Enterprise AI Assistants are increasingly deployed in domains where accuracy is paramount, making each erroneous output a potentially significant incident. This paper presents a comprehensive framework for monitoring, benchmarking, and continuously improving such complex, multi-component systems under active development by multiple teams. Our approach encompasses three key elements: (1) a hierarchical ``severity'' framework for incident detection that identifies and categorizes errors while attributing component-specific error rates, facilitating targeted improvements; (2) a scalable and principled methodology for benchmark construction, evaluation, and deployment, designed to accommodate multiple development teams, mitigate overfitting risks, and assess the downstream impact of system modifications; and (3) a continual improvement strategy leveraging multidimensional evaluation, enabling the identification and implementation of diverse enhancement opportunities. By adopting this holistic framework, organizations can systematically enhance the reliability and performance of their AI Assistants, ensuring their efficacy in critical enterprise environments. We conclude by discussing how this multifaceted evaluation approach opens avenues for various classes of enhancements, paving the way for more robust and trustworthy AI systems.

While large language models (LLMs) are increasingly capable of handling longer contexts, recent work has demonstrated that they exhibit the "lost in the middle" phenomenon (Liu et al., 2024) of unevenly attending to different parts of the provided context.This hinders their ability to cover diverse source material in multidocument summarization, as noted in the DI-VERSESUMM benchmark (Huang et al., 2024).In this work, we contend that principled content selection is a simple way to increase source coverage on this task.As opposed to prompting an LLM to perform the summarization in a single step, we explicitly divide the task into three steps-(1) reducing document collections to atomic key points, (2) using determinantal point processes (DPP) to perform select key points that prioritize diverse content, and (3) rewriting to the final summary.By combining prompting steps, for extraction and rewriting, with principled techniques, for content selection, we consistently improve source coverage on the DIVERSESUMM benchmark across various LLMs.Finally, we also show that by incorporating relevance to a provided user intent into the DPP kernel, we can generate personalized summaries that cover relevant source information while retaining coverage.

Image Difference Captioning (IDC) aims to generate natural language descriptions that highlight subtle differences between two visually similar images.While recent advances leverage pre-trained vision-language models to align fine-grained visual differences with textual semantics, existing supervised approaches often overly focus on dataset-specific language patterns and fail to capture fine-grained and context-aware preferences on IDC, due to limited annotation diversity and a lack of semantically informative negative examples during training, To address these limitations, we propose an adversarial direct preference optimization (ADPO) framework for IDC, which formulates IDC as a preference optimization problem under the Bradley-Terry-Luce model, directly aligning the captioning policy with pairwise difference preferences via Direct Preference Optimization (DPO).To model more accurate and diverse IDC preferences, we introduce an adversarially trained hard negative retriever that selects counterfactual captions, This results in a minimax optimization problem, which we solve via policy-gradient reinforcement learning, enabling the policy and retriever to improve jointly.By dynamically generating semantically challenging negatives, our method reduces reliance on dataset-specific patterns.Experiments on benchmark IDC datasets show that our approach outperforms existing baselines, especially in generating fine-grained and accurate difference descriptions.

Large Language Models (LLMs) enable a new form of digital experimentation where treatments combine human and model-generated content in increasingly sophisticated ways. The main methodological challenge in this setting is representing these high-dimensional treatments without losing their semantic meaning or rendering analysis intractable. Here we address this problem by focusing on learning low-dimensional representations that capture the underlying structure of such treatments. These representations enable downstream applications such as guiding generative models to produce meaningful treatment variants and facilitating adaptive assignment in online experiments. We propose double kernel representation learning, which models the causal effect through the inner product of kernel-based representations of treatments and user covariates. We develop an alternating-minimization algorithm that learns these representations efficiently from data and provide convergence guarantees under a low-rank factor model. As an application of this framework, we introduce an adaptive design strategy for online experimentation and demonstrate the method's effectiveness through numerical experiments.

