Consider the problem of finding the best matching in a weighted graph where we only have access to predictions of the actual stochastic weights, based on an underlying context. If the predictor is the Bayes optimal one, then computing the best matching based on the predicted weights is optimal. However, in practice, this perfect information scenario is not realistic. Given an imperfect predictor, a suboptimal decision rule may compensate for the induced error and thus outperform the standard optimal rule. In this paper, we propose multicalibration as a way to address this problem. This fairness notion requires a predictor to be unbiased on each element of a family of protected sets of contexts. Given a class of matching algorithms 𝒞 and any predictor γ of the edge-weights, we show how to construct a specific multicalibrated predictor γ̂, with the following property. Picking the best matching based on the output of γ̂ is competitive with the best decision rule in 𝒞 applied onto the original predictor γ. We complement this result by providing sample complexity bounds, and by performing numerical experiments.

Estimating the prevalence of a category in a population using imperfect measurement devices (diagnostic tests, classifiers, or large language models) is fundamental to science, public health, and online trust and safety. Standard approaches correct for known device error rates but assume these rates remain stable across populations. We show this assumption fails under covariate shift and that multicalibration, which enforces calibration conditional on the input features rather than just on average, is sufficient for unbiased prevalence estimation under such shift. Standard calibration and quantification methods fail to provide this guarantee. Our work connects recent theoretical work on fairness to a longstanding measurement problem spanning nearly all academic disciplines. A simulation confirms that standard methods exhibit bias growing with shift magnitude, while a multicalibrated estimator maintains near-zero bias. While we focus the discussion mostly on LLMs, our theoretical results apply to any classification model. Two empirical applications – estimating employment prevalence across U.S. states using the American Community Survey, and classifying political texts across four countries using an LLM – demonstrate that multicalibration substantially reduces bias in practice, while highlighting that calibration data should cover the key feature dimensions along which target populations may differ.

We propose MCGrad, a novel and scalable multicalibration algorithm. Multicalibration - calibration in subgroups of the data - is an important property for the performance of machine learning-based systems. Existing multicalibration methods have thus far received limited traction in industry. We argue that this is because existing methods (1) require such subgroups to be manually specified, which ML practitioners often struggle with, (2) are not scalable, or (3) may harm other notions of model performance such as log loss and Area Under the Precision-Recall Curve (PRAUC). MCGrad does not require explicit specification of protected groups, is scalable, and often improves other ML evaluation metrics instead of harming them. MCGrad has been in production at Meta, and is now part of hundreds of production models. We present results from these deployments as well as results on public datasets. We provide an open source implementation of MCGrad at https://github.com/facebookincubator/MCGrad.

Multicalibration gradient boosting has recently emerged as a scalable method that empirically produces approximately multicalibrated predictors and has been deployed at web scale. Despite this empirical success, its convergence properties are not well understood. In this paper, we bridge the gap by providing convergence guarantees for multicalibration gradient boosting in regression with squared-error loss. We show that the magnitude of successive prediction updates decays at O(1/√(T)), which implies the same convergence rate bound for the multicalibration error over rounds. Under additional smoothness assumptions on the weak learners, this rate improves to linear convergence. We further analyze adaptive variants, showing local quadratic convergence of the training loss, and we study rescaling schemes that preserve convergence. Experiments on real-world datasets support our theory and clarify the regimes in which the method achieves fast convergence and strong multicalibration.

This paper presents a manifesto that articulates the conceptual foundations of Agentic Business Process Management (APM), an extension of Business Process Management (BPM) for governing autonomous agents executing processes in organizations. From a management perspective, APM represents a paradigm shift from the traditional view on business processes. This shift is driven by the realization of process awareness by agent-oriented abstractions: software and human agents act as primary functional entities that perceive, reason, and act within explicit process frames. Thus, APM moves away from automation-oriented BPM towards systems in which autonomy is constrained, aligned, and made operational through process aware agents.We introduce the core abstractions and architectural elements required to realize APM systems and elaborate on four key capabilities that agents in APM systems must support: framed autonomy, explainability, conversational actionability, and self-modification. These capabilities jointly ensure that agents’ goals are aligned with organizational goals and that agents behave in a framed yet proactive manner in pursuing those goals. We discuss the extent to which the capabilities can be realized and identify research challenges whose resolution requires further advances in BPM, AI, and multi-agent systems. The manifesto thus serves as a roadmap for bridging these communities and for guiding the development of APM systems in practice.

