scales (e.g., individual schools or towns). For example, case counts may be sparse when only a small fraction of infections are caught by a testing program. Or, whether an infected individual tests positive may depend on the kind of test and the point in time when they are tested. We propose a Bayesian framework which accommodates partial observability in a principled manner. Our model places a Gaussian process prior over the unknown reproduction number at each time step and models observations sampled from the distribution of a specific testing program. For example, our framework can accommodate a variety of kinds of tests (viral RNA, antibody, antigen, etc.) and sampling schemes (e.g., longitudinal or cross-sectional screening). Inference in this framework is complicated by the presence of tens or hundreds of thousands of discrete latent variables. To address this challenge, we propose an efficient stochastic variational inference method which relies on a novel gradient estimator for the variational objective. Experimental results for an example motivated by COVID-19 show that our method produces an accurate and well-calibrated posterior, while standard methods for estimating the reproduction number can fail badly.

Kai Wang, Bryan Wilder, Andrew Perrault et al.

Solving optimization problems with unknown parameters often requires learning a predictive model to predict the values of the unknown parameters and then solving the problem using these values. Recent work has shown that including the optimization problem as a layer in the model training pipeline results in predictions of the unobserved parameters that lead to higher decision quality. Unfortunately, this process comes at a large computational cost because the optimization problem must be solved and differentiated through in each training iteration; furthermore, it may also sometimes fail to improve solution quality due to non-smoothness issues that arise when training through a complex optimization layer. To address these shortcomings, we learn a low-dimensional surrogate model of a large optimization problem by representing the feasible space in terms of meta-variables, each of which is a linear combination of the original variables. By training a low-dimensional surrogate model end-to-end, and jointly with the predictive model, we achieve: i) a large reduction in training and inference time; and ii) improved performance by focusing attention on the more important variables in the optimization and learning in a smoother space. Empirically, we demonstrate these improvements on a non-convex adversary modeling task, a submodular recommendation task and a convex portfolio optimization task.

Thomas Davies, Jack Aspinall, Bryan Wilder et al.

Persistence diagrams concisely represent the topology of a point cloud whilst having strong theoretical guarantees, but the question of how to best integrate this information into machine learning workflows remains open. In this paper we extend the ubiquitous Fuzzy c-Means (FCM) clustering algorithm to the space of persistence diagrams, enabling unsupervised learning that automatically captures the topological structure of data without the topological prior knowledge or additional processing of persistence diagrams that many other techniques require. We give theoretical convergence guarantees that correspond to the Euclidean case, and empirically demonstrate the capability of our algorithm to capture topological information via the fuzzy RAND index. We end with experiments on two datasets that utilise both the topological and fuzzy nature of our algorithm: pre-trained model selection in machine learning and lattices structures from materials science. As pre-trained models can perform well on multiple tasks, selecting the best model is a naturally fuzzy problem; we show that fuzzy clustering persistence diagrams allows for model selection using the topology of decision boundaries. In materials science, we classify transformed lattice structure datasets for the first time, whilst the probabilistic membership values let us rank candidate lattices in a scenario where further investigation requires expensive laboratory time and expertise.

Bryan Wilder, Eric Horvitz, Ece Kamar

A rising vision for AI in the open world centers on the development of systems that can complement humans for perceptual, diagnostic, and reasoning tasks. To date, systems aimed at complementing the skills of people have employed models trained to be as accurate as possible in isolation. We demonstrate how an end-to-end learning strategy can be harnessed to optimize the combined performance of human-machine teams by considering the distinct abilities of people and machines. The goal is to focus machine learning on problem instances that are difficult for humans, while recognizing instances that are difficult for the machine and seeking human input on them. We demonstrate in two real-world domains (scientific discovery and medical diagnosis) that human-machine teams built via these methods outperform the individual performance of machines and people. We then analyze conditions under which this complementarity is strongest, and which training methods amplify it. Taken together, our work provides the first systematic investigation of how machine learning systems can be trained to complement human reasoning.