Human annotations play a crucial role in evaluating the performance of GenAI models. Two common challenges in practice, however, are missing annotations (the response variable of interest) and cluster dependence among human-AI interactions (e.g., questions asked by the same user may be highly correlated). Reliable inference must address both issues to achieve unbiased estimation and appropriately quantify uncertainty when estimating average scores from human annotations. In this paper, we analyze the doubly robust estimator, a widely used method in missing data analysis and causal inference, applied to this setting and establish novel theoretical properties under cluster dependence. We further illustrate our findings through simulations and a real-world conversation quality dataset. Our theoretical and empirical results underscore the importance of incorporating cluster dependence in missing response problems to perform valid statistical inference.

Estimating causal effects from real-world relational data can be challenging when the underlying causal model and potential confounders are unknown. While several causal discovery algorithms exist for learning causal models with latent confounders from data, they assume that the data is independent and identically distributed (i.i.d.) and are not well-suited for learning from relational data. Similarly, existing relational causal discovery algorithms assume causal sufficiency, which is unrealistic for many real-world datasets. To address this gap, we propose RelFCI, a sound and complete causal discovery algorithm for relational data with latent confounders. Our work builds upon the Fast Causal Inference (FCI) and Relational Causal Discovery (RCD) algorithms and it defines new graphical models, necessary to support causal discovery in relational domains. We also establish soundness and completeness guarantees for relational d-separation with latent confounders. We present experimental results demonstrating the effectiveness of RelFCI in identifying the correct causal structure in relational causal models with latent confounders.

Recommendation strategies are typically evaluated by using previously logged data, employing off-policy evaluation methods to estimate their expected performance. However, for strategies that present users with slates of multiple items, the resulting combinatorial action space renders many of these methods impractical. Prior work has developed estimators that leverage the structure in slates to estimate the expected off-policy performance, but the estimation of the entire performance distribution remains elusive. Estimating the complete distribution allows for a more comprehensive evaluation of recommendation strategies, particularly along the axes of risk and fairness that employ metrics computable from the distribution. In this paper, we propose an estimator for the complete off-policy performance distribution for slates and establish conditions under which the estimator is unbiased and consistent. This builds upon prior work on off-policy evaluation for slates and off-policy distribution estimation in reinforcement learning. We validate the efficacy of our method empirically on synthetic data as well as on a slate recommendation simulator constructed from real-world data (MovieLens-20M). Our results show a significant reduction in estimation variance and improved sample efficiency over prior work across a range of slate structures.

Most real-world networks are noisy and incomplete samples from an unknown target distribution. Refining them by correcting corruptions or inferring unobserved regions typically improves downstream performance. Inspired by the impressive generative capabilities that have been used to correct corruptions in images, and the similarities between "in-painting" and filling in missing nodes and edges conditioned on the observed graph, we propose a novel graph generative framework, SGDM, which is based on subgraph diffusion. Our framework not only improves the scalability and fidelity of graph diffusion models, but also leverages the reverse process to perform novel, conditional generation tasks. In particular, through extensive empirical analysis and a set of novel metrics, we demonstrate that our proposed model effectively supports the following refinement tasks for partially observable networks: (T1) denoising extraneous subgraphs, (T2) expanding existing subgraphs and (T3) performing ``style" transfer by regenerating a particular subgraph to match the characteristics of a different node or subgraph.

In many real-world causal inference applications, the primary outcomes (labels) are often partially missing, especially if they are expensive or difficult to collect. If the missingness depends on covariates (i.e., missingness is not completely at random), analyses based on fully observed samples alone may be biased. Incorporating surrogates, which are fully observed post-treatment variables related to the primary outcome, can improve estimation in this case. In this paper, we study the role of surrogates in estimating continuous treatment effects and propose a doubly robust method to efficiently incorporate surrogates in the analysis, which uses both labeled and unlabeled data and does not suffer from the above selection bias problem. Importantly, we establish the asymptotic normality of the proposed estimator and show possible improvements on the variance compared with methods that solely use labeled data. Extensive simulations show our methods enjoy appealing empirical performance.

