Policy Mirror Descent (PMD) is a powerful and theoretically sound methodology for sequential decision-making. However, it is not directly applicable to Reinforcement Learning (RL) due to the inaccessibility of explicit action-value functions. We address this challenge by introducing a novel approach based on learning a world model of the environment using conditional mean embeddings. We then leverage the operatorial formulation of RL to express the action-value function in terms of this quantity in closed form via matrix operations. Combining these estimators with PMD leads to POWR, a new RL algorithm for which we prove convergence rates to the global optimum. Preliminary experiments in finite and infinite state settings support the effectiveness of our method.

Sequential Bayesian Filtering aims to estimate the current state distribution of a Hidden Markov Model, given the past observations. The problem is well-known to be intractable for most application domains, except in notable cases such as the tabular setting or for linear dynamical systems with gaussian noise. In this work, we propose a new class of filters based on Gaussian PSD Models, which offer several advantages in terms of density approximation and computational efficiency. We show that filtering can be efficiently performed in closed form when transitions and observations are Gaussian PSD Models. When the transition and observations are approximated by Gaussian PSD Models, we show that our proposed estimator enjoys strong theoretical guarantees, with estimation error that depends on the quality of the approximation and is adaptive to the regularity of the transition probabilities. In particular, we identify regimes in which our proposed filter attains a TV $\epsilon$-error with memory and computational complexity of $O(\epsilon^{-1})$ and $O(\epsilon^{-3/2})$ respectively, including the offline learning step, in contrast to the $O(\epsilon^{-2})$ complexity of sampling methods such as particle filtering.

Few-shot learning (FSL) is a central problem in meta-learning, where learners must efficiently learn from few labeled examples. Within FSL, feature pre-training has recently become an increasingly popular strategy to significantly improve generalization performance. However, the contribution of pre-training is often overlooked and understudied, with limited theoretical understanding of its impact on meta-learning performance. Further, pre-training requires a consistent set of global labels shared across training tasks, which may be unavailable in practice. In this work, we address the above issues by first showing the connection between pre-training and meta-learning. We discuss why pre-training yields more robust meta-representation and connect the theoretical analysis to existing works and empirical results. Secondly, we introduce Meta Label Learning (MeLa), a novel meta-learning algorithm that learns task relations by inferring global labels across tasks. This allows us to exploit pre-training for FSL even when global labels are unavailable or ill-defined. Lastly, we introduce an augmented pre-training procedure that further improves the learned meta-representation. Empirically, MeLa outperforms existing methods across a diverse range of benchmarks, in particular under a more challenging setting where the number of training tasks is limited and labels are task-specific. We also provide extensive ablation study to highlight its key properties.

We study a class of dynamical systems modelled as Markov chains that admit an invariant distribution via the corresponding transfer, or Koopman, operator. While data-driven algorithms to reconstruct such operators are well known, their relationship with statistical learning is largely unexplored. We formalize a framework to learn the Koopman operator from finite data trajectories of the dynamical system. We consider the restriction of this operator to a reproducing kernel Hilbert space and introduce a notion of risk, from which different estimators naturally arise. We link the risk with the estimation of the spectral decomposition of the Koopman operator. These observations motivate a reduced-rank operator regression (RRR) estimator. We derive learning bounds for the proposed estimator, holding both in i.i.d. and non i.i.d. settings, the latter in terms of mixing coefficients. Our results suggest RRR might be beneficial over other widely used estimators as confirmed in numerical experiments both for forecasting and mode decomposition.

One of the overarching goals of most machine learning algorithms is to generalize to unseen data. Ensuring and quantifying generalization is of course challenging, especially in the high-dimensional setting. One way of reducing the hardness of a learning problem is to study the invariances that may exist with respect to the distribution of data, effectively reducing its dimension. Handling invariances in data has attracted considerable attention over time in machine learning and applied mathematics more broadly. Two notable examples are image registration [De Castro and Morandi, 1987, Reddy and Chatterji, 1996] and time series alignment [Sakoe and Chiba, 1978, Cuturi and Blondel, 2017, Vayer et al., 2020, Blondel et al., 2021, Cohen et al., 2021]. In practice, data augmentation is a central tool in the machine learning practitioner's toolbox. In computer vision for instance, images are randomly cropped, color spaces are changed, artifacts are added.