A suitable scalar metric can help measure multi-calibration, defined as follows. When the expected values of observed responses are equal to corresponding predicted probabilities, the probabilistic predictions are known as "perfectly calibrated." When the predicted probabilities are perfectly calibrated simultaneously across several subpopulations, the probabilistic predictions are known as "perfectly multi-calibrated." In practice, predicted probabilities are seldom perfectly multi-calibrated, so a statistic measuring the distance from perfect multi-calibration is informative. A recently proposed metric for calibration, based on the classical Kuiper statistic, is a natural basis for a new metric of multi-calibration and avoids well-known problems of metrics based on binning or kernel density estimation. The newly proposed metric weights the contributions of different subpopulations in proportion to their signal-to-noise ratios; data analyses' ablations demonstrate that the metric becomes noisy when omitting the signal-to-noise ratios from the metric. Numerical examples on benchmark data sets illustrate the new metric.

Active learning can improve the efficiency of training prediction models by identifying the most informative new labels to acquire. However, non-response to label requests can impact active learning's effectiveness in real-world contexts. We conceptualise this degradation by considering the type of non-response present in the data, demonstrating that biased non-response is particularly detrimental to model performance. We argue that biased non-response is likely in contexts where the labelling process, by nature, relies on user interactions. To mitigate the impact of biased non-response, we propose a cost-based correction to the sampling strategy--the Upper Confidence Bound of the Expected Utility (UCB-EU)--that can, plausibly, be applied to any active learning algorithm. Through experiments, we demonstrate that our method successfully reduces the harm from labelling non-response in many settings. However, we also characterise settings where the non-response bias in the annotations remains detrimental under UCB-EU for specific sampling methods and data generating processes. Finally, we evaluate our method on a real-world dataset from an e-commerce platform. We show that UCB-EU yields substantial performance improvements to conversion models that are trained on clicked impressions. Most generally, this research serves to both better conceptualise the interplay between types of non-response and model improvements via active learning, and to provide a practical, easy-to-implement correction that mitigates model degradation.

We investigate the convergence rates and data sample sizes required for training a machine learning model using a stochastic gradient descent (SGD) algorithm, where data points are sampled based on either their loss value or uncertainty value. These training methods are particularly relevant for active learning and data subset selection problems. For SGD with a constant step size update, we present convergence results for linear classifiers and linearly separable datasets using squared hinge loss and similar training loss functions. Additionally, we extend our analysis to more general classifiers and datasets, considering a wide range of loss-based sampling strategies and smooth convex training loss functions. We propose a novel algorithm called Adaptive-Weight Sampling (AWS) that utilizes SGD with an adaptive step size that achieves stochastic Polyak's step size in expectation. We establish convergence rate results for AWS for smooth convex training loss functions. Our numerical experiments demonstrate the efficiency of AWS on various datasets by using either exact or estimated loss values.

In online learning, a decision maker repeatedly selects one of a set of actions, with the goal of minimizing the overall loss incurred. Following the recent line of research on algorithms endowed with additional predictive features, we revisit this problem by allowing the decision maker to acquire additional information on the actions to be selected. In particular, we study the power of \emph{best-action queries}, which reveal beforehand the identity of the best action at a given time step. In practice, predictive features may be expensive, so we allow the decision maker to issue at most $k$ such queries. We establish tight bounds on the performance any algorithm can achieve when given access to $k$ best-action queries for different types of feedback models. In particular, we prove that in the full feedback model, $k$ queries are enough to achieve an optimal regret of $\Theta(\min\{\sqrt T, \frac{T}{k}\})$. This finding highlights the significant multiplicative advantage in the regret rate achievable with even a modest (sublinear) number $k \in \Omega(\sqrt{T})$ of queries. Additionally, we study the challenging setting in which the only available feedback is obtained during the time steps corresponding to the $k$ best-action queries. There, we provide a tight regret rate of $\Theta(\min\{\frac{T}{\sqrt k},\frac{T^2}{k^2}\})$, which improves over the standard $\Theta(\frac{T}{\sqrt k})$ regret rate for label efficient prediction for $k \in \Omega(T^{2/3})$.

This paper proposes a new metric to measure the calibration error of probabilistic binary classifiers, called test-based calibration error (TCE). TCE incorporates a novel loss function based on a statistical test to examine the extent to which model predictions differ from probabilities estimated from data. It offers (i) a clear interpretation, (ii) a consistent scale that is unaffected by class imbalance, and (iii) an enhanced visual representation with repect to the standard reliability diagram. In addition, we introduce an optimality criterion for the binning procedure of calibration error metrics based on a minimal estimation error of the empirical probabilities. We provide a novel computational algorithm for optimal bins under bin-size constraints. We demonstrate properties of TCE through a range of experiments, including multiple real-world imbalanced datasets and ImageNet 1000.