In real-world phenomena which involve mutual influence or causal effects between interconnected units, equilibrium states are typically represented with cycles in graphical models. An expressive class of graphical models, relational causal models, can represent and reason about complex dynamic systems exhibiting such cycles or feedback loops. Existing cyclic causal discovery algorithms for learning causal models from observational data assume that the data instances are independent and identically distributed which makes them unsuitable for relational causal models. At the same time, causal discovery algorithms for relational causal models assume acyclicity. In this work, we examine the necessary and sufficient conditions under which a constraint-based relational causal discovery algorithm is sound and complete for cyclic relational causal models. We introduce relational acyclification, an operation specifically designed for relational models that enables reasoning about the identifiability of cyclic relational causal models. We show that under the assumptions of relational acyclification and sigma-faithfulness, the relational causal discovery algorithm RCD is sound and complete for cyclic relational models. We present experimental results to support our claim.

Anomaly detection for time series data is critical for monitoring the status of ever-growing data sources, e.g., health metrics of data servers. An important element of anomaly detection is time series forecasting, which is often the most time-consuming stage in the system. Existing approaches based on batch processing usually take several seconds or more to process one forecasting task, thus not applicable for large-scale real-time applications. In this paper, we present an online-learning based forecasting algorithm to address the computation bottleneck. It readily handles missing observations and has a constant time complexity, independent of the number of past observations. Our experiments show that it achieves similar level of accuracy as existing approaches while only takes a fraction of computational resources. The proposed algorithm can help us easily monitor tens of thousands of data sources simultaneously with only a small hardware cost.

The design and analysis of randomized experiments is fundamental to many areas, from the physical and social sciences to industrial settings. Regression adjustment is a popular technique to reduce the variance of estimates obtained from experiments, by utilizing information contained in auxiliary covariates. While there is a large literature within the statistics community studying various approaches to regression adjustment and their asymptotic properties, little focus has been given to approaches in the finite population setting with non-asymptotic accuracy bounds. Further, prior work typically assumes that an entire population is exposed to an experiment, whereas practitioners often seek to minimize the number of subjects exposed to an experiment, for ethical and pragmatic reasons. In this work, we study the problems of estimating the sample mean, individual treatment effects, and average treatment effect with regression adjustment. We propose approaches that use techniques from randomized numerical linear algebra to sample a subset of the population on which to perform an experiment. We give non-asymptotic accuracy bounds for our methods and demonstrate that they compare favorably with prior approaches.

Several cloud-based applications, such as cloud gaming, rent servers to execute jobs which arrive in an online fashion. Each job has a resource demand, such as GPU requirement, and must be dispatched to a cloud server which has enough resources to execute the job, which departs after its completion. Under the "pay-as-you-go'' billing model, the server rental cost is proportional to the total time that servers are actively running jobs. The problem of efficiently allocating a sequence of online jobs to servers without exceeding the resource capacity of any server while minimizing total server usage time can be modelled as a variant of the dynamic bin packing problem (DBP), called MinUsageTime DBP [10].

Online reinforcement learning (RL) algorithms are often difficult to deploy in complex human-facing applications as they may learn slowly and have poor early performance. To address this, we introduce a practical algorithm for incorporating human insight to speed learning. Our algorithm, Constraint Sampling Reinforcement Learning (CSRL), incorporates prior domain knowledge as constraints/restrictions on the RL policy. It takes in multiple potential policy constraints to maintain robustness to misspecification of individual constraints while leveraging helpful ones to learn quickly. Given a base RL learning algorithm (ex. UCRL, DQN, Rainbow) we propose an upper confidence with elimination scheme that leverages the relationship between the constraints, and their observed performance, to adaptively switch among them. We instantiate our algorithm with DQN-type algorithms and UCRL as base algorithms, and evaluate our algorithm in four environments, including three simulators based on real data: recommendations, educational activity sequencing, and HIV treatment sequencing. In all cases, CSRL learns a good policy faster than baselines.

