We derive closed-form expressions for the Bayes optimal decision boundaries in binary classification of high dimensional overlapping Gaussian mixture model (GMM) data, and show how they depend on the eigenstructure of the class covariances, for particularly interesting structured data. We empirically demonstrate, through experiments on synthetic GMMs inspired by real-world data, that deep neural networks trained for classification, learn predictors which approximate the derived optimal classifiers. We further extend our study to networks trained on authentic data, observing that decision thresholds correlate with the covariance eigenvectors rather than the eigenvalues, mirroring our GMM analysis. This provides theoretical insights regarding neural networks' ability to perform probabilistic inference and distill statistical patterns from intricate distributions.

In this paper, we study a natural policy gradient method based on recurrent neural networks (RNNs) for partially-observable Markov decision processes, whereby RNNs are used for policy parameterization and policy evaluation to address curse of dimensionality in non-Markovian reinforcement learning. We present finite-time and finite-width analyses for both the critic (recurrent temporal difference learning), and correspondingly-operated recurrent natural policy gradient method in the near-initialization regime. Our analysis demonstrates the efficiency of RNNs for problems with short-term memory with explicit bounds on the required network widths and sample complexity, and points out the challenges in the case of long-term dependencies.

Forward stagewise regression is a simple algorithm that can be used to estimate regularized models. The updating rule adds a small constant to a regression coefficient in each iteration, such that the underlying optimization problem is solved slowly with small improvements. This is similar to gradient boosting, with the essential difference that the step size is determined by the product of the gradient and a step length parameter in the latter algorithm. One often overlooked challenge in gradient boosting for distributional regression is the issue of a vanishing small gradient, which practically halts the algorithm's progress. We show that gradient boosting in this case oftentimes results in suboptimal models, especially for complex problems certain distributional parameters are never updated due to the vanishing gradient. Therefore, we propose a stagewise boosting-type algorithm for distributional regression, combining stagewise regression ideas with gradient boosting. Additionally, we extend it with a novel regularization method, correlation filtering, to provide additional stability when the problem involves a large number of covariates. Furthermore, the algorithm includes best-subset selection for parameters and can be applied to big data problems by leveraging stochastic approximations of the updating steps. Besides the advantage of processing large datasets, the stochastic nature of the approximations can lead to better results, especially for complex distributions, by reducing the risk of being trapped in a local optimum. The performance of our proposed stagewise boosting distributional regression approach is investigated in an extensive simulation study and by estimating a full probabilistic model for lightning counts with data of more than 9.1 million observations and 672 covariates.

Supervised classification is one of the most ubiquitous tasks in machine learning. Generative classifiers based on Bayesian networks are often used because of their interpretability and competitive accuracy. The widely used naive and TAN classifiers are specific instances of Bayesian network classifiers with a constrained underlying graph. This paper introduces novel classes of generative classifiers extending TAN and other famous types of Bayesian network classifiers. Our approach is based on staged tree models, which extend Bayesian networks by allowing for complex, context-specific patterns of dependence. We formally study the relationship between our novel classes of classifiers and Bayesian networks. We introduce and implement data-driven learning routines for our models and investigate their accuracy in an extensive computational study. The study demonstrates that models embedding asymmetric information can enhance classification accuracy.

Staged trees are probabilistic graphical models capable of representing any class of non-symmetric independence via a coloring of its vertices. Several structural learning routines have been defined and implemented to learn staged trees from data, under the frequentist or Bayesian paradigm. They assume a data set has been observed fully and, in practice, observations with missing entries are either dropped or imputed before learning the model. Here, we introduce the first algorithms for staged trees that handle missingness within the learning of the model. To this end, we characterize the likelihood of staged tree models in the presence of missing data and discuss pseudo-likelihoods that approximate it. A structural expectation-maximization algorithm estimating the model directly from the full likelihood is also implemented and evaluated. A computational experiment showcases the performance of the novel learning algorithms, demonstrating that it is feasible to account for different missingness patterns when learning staged trees.