Researchers in explainable artificial intelligence have developed numerous methods for helping users understand the predictions of complex supervised learning models. By contrast, explaining the $\textit{uncertainty}$ of model outputs has received relatively little attention. We adapt the popular Shapley value framework to explain various types of predictive uncertainty, quantifying each feature's contribution to the conditional entropy of individual model outputs. We consider games with modified characteristic functions and find deep connections between the resulting Shapley values and fundamental quantities from information theory and conditional independence testing. We outline inference procedures for finite sample error rate control with provable guarantees, and implement efficient algorithms that perform well in a range of experiments on real and simulated data. Our method has applications to covariate shift detection, active learning, feature selection, and active feature-value acquisition.

Process mining aims to provide analysts with insights, such that business processes supported by information systems can be improved. Traditionally, insights from process mining projects and techniques have been associational rather than causal, thus only describing the current state of the process, without predictive capabilities over effects of hypothetical process changes, which inherently limits business process optimisation efforts. In this paper, we introduce causal analysis for control-flow decisions taken during the execution of process models: using an event log and the structure of a process model, we (i) extract the set of decision points in the process, (ii) apply a causal discovery approach to obtain a collection of causal graphs that are consistent with the observations in the event log, (iii) extract ordered pairs of decision points between which a causal connection can be ruled out based on the temporal ordering that is implied by the process model specification, and use these to narrow down the set of possible causal graphs. This technique addresses the problem of mining dependencies, which has long been a challenge in the process discovery field. The technique has been implemented in the Visual Miner as part of the ProM framework. We illustrate the technique using examples and demonstrate its applicability on real-life logs.

Predictive process monitoring is a family of techniques to analyze events produced during the execution of a business process in order to predict the future state or the final outcome of running process instances. Existing techniques in this field are able to predict, at each step of a process instance, the likelihood that it will lead to an undesired outcome. These techniques, however, focus on generating predictions and do not prescribe when and how process workers should intervene to decrease the cost of undesired outcomes. This paper proposes a framework for prescriptive process monitoring, which extends predictive monitoring with the ability to generate alarms that trigger interventions to prevent an undesired outcome or mitigate its effect. The framework incorporates a parameterized cost model to assess the cost–benefit trade-off of generating alarms. We show how to optimize the generation of alarms given an event log of past process executions and a set of cost model parameters. The proposed approaches are empirically evaluated using a range of real-life event logs. The experimental results show that the net cost of undesired outcomes can be minimized by changing the threshold for generating alarms, as the process instance progresses. Moreover, introducing delays for triggering alarms, instead of triggering them as soon as the probability of an undesired outcome exceeds a threshold, leads to lower net costs.

Fraud detection and prevention play an important part in ensuring the sustained operation of any e-commerce business. Machine learning (ML) often plays an important role in these anti-fraud operations, but the organizational context in which these ML models operate cannot be ignored. In this paper, we take an organization-centric view on the topic of fraud detection by formulating an operational model of the anti-fraud departments in e-commerce organizations. We derive 6 research topics and 12 practical challenges for fraud detection from this operational model. We summarize the state of the literature for each research topic, discuss potential solutions to the practical challenges, and identify 22 open research challenges.

Data of sequential nature arise in many application domains in the form of, e.g., textual data, DNA sequences, and software execution traces. Different research disciplines have developed methods to learn sequence models from such datasets: (i) In the machine learning field methods such as (hidden) Markov models and recurrent neural networks have been developed and successfully applied to a wide range of tasks, (ii) in process mining process discovery methods aim to generate human-interpretable descriptive models, and (iii) in the grammar inference field the focus is on finding descriptive models in the form of formal grammars. Despite their different focuses, these fields share a common goal: learning a model that accurately captures the sequential behavior in the underlying data. Those sequence models are generative, i.e., they are able to predict what elements are likely to occur after a given incomplete sequence. So far, these fields have developed mainly in isolation from each other and no comparison exists. This paper presents an interdisciplinary experimental evaluation that compares sequence modeling methods on the task of next-element prediction on four real-life sequence datasets. The results indicate that machine learning methods, which generally do not aim at model interpretability, tend to outperform methods from the process mining and grammar inference fields in terms of accuracy.