We consider the experimental design problem in an online environment, an important practical task for reducing the variance of estimates in randomized experiments which allows for greater precision, and in turn, improved decision making. In this work, we present algorithms that build on recent advances in online discrepancy minimization which accommodate both arbitrary treatment probabilities and multiple treatments. The proposed algorithms are computational efficient, minimize covariate imbalance, and include randomization which enables robustness to misspecification. We provide worst case bounds on the expected mean squared error of the causal estimate and show that the proposed estimator is no worse than an implicit ridge regression, which are within a logarithmic factor of the best known results for offline experimental design. We conclude with a detailed simulation study showing favorable results relative to complete randomization as well as to offline methods for experimental design with time complexities exceeding our algorithm, which has a linear dependence on the number of observations, by polynomial factors.

Treatment effect estimation is a fundamental problem in causal inference. We focus on designing efficient randomized controlled trials, to accurately estimate the effect of some treatment on a population of $n$ individuals. In particular, we study \textit{sample-constrained treatment effect estimation}, where we must select a subset of $s \ll n$ individuals from the population to experiment on. This subset must be further partitioned into treatment and control groups. Algorithms for partitioning the entire population into treatment and control groups, or for choosing a single representative subset, have been well-studied. The key challenge in our setting is jointly choosing a representative subset and a partition for that set. We focus on both individual and average treatment effect estimation, under a linear effects model. We give provably efficient experimental designs and corresponding estimators, by identifying connections to discrepancy minimization and leverage-score-based sampling used in randomized numerical linear algebra. Our theoretical results obtain a smooth transition to known guarantees when $s$ equals the population size. We also empirically demonstrate the performance of our algorithms.

Search engines and recommendation systems attempt to continually improve the quality of the experience they afford to their users. Refining the ranker that produces the lists displayed in response to user requests is an important component of this process. A common practice is for the service providers to make changes (e.g. new ranking features, different ranking models) and A/B test them on a fraction of their users to establish the value of the change. An alternative approach estimates the effectiveness of the proposed changes offline, utilising previously collected clickthrough data on the old ranker to posit what the user behaviour on ranked lists produced by the new ranker would have been. A majority of offline evaluation approaches invoke the well studied inverse propensity weighting to adjust for biases inherent in logged data. In this paper, we propose the use of parametric estimates for these propensities. Specifically, by leveraging well known learning-to-rank methods as subroutines, we show how accurate offline evaluation can be achieved when the new rankings to be evaluated differ from the logged ones.

Independence testing plays a central role in statistical and causal inference from observational data. Standard independence tests assume that the data samples are independent and identically distributed (i.i.d.) but that assumption is violated in many real-world datasets and applications centered on relational systems. This work examines the problem of estimating independence in data drawn from relational systems by defining sufficient representations for the sets of observations influencing individual instances. Specifically, we define marginal and conditional independence tests for relational data by considering the kernel mean embedding as a flexible aggregation function for relational variables. We propose a consistent, non-parametric, scalable kernel test to operationalize the relational independence test for non-i.i.d. observational data under a set of structural assumptions. We empirically evaluate our proposed method on a variety of synthetic and semi-synthetic networks and demonstrate its effectiveness compared to state-of-the-art kernel-based independence tests.

In our society, generations of systemic biases have led to some professions being more common among certain genders and races. This bias is also reflected in image search on stock image repositories and search engines, e.g., a query like “male Asian administrative assistant” may produce limited results. The pursuit of a utopian world demands providing content users with an opportunity to present any profession with diverse racial and gender characteristics. The limited choice of existing content for certain combinations of profession, race, and gender presents a challenge to content providers. Current research dealing with bias in search mostly focuses on re-ranking algorithms. However, these methods cannot create new content or change the overall distribution of protected attributes in photos. To remedy these problems, we propose a new task of high-fidelity image generation conditioning on multiple attributes from imbalanced datasets. Our proposed task poses new sets of challenges for the state-of-the-art Generative Adversarial Networks (GANs). In this paper, we also propose a new training framework to better address the challenges. We evaluate our framework rigorously on a real-world dataset and perform user studies that show our model is preferable to the alternatives.

