In this thesis, we disentangle the generalized Gauss-Newton and approximate inference for Bayesian deep learning. The generalized Gauss-Newton method is an optimization method that is used in several popular Bayesian deep learning algorithms. Algorithms that combine the Gauss-Newton method with the Laplace and Gaussian variational approximation have recently led to state-of-the-art results in Bayesian deep learning. While the Laplace and Gaussian variational approximation have been studied extensively, their interplay with the Gauss-Newton method remains unclear. Recent criticism of priors and posterior approximations in Bayesian deep learning further urges the need for a deeper understanding of practical algorithms. The individual analysis of the Gauss-Newton method and Laplace and Gaussian variational approximations for neural networks provides both theoretical insight and new practical algorithms. We find that the Gauss-Newton method simplifies the underlying probabilistic model significantly. In particular, the combination of the Gauss-Newton method with approximate inference can be cast as inference in a linear or Gaussian process model. The Laplace and Gaussian variational approximation can subsequently provide a posterior approximation to these simplified models. This new disentangled understanding of recent Bayesian deep learning algorithms also leads to new methods: first, the connection to Gaussian processes enables new function-space inference algorithms. Second, we present a marginal likelihood approximation of the underlying probabilistic model to tune neural network hyperparameters. Finally, the identified underlying models lead to different methods to compute predictive distributions. In fact, we find that these prediction methods for Bayesian neural networks often work better than the default choice and solve a common issue with the Laplace approximation.

We investigate the problem of multiple time series forecasting, with the objective to improve multiple-step-ahead predictions. We propose a multi-task Gaussian process framework to simultaneously model batches of individuals with a common mean function and a specific covariance structure. This common mean is defined as a Gaussian process for which the hyper-posterior distribution is tractable. Therefore an EM algorithm can be derived for simultaneous hyper-parameters optimisation and hyper-posterior computation. Unlike previous approaches in the literature, we account for uncertainty and handle uncommon grids of observations while maintaining explicit formulations, by modelling the mean process in a non-parametric probabilistic framework. We also provide predictive formulas integrating this common mean process. This approach greatly improves the predictive performance far from observations, where information shared across individuals provides a relevant prior mean. Our overall algorithm is called \textsc{Magma} (standing for Multi tAsk Gaussian processes with common MeAn), and publicly available as a R package. The quality of the mean process estimation, predictive performances, and comparisons to alternatives are assessed in various simulated scenarios and on real datasets.

We extend the WSINDy (Weak SINDy) method of sparse recovery introduced previously by the authors (arXiv:2005.04339) to the setting of partial differential equations (PDEs). As in the case of ODE discovery, the weak form replaces pointwise approximation of derivatives with local integrations against test functions and achieves effective machine-precision recovery of weights from noise-free data (i.e. below the tolerance of the simulation scheme) as well as natural robustness to noise without the use of noise filtering. The resulting WSINDy_PDE algorithm uses separable test functions implemented efficiently via convolutions for discovery of PDE models with computational complexity $O(NM)$ from data points with $M = N^{D+1}$ points, or $N$ points in each of $D+1$ dimensions. We demonstrate on several notoriously challenging PDEs the speed and accuracy with which WSINDy_PDE recovers the correct models from datasets with surprisingly large levels noise (often with levels of noise much greater than 10%).

Ideal point models analyze lawmakers' votes to quantify their political positions, or ideal points. But votes are not the only way to express a political position. Lawmakers also give speeches, release press statements, and post tweets. In this paper, we introduce the text-based ideal point model (TBIP), an unsupervised probabilistic topic model that analyzes texts to quantify the political positions of its authors. We demonstrate the TBIP with two types of politicized text data: U.S. Senate speeches and senator tweets. Though the model does not analyze their votes or political affiliations, the TBIP separates lawmakers by party, learns interpretable politicized topics, and infers ideal points close to the classical vote-based ideal points. One benefit of analyzing texts, as opposed to votes, is that the TBIP can estimate ideal points of anyone who authors political texts, including non-voting actors. To this end, we use it to study tweets from the 2020 Democratic presidential candidates. Using only the texts of their tweets, it identifies them along an interpretable progressive-to-moderate spectrum.