This thesis focuses on process mining on event data where such a normative specification is absent and, as a result, the event data is less structured. The thesis puts special emphasis on one application domain that fits this description: the analysis of smart home data where sequences of daily activities are recorded. In this thesis we propose a set of techniques to analyze such data, which can be grouped into two categories of techniques. The first category of methods focuses on preprocessing event logs in order to enable process discovery techniques to extract insights into unstructured event data. In this category we have developed the following techniques: - An unsupervised approach to refine event labels based on the time at which the event took place, allowing for example to distinguish recorded eating events into breakfast, lunch, and dinner. - An approach to detect and filter from event logs so-called chaotic activities, which are activities that cause process discovery methods to overgeneralize. - A supervised approach to abstract low-level events into more high-level events, where we show that there exist situations where process discovery approaches overgeneralize on the low-level event data but are able to find precise models on the high-level event data. The second category focuses on mining local process models, i.e., collections of process model patterns that each describe some frequent pattern, in contrast to the single global process model that is obtained with existing process discovery techniques. Several techniques are introduced in the area of local process model mining, including a basic method, fast but approximate heuristic methods, and constraint-based techniques.

Process mining sheds new light on the relationship between process models and real-life processes. Process discovery can be used to learn process models from event logs. Conformance checking is concerned with quantifying the quality of a business process model in relation to event data that was logged during the execution of the business process. There exist different categories of conformance measures. Recall, also called fitness, is concerned with quantifying how much of the behavior that was observed in the event log fits the process model. Precision is concerned with quantifying how much behavior a process model allows for that was never observed in the event log. Generalization is concerned with quantifying how well a process model generalizes to behavior that is possible in the business process but was never observed in the event log. Many recall, precision, and generalization measures have been developed throughout the years, but they are often defined in an ad-hoc manner without formally defining the desired properties up front. To address these problems, we formulate 21 conformance propositions and we use these propositions to evaluate current and existing conformance measures. The goal is to trigger a discussion by clearly formulating the challenges and requirements (rather than proposing new measures). Additionally, this paper serves as an overview of the conformance checking measures that are available in the process mining area.

Process mining is a research field focused on the analysis of event data with the aim of extracting insights related to dynamic behavior. Applying process mining techniques on data from smart home environments has the potential to provide valuable insights in (un)healthy habits and to contribute to ambient assisted living solutions. Finding the right event labels to enable the application of process mining techniques is however far from trivial, as simply using the triggering sensor as the label for sensor events results in uninformative models that allow for too much behavior (overgeneralizing). Refinements of sensor level event labels suggested by domain experts have been shown to enable discovery of more precise and insightful process models. However, there exists no automated approach to generate refinements of event labels in the context of process mining. In this paper we propose a framework for the automated generation of label refinements based on the time attribute of events, allowing us to distinguish behaviourally different instances of the same event type based on their time attribute. We show on a case study with real life smart home event data that using automatically generated refined labels in process discovery, we can find more specific, and therefore more insightful, process models. We observe that one label refinement could have an effect on the usefulness of other label refinements when used together. Therefore, we explore four strategies to generate useful combinations of multiple label refinements and evaluate those on three real life smart home event logs.

Process Discovery is concerned with the automatic generation of a process model that describes a business process from execution data of that business process. Real life event logs can contain chaotic activities. These activities are independent of the state of the process and can, therefore, happen at rather arbitrary points in time. We show that the presence of such chaotic activities in an event log heavily impacts the quality of the process models that can be discovered with process discovery techniques. The current modus operandi for filtering activities from event logs is to simply filter out infrequent activities. We show that frequency-based filtering of activities does not solve the problems that are caused by chaotic activities. Moreover, we propose a novel technique to filter out chaotic activities from event logs. We evaluate this technique on a collection of seventeen real-life event logs that originate from both the business process management domain and the smart home environment domain. As demonstrated, the developed activity filtering methods enable the discovery of process models that are more behaviorally specific compared to process models that are discovered using standard frequency-based filtering.

Mining local patterns of process behavior is a vital tool for the analysis of event data that originates from flexible processes, which in general cannot be described by a single process model without overgeneralizing the allowed behavior. Several techniques for mining local patterns have been developed over the years, including Local Process Model (LPM) mining, episode mining, and the mining of frequent subtraces. These pattern mining techniques can be considered to be orthogonal, i.e., they provide different types of insights on the behavior observed in an event log. In this work, we demonstrate that the joint application of LPM mining and other patter mining techniques provides benefits over applying only one of them. First, we show how the output of a subtrace mining approach can be used to mine LPMs more efficiently. Secondly, we show how instances of LPMs can be correlated together to obtain larger LPMs, thus providing a more comprehensive overview of the overall process. We demonstrate both effects on a collection of real-life event logs.