The problem of learning functions over spaces of probabilities - or distribution regression - is gaining significant interest in the machine learning community. A key challenge behind this problem is to identify a suitable representation capturing all relevant properties of the underlying functional mapping. A principled approach to distribution regression is provided by kernel mean embeddings, which lifts kernel-induced similarity on the input domain at the probability level. This strategy effectively tackles the two-stage sampling nature of the problem, enabling one to derive estimators with strong statistical guarantees, such as universal consistency and excess risk bounds. However, kernel mean embeddings implicitly hinge on the maximum mean discrepancy (MMD), a metric on probabilities, which may fail to capture key geometrical relations between distributions. In contrast, optimal transport (OT) metrics, are potentially more appealing, as documented by the recent literature on the topic. In this work, we propose the first OT-based estimator for distribution regression. We build on the Sliced Wasserstein distance to obtain an OT-based representation. We study the theoretical properties of a kernel ridge regression estimator based on such representation, for which we prove universal consistency and excess risk bounds. Preliminary experiments complement our theoretical findings by showing the effectiveness of the proposed approach and compare it with MMD-based estimators.

Few-shot learning (FSL) is a central problem in meta-learning, where learners must quickly adapt to new tasks given limited training data. Surprisingly, recent works have outperformed meta-learning methods tailored to FSL by casting it as standard supervised learning to jointly classify all classes shared across tasks. However, this approach violates the standard FSL setting by requiring global labels shared across tasks, which are often unavailable in practice. In this paper, we show why solving FSL via standard classification is theoretically advantageous. This motivates us to propose Meta Label Learning (MeLa), a novel algorithm that infers global labels and obtains robust few-shot models via standard classification. Empirically, we demonstrate that MeLa outperforms meta-learning competitors and is comparable to the oracle setting where ground truth labels are given. We provide extensive ablation studies to highlight the key properties of the proposed strategy.

Finding a good way to model probability densities is key to probabilistic inference. An ideal model should be able to concisely approximate any probability, while being also compatible with two main operations: multiplications of two models (product rule) and marginalization with respect to a subset of the random variables (sum rule). In this work, we show that a recently proposed class of positive semi-definite (PSD) models for non-negative functions is particularly suited to this end. In particular, we characterize both approximation and generalization capabilities of PSD models, showing that they enjoy strong theoretical guarantees. Moreover, we show that we can perform efficiently both sum and product rule in closed form via matrix operations, enjoying the same versatility of mixture models. Our results open the way to applications of PSD models to density estimation, decision theory and inference. Preliminary empirical evaluation supports our findings.

Biased regularization and fine-tuning are two recent meta-learning approaches. They have been shown to be effective to tackle distributions of tasks, in which the tasks' target vectors are all close to a common meta-parameter vector. However, these methods may perform poorly on heterogeneous environments of tasks, where the complexity of the tasks' distribution cannot be captured by a single meta-parameter vector. We address this limitation by conditional meta-learning, inferring a conditioning function mapping task's side information into a meta-parameter vector that is appropriate for that task at hand. We characterize properties of the environment under which the conditional approach brings a substantial advantage over standard meta-learning and we highlight examples of environments, such as those with multiple clusters, satisfying these properties. We then propose a convex meta-algorithm providing a comparable advantage also in practice. Numerical experiments confirm our theoretical findings.

Generative Adversarial Networks (GAN) are powerful methods for learning probability measures and performing realistic sampling [21]. Algorithms in this class aim to reproduce the sampling behavior of the target distribution, rather than explicitly fitting a density function. This is done by modeling the target probability as the pushforward of a probability measure in a latent space. Since their introduction, GANs have achieved remarkable progress. From a practical perspective, a large number of model architectures have been explored, leading to impressive results in data generation [48, 24, 26]. From the side of theory, attention has been devoted to identify rich metrics for generator training, such as f-divergences [36], integral probability metrics (IPM) [16] or optimal transport distances [3], as well as studying their approximation properties [28, 5, 51]. From the statistical perspective, progression has been slower. While recent work has set the first steps towards a characterization of the generalization properties of GANs with IPM loss functions [4, 51, 27, 41, 45], a full theoretical understanding of the main building blocks of the GAN framework is still missing. In this work, we focus on optimal transport-based loss functions [20] and study the impact of two key quantities of the GANs paradigm on the overall generalization performance.