In this article, I will explain what the maximum entropy principle is, how to apply it and why it's useful in the context of Bayesian inference. The code to reproduce the results and figures can be found in this notebook. The maximum entropy principle is a method to create probability distributions that is most consistent with a given set of assumptions and nothing more. The rest of the article will explain what this means. First, we need to a way to measure the uncertainty in a probability distribution.

In this paper, three recently introduced reinforcement learning (RL) methods are used to generate human-interpretable policies for the cart-pole balancing benchmark. The novel RL methods learn human-interpretable policies in the form of compact fuzzy controllers and simple algebraic equations. The representations as well as the achieved control performances are compared with two classical controller design methods and three non-interpretable RL methods. All eight methods utilize the same previously generated data batch and produce their controller offline - without interaction with the real benchmark dynamics. The experiments show that the novel RL methods are able to automatically generate well-performing policies which are at the same time human-interpretable. Furthermore, one of the methods is applied to automatically learn an equation-based policy for a hardware cart-pole demonstrator by using only human-player-generated batch data. The solution generated in the first attempt already represents a successful balancing policy, which demonstrates the methods applicability to real-world problems.

Few-shot classification (FSC), the task of adapting a classifier to unseen classes given a small labeled dataset, is an important step on the path toward human-like machine learning. Bayesian methods are well-suited to tackling the fundamental issue of overfitting in the few-shot scenario because they allow practitioners to specify prior beliefs and update those beliefs in light of observed data. Contemporary approaches to Bayesian few-shot classification maintain a posterior distribution over model parameters, which is slow and requires storage that scales with model size. Instead, we propose a Gaussian process classifier based on a novel combination of P\'olya-gamma augmentation and the one-vs-each softmax approximation that allows us to efficiently marginalize over functions rather than model parameters. We demonstrate improved accuracy and uncertainty quantification on both standard few-shot classification benchmarks and few-shot domain transfer tasks.

In this paper, we show that slow feature analysis (SFA), a common time series decomposition method, naturally fits into the flow-based models (FBM) framework, a type of invertible neural latent variable models. Building upon recent advances on blind source separation, we show that such a fit makes the time series decomposition identifiable.

Deep Learning methods are known to suffer from calibration issues: they typically produce over-confident estimates. These problems are exacerbated in the low data regime. Although the calibration of probabilistic models is well studied, calibrating extremely over-parametrized models in the low-data regime presents unique challenges. We show that deep-ensembles do not necessarily lead to improved calibration properties. In fact, we show that standard ensembling methods, when used in conjunction with modern techniques such as mixup regularization, can lead to less calibrated models. In this text, we examine the interplay between three of the most simple and commonly used approaches to leverage deep learning when data is scarce: data-augmentation, ensembling, and post-processing calibration methods. We demonstrate that, although standard ensembling techniques certainly help to boost accuracy, the calibration of deep-ensembles relies on subtle trade-offs. Our main finding is that calibration methods such as temperature scaling need to be slightly tweaked when used with deep-ensembles and, crucially, need to be executed after the averaging process. Our simulations indicate that, in the low data regime, this simple strategy can halve the Expected Calibration Error (ECE) on a range of benchmark classification problems when compared to standard deep-ensembles.

Exchange algorithm is one of the most popular extensions of Metropolis-Hastings algorithm to sample from doubly-intractable distributions. However, theoretical exploration of exchange algorithm is very limited. For example, natural questions like `Does exchange algorithm converge at a geometric rate?' or `Does the exchange algorithm admit a Central Limit Theorem?' have not been answered. In this paper, we study the theoretical properties of exchange algorithm, in terms of asymptotic variance and convergence speed. We compare the exchange algorithm with the original Metropolis-Hastings algorithm and provide both necessary and sufficient conditions for geometric ergodicity of the exchange algorithm, which can be applied to various practical applications such as exponential random graph models and Ising models. A central limit theorem for the exchange algorithm is also established. Meanwhile, a concrete example, involving the Binomial model with conjugate and non-conjugate priors, is treated in detail with sharp convergence rates. Our results justify the theoretical usefulness of the exchange algorithm.