In this work, we address the problem of balanced treatment assignment for experiments by considering an interpretation of the problem as optimization of a two-sample test between test and control units. Using this lens we provide an assignment algorithm that is optimal with respect to the minimum spanning tree test of Friedman and Rafsky [1979]. This assignment to treatment groups may be performed exactly in polynomial time and allows for the design of experiments explicitly targeting the individual treatment effect. We provide a probabilistic interpretation of this process in terms of the most probable element of designs drawn from a determinantal point process. We provide a novel formulation of estimation as transductive inference and show how the tree structures used in design can also be used in an adjustment estimator. We conclude with a simulation study demonstrating the improved efficacy of our method.

We consider the problem of designing a randomized experiment on a source population to estimate the Average Treatment Effect (ATE) on a target population. We propose a novel approach which explicitly considers the target when designing the experiment on the source. Under the covariate shift assumption, we design an unbiased importance-weighted estimator for the target population’s ATE. To reduce the variance of our estimator, we design a covariate balance condition (Target Balance) between the treatment and control groups based on the target population. We show that Target Balance achieves a higher variance reduction asymptotically than methods that do not consider the target population during the design phase. Our experiments illustrate that Target Balance reduces the variance even for small sample sizes.

A commonly applied approach for estimating causal effects from observational data is to apply weights which render treatments independent of observed pre-treatment covariates. Recently emphasis has been placed on deriving balancing weights which explicitly target this independence condition. In this work we introduce permutation weighting, a method for estimating balancing weights using a standard binary classifier (regardless of cardinality of treatment). A large class of probabilistic classifiers may be used in this method; the choice of loss for the classifier implies the particular definition of balance. We bound bias and variance in terms of the excess risk of the classifier, show that these disappear asymptotically, and demonstrate that our classification problem directly minimizes imbalance. Additionally, hyper-parameter tuning and model selection can be performed with standard cross-validation methods. Empirical evaluations indicate that permutation weighting provides favorable performance in comparison to existing methods.

We present BOhance, an efficient solution for optimizing digital content like images. Our approach enhances the standard and widely-used method for optimizing content, A/B testing, by using Bayesian Optimization. Our work effectively extends A/B testing in the continuous domain where A/B testing cannot efficiently test infinitely many variants. We test our approach on an image enhancement task where we use iterative human feedback on different variants of an image to arrive at the optimal variant. BOhance auto-generates candidate content variants to be tested based on the human feedback on prior variants. We demonstrate with user-studies conducted on Amazon Mechanical Turk that BOhance can be both time and cost-efficient; and a superior alternative to existing solutions. Furthermore, we conduct a Visual Turing Test to obtain human impressions on the optimum variants generated by BOhance. Our experiments show that given a human-enhanced image and an image generated by BOhance, 53% users think that the BOhance image was generated by a human expert.

This tutorial presents state-of-the-art research on causal inference from network data in the presence of interference. We start by motivating research in this area with real-world applications, such as measuring influence in social networks and market experimentation. We discuss the challenges of applying existing causal inference techniques designed for independent and identically distributed (i.i.d.) data to relational data, some of the solutions that currently exist and the gaps and opportunities for future research. We present existing network experiment designs for measuring different possible effects of interest. Then we focus on causal inference from observational data, its representation, identification, and estimation. We conclude with research on causal discovery in networks.

In this article, we introduce a generalization of graphlets to heterogeneous networks called typed graphlets . Informally, typed graphlets are small typed induced subgraphs. Typed graphlets generalize graphlets to rich heterogeneous networks as they explicitly capture the higher-order typed connectivity patterns in such networks. To address this problem, we describe a general framework for counting the occurrences of such typed graphlets. The proposed algorithms leverage a number of combinatorial relationships for different typed graphlets. For each edge, we count a few typed graphlets, and with these counts along with the combinatorial relationships, we obtain the exact counts of the other typed graphlets in o (1) constant time. Notably, the worst-case time complexity of the proposed approach matches the time complexity of the best known untyped algorithm. In addition, the approach lends itself to an efficient lock-free and asynchronous parallel implementation. While there are no existing methods for typed graphlets, there has been some work that focused on computing a different and much simpler notion called colored graphlet. The experiments confirm that our proposed approach is orders of magnitude faster and more space-efficient than methods for computing the simpler notion of colored graphlet. Unlike these methods that take hours on small networks, the proposed approach takes only seconds on large networks with millions of edges. Notably, since typed graphlet is more general than colored graphlet (and untyped graphlets), the counts of various typed graphlets can be combined to obtain the counts of the much simpler notion of colored graphlets. The proposed methods give rise to new opportunities and applications for typed graphlets.