Conformal prediction methodologies have significantly advanced the quantification of uncertainties in predictive models. Yet, the construction of confidence regions for model parameters presents a notable challenge, often necessitating stringent assumptions regarding data distribution or merely providing asymptotic guarantees. We introduce a novel approach termed CCR, which employs a combination of conformal prediction intervals for the model outputs to establish confidence regions for model parameters. We present coverage guarantees under minimal assumptions on noise and that is valid in finite sample regime. Our approach is applicable to both split conformal predictions and black-box methodologies including full or cross-conformal approaches. In the specific case of linear models, the derived confidence region manifests as the feasible set of a Mixed-Integer Linear Program (MILP), facilitating the deduction of confidence intervals for individual parameters and enabling robust optimization. We empirically compare CCR to recent advancements in challenging settings such as with heteroskedastic and non-Gaussian noise.

In this paper, we study the problem of estimator selection and hyper-parameter tuning in off-policy evaluation. Although cross-validation is the most popular method for model selection in supervised learning, off-policy evaluation relies mostly on theory-based approaches, which provide only limited guidance to practitioners. We show how to use cross-validation for off-policy evaluation. This challenges a popular belief that cross-validation in off-policy evaluation is not feasible. We evaluate our method empirically and show that it addresses a variety of use cases.

Knowledge Graphs (KGs) are a major asset for companies thanks to their great flexibility in data representation and their numerous applications, e.g., vocabulary sharing, Q/A or recommendation systems. To build a KG it is a common practice to rely on automatic methods for extracting knowledge from various heterogeneous sources. But in a noisy and uncertain world, knowledge may not be reliable and conflicts between data sources may occur. Integrating unreliable data would directly impact the use of the KG, therefore such conflicts must be resolved. This could be done manually by selecting the best data to integrate. This first approach is highly accurate, but costly and time-consuming. That is why recent efforts focus on automatic approaches, which represents a challenging task since it requires handling the uncertainty of extracted knowledge throughout its integration into the KG. We survey state-of-the-art approaches in this direction and present constructions of both open and enterprise KGs and how their quality is maintained. We then describe different knowledge extraction methods, introducing additional uncertainty. We also discuss downstream tasks after knowledge acquisition, including KG completion using embedding models, knowledge alignment, and knowledge fusion in order to address the problem of knowledge uncertainty in KG construction. We conclude with a discussion on the remaining challenges and perspectives when constructing a KG taking into account uncertainty.

In recent years, there has been increasing excitement about the potential of decision support systems based on machine learning to help human experts make more accurate predictions in a variety of application domains, including medicine, education and science [1-3]. In this context, the ultimate goal is human-AI complementarity--the predictions made by the human expert who uses a decision support system are more accurate than the predictions made by the expert on their own and by the classifier used by the decision support system [4-8]. The conventional wisdom is that to achieve human-AI complementarity, decision support systems should help humans understand when and how to use their predictions to update their own. As a result, a flurry of empirical studies has analyzed how factors such as confidence, explanations, or calibration influence when and how humans use the predictions provided by a decision support system [9-12]. Unfortunately, these studies have been so far inconclusive and it is yet unclear how to design decision support systems that achieve human-AI complementarity [13-17]. In this context, Straitouri et al. [18, 19] have recently argued, both theoretically and empirically, that an alternative type of decision support systems may achieve human-AI complementarity, by design. Rather than providing a single label prediction and letting a human expert decide when and how to use the predicted label to update their own prediction, these systems provide a set of label predictions, namely a prediction set, and ask the expert to predict a label value from the set.

Existing AI alignment approaches assume that preferences are static, which is unrealistic: our preferences change, and may even be influenced by our interactions with AI systems themselves. To clarify the consequences of incorrectly assuming static preferences, we introduce Dynamic Reward Markov Decision Processes (DR-MDPs), which explicitly model preference changes and the AI's influence on them. We show that despite its convenience, the static-preference assumption may undermine the soundness of existing alignment techniques, leading them to implicitly reward AI systems for influencing user preferences in ways users may not truly want. We then explore potential solutions. First, we offer a unifying perspective on how an agent's optimization horizon may partially help reduce undesirable AI influence. Then, we formalize different notions of AI alignment that account for preference change from the outset. Comparing the strengths and limitations of 8 such notions of alignment, we find that they all either err towards causing undesirable AI influence, or are overly risk-averse, suggesting that a straightforward solution to the problems of changing preferences may not exist. As there is no avoiding grappling with changing preferences in real-world settings, this makes it all the more important to handle these issues with care, balancing risks and capabilities. We hope our work can provide conceptual clarity and constitute a first step towards AI alignment practices which explicitly account for (and contend with) the changing and influenceable nature of human preferences.