Aaron Ferber, Bryan Wilder, Bistra Dilkina et al.

Machine learning components commonly appear in larger decision-making pipelines; however, the model training process typically focuses only on a loss that measures accuracy between predicted values and ground truth values. Decision-focused learning explicitly integrates the downstream decision problem when training the predictive model, in order to optimize the quality of decisions induced by the predictions. It has been successfully applied to several limited combinatorial problem classes, such as those that can be expressed as linear programs (LP), and submodular optimization. However, these previous applications have uniformly focused on problems from specific classes with simple constraints. Here, we enable decision-focused learning for the broad class of problems that can be encoded as a Mixed Integer Linear Program (MIP), hence supporting arbitrary linear constraints over discrete and continuous variables. We show how to differentiate through a MIP by employing a cutting planes solution approach, which is an exact algorithm that iteratively adds constraints to a continuous relaxation of the problem until an integral solution is found. We evaluate our new end-to-end approach on several real world domains and show that it outperforms the standard two phase approaches that treat prediction and prescription separately, as well as a baseline approach of simply applying decision-focused learning to the LP relaxation of the MIP.

Harshavardhan Kamarthi, Priyesh Vijayan, Bryan Wilder et al.

A serious challenge when finding influential actors in real-world social networks is the lack of knowledge about the structure of the underlying network. Current state-of-the-art methods rely on hand-crafted sampling algorithms; these methods sample nodes and their neighbours in a carefully constructed order and choose opinion leaders from this discovered network to maximize influence spread in the (unknown) complete network. In this work, we propose a reinforcement learning framework for network discovery that automatically learns useful node and graph representations that encode important structural properties of the network. At training time, the method identifies portions of the network such that the nodes selected from this sampled subgraph can effectively influence nodes in the complete network. The realization of such transferable network structure based adaptable policies is attributed to the meticulous design of the framework that encodes relevant node and graph signatures driven by an appropriate reward scheme. We experiment with real-world social networks from four different domains and show that the policies learned by our RL agent provide a 10-36% improvement over the current state-of-the-art method.

Po-Wei Wang, Priya L. Donti, Bryan Wilder et al.

Integrating logical reasoning within deep learning architectures has been a major goal of modern AI systems. In this paper, we propose a new direction toward this goal by introducing a differentiable (smoothed) maximum satisfiability (MAXSAT) solver that can be integrated into the loop of larger deep learning systems. Our (approximate) solver is based upon a fast coordinate descent approach to solving the semidefinite program (SDP) associated with the MAXSAT problem. We show how to analytically differentiate through the solution to this SDP and efficiently solve the associated backward pass. We demonstrate that by integrating this solver into end-to-end learning systems, we can learn the logical structure of challenging problems in a minimally supervised fashion. In particular, we show that we can learn the parity function using single-bit supervision (a traditionally hard task for deep networks) and learn how to play 9x9 Sudoku solely from examples. We also solve a "visual Sudok" problem that maps images of Sudoku puzzles to their associated logical solutions by combining our MAXSAT solver with a traditional convolutional architecture. Our approach thus shows promise in integrating logical structures within deep learning.

Andrew Perrault, Bryan Wilder, Eric Ewing et al.

Stackelberg security games are a critical tool for maximizing the utility of limited defense resources to protect important targets from an intelligent adversary. Motivated by green security, where the defender may only observe an adversary's response to defense on a limited set of targets, we study the problem of learning a defense that generalizes well to a new set of targets with novel feature values and combinations. Traditionally, this problem has been addressed via a two-stage approach where an adversary model is trained to maximize predictive accuracy without considering the defender's optimization problem. We develop an end-to-end game-focused approach, where the adversary model is trained to maximize a surrogate for the defender's expected utility. We show both in theory and experimental results that our game-focused approach achieves higher defender expected utility than the two-stage alternative when there is limited data.