In many applied fields, researchers are ofteninterested in tailoring treatments to unit-levelcharacteristics in order to optimize an outcomeof interest. Methods for identifying andestimating treatment policies are the subjectof the dynamic treatment regime literature. Separately, in many settings the assumptionthat data are independent and identically distributeddoes not hold due to inter-subjectdependence. The phenomenon where a subject’s outcome is dependent on his neighbor’s exposure is known as interference. These areasintersect in myriad real-world settings. Inthis paper we consider the problem of identifyingoptimal treatment policies in the presenceof interference. Using a general representationof interference, via Lauritzen-Wermuth-Freydenburg chain graphs (Lauritzen andRichardson, 2002), we formalize a variety ofpolicy interventions under interference andextend existing identification theory (Tian,2008; Sherman and Shpitser, 2018). Finally, we illustrate the efficacy of policy maximization under interference in a simulation study.

Estimation of importance sampling weights for off-policy evaluation of contextual bandits often results in imbalance—a mismatch between the desired and the actual distribution of state-action pairs after weighting. In this work we present balanced off-policy evaluation (B-OPE), a generic method for estimating weights which minimize this imbalance. Estimation of these weights reduces to a binary classification problem regardless of action type. We show that minimizing the risk of the classifier implies minimization of imbalance to the desired counterfactual distribution. In turn, this is tied to the error of the off-policy estimate, allowing for easy tuning of hyperparameters. We provide experimental evidence that B-OPE improves weighting-based approaches for offline policy evaluation in both discrete and continuous action spaces.

Aggregate advertising-presenting a single ad to large groups of individuals through traditional media such as television and print-presents a unique challenge to measuring efficacy because treatment and outcome are observed from two disparate sources (interaction and revenue realization). In this work, we propose a Bayesian model to estimate the impact of an ad on observable web metrics that are readily available in many modern analytics suites. The proposed model controls for three sources of possible confounding: the time, geography, and content of the advertisement. The proposed model is easily applicable to a wide variety of problems and readily generates error bounds for the estimates. We evaluate our approach on a real dataset for a set of TV ads for an advertiser.

We present Relational Covariate Adjustment (RCA), a general method for estimating causal effects in relational data. Relational Covariate Adjustment is implemented through two high-level operations: identification of an adjustment set and relational regression adjustment. The former is achieved through an extension of Pearl's back-door criterion to relational domains. We demonstrate how this extended definition can be used to estimate causal effects in the presence of network interference and confounding. RCA is agnostic to functional form, and it can easily model both discrete and continuous treatments as well as estimate the effects of a wider array of network interventions than existing experimental approaches. We show that RCA can yield robust estimates of causal effects using common regression models without extensive parameter tuning. Through a series of simulation experiments on a variety of synthetic and real-world network structures, we show that causal effects estimated on observational data with RCA are nearly as accurate as those estimated from well-designed network experiments

Our research into automatic recording of the complete classroom experience has led to the development of many software systems, one of which captures an image stream of all content presented on a computer. We have just completed a first deployment of this computer capture system in which 3 separate courses were recorded for an entire semester with the content being presented to students within 24 hours of the class meeting time. This system has been envisioned as a component of a complete lecture capture system but a component with real value even when used as a stand alone. In this paper we discuss student feedback to this computer capture system, revision of system functionality, and thoughts on the usefulness of capturing computer content in computer science courses in general.

This paper describes our experiences with the first partial deployment of Presentations Automatically Organized from Lectures (PAOL), a lecture recording system developed and tested at the University of Massachusetts Amherst. PAOL automatically records all information presented during lectures using any combination of computer, whiteboard, and overhead presentation and compiles the captured lectures into indexed presentations. We discuss lessons learned from this deployment that have application in lecture recording specifically and classroom technology in general. We also discuss our initial evaluation of created presentations as determined by a small focus group study.