Corruption is frequently observed in collected data and has been extensively studied in machine learning under different corruption models. Despite this, there remains a limited understanding of how these models relate such that a unified view of corruptions and their consequences on learning is still lacking. In this work, we formally analyze corruption models at the distribution level through a general, exhaustive framework based on Markov kernels. We highlight the existence of intricate joint and dependent corruptions on both labels and attributes, which are rarely touched by existing research. Further, we show how these corruptions affect standard supervised learning by analyzing the resulting changes in Bayes Risk. Our findings offer qualitative insights into the consequences of "more complex" corruptions on the learning problem, and provide a foundation for future quantitative comparisons. Applications of the framework include corruption-corrected learning, a subcase of which we study in this paper by theoretically analyzing loss correction with respect to different corruption instances.

Supervised learning in Neural Machine Translation (NMT) typically follows a teacher forcing paradigm where reference tokens constitute the conditioning context in the model's prediction, instead of its own previous predictions. In order to alleviate this lack of exploration in the space of translations, we present a simple extension of standard maximum likelihood estimation by a contrastive marking objective. The additional training signals are extracted automatically from reference translations by comparing the system hypothesis against the reference, and used for up/down-weighting correct/incorrect tokens. The proposed new training procedure requires one additional translation pass over the training set per epoch, and does not alter the standard inference setup. We show that training with contrastive markings yields improvements on top of supervised learning, and is especially useful when learning from postedits where contrastive markings indicate human error corrections to the original hypotheses. Code is publicly released.

Combining complex mathematical models with observational data is an extremely challenging yet ubiquitous problem in the field of modern applied mathematics and data science. Inverse problems, where one is interested in learning inputs to a mathematical model such as physical parameters and initial conditions given partial and noisy observation of model outputs, are hence of frequent interest. Adopting a Bayesian approach[15, 32], we incorporate our prior knowledge on the inputs into a probability distribution, the prior distribution, and obtain a more accurate representation of the model inputs in the posterior distribution, which results from conditioning the prior distribution on the observed data. The posterior distribution contains all the necessary information about the characteristics of our inputs.

Mobile robots that navigate in unknown environments need to be constantly aware of the dynamic objects in their surroundings for mapping, localization, and planning. It is key to reason about moving objects in the current observation and at the same time to also update the internal model of the static world to ensure safety. In this paper, we address the problem of jointly estimating moving objects in the current 3D LiDAR scan and a local map of the environment. We use sparse 4D convolutions to extract spatio-temporal features from scan and local map and segment all 3D points into moving and non-moving ones. Additionally, we propose to fuse these predictions in a probabilistic representation of the dynamic environment using a Bayes filter. This volumetric belief models, which parts of the environment can be occupied by moving objects. Our experiments show that our approach outperforms existing moving object segmentation baselines and even generalizes to different types of LiDAR sensors. We demonstrate that our volumetric belief fusion can increase the precision and recall of moving object segmentation and even retrieve previously missed moving objects in an online mapping scenario.

Gaussian Processes (GPs) offer an attractive method for regression over small, structured and correlated datasets. However, their deployment is hindered by computational costs and limited guidelines on how to apply GPs beyond simple low-dimensional datasets. We propose a framework to identify the suitability of GPs to a given problem and how to set up a robust and well-specified GP model. The guidelines formalise the decisions of experienced GP practitioners, with an emphasis on kernel design and options for computational scalability. The framework is then applied to a case study of glacier elevation change yielding more accurate results at test time.

Causal phenomena associated with rare events occur across a wide range of engineering problems, such as risk-sensitive safety analysis, accident analysis and prevention, and extreme value theory. However, current methods for causal discovery are often unable to uncover causal links, between random variables in a dynamic setting, that manifest only when the variables first experience low-probability realizations. To address this issue, we introduce a novel statistical independence test on data collected from time-invariant dynamical systems in which rare but consequential events occur. In particular, we exploit the time-invariance of the underlying data to construct a superimposed dataset of the system state before rare events happen at different timesteps. We then design a conditional independence test on the reorganized data. We provide non-asymptotic sample complexity bounds for the consistency of our method, and validate its performance across various simulated and real-world datasets, including incident data collected from the Caltrans Performance Measurement System (PeMS). Code containing the datasets and experiments is publicly available.

Variational inference with Gaussian mixture models (GMMs) enables learning of highly tractable yet multi-modal approximations of intractable target distributions with up to a few hundred dimensions. The two currently most effective methods for GMM-based variational inference, VIPS and iBayes-GMM, both employ independent natural gradient updates for the individual components and their weights. We show for the first time, that their derived updates are equivalent, although their practical implementations and theoretical guarantees differ. We identify several design choices that distinguish both approaches, namely with respect to sample selection, natural gradient estimation, stepsize adaptation, and whether trust regions are enforced or the number of components adapted. We argue that for both approaches, the quality of the learned approximations can heavily suffer from the respective design choices: By updating the individual components using samples from the mixture model, iBayes-GMM often fails to produce meaningful updates to low-weight components, and by using a zero-order method for estimating the natural gradient, VIPS scales badly to higher-dimensional problems. Furthermore, we show that information-geometric trust-regions (used by VIPS) are effective even when using first-order natural gradient estimates, and often outperform the improved Bayesian learning rule (iBLR) update used by iBayes-GMM. We systematically evaluate the effects of design choices and show that a hybrid approach significantly outperforms both prior works. Along with this work, we publish our highly modular and efficient implementation for natural gradient variational inference with Gaussian mixture models, which supports 432 different combinations of design choices, facilitates the reproduction of all our experiments, and may prove valuable for the practitioner.

Uncertainty quantification is crucial for assessing the predictive ability of AI algorithms. Much research has been devoted to describing the predictive distribution (PD) $F(y|\mathbf{x})$ of a target variable $y \in \mathbb{R}$ given complex input features $\mathbf{x} \in \mathcal{X}$. However, off-the-shelf PDs (from, e.g., normalizing flows and Bayesian neural networks) often lack conditional calibration with the probability of occurrence of an event given input $\mathbf{x}$ being significantly different from the predicted probability. Current calibration methods do not fully assess and enforce conditionally calibrated PDs. Here we propose \texttt{Cal-PIT}, a method that addresses both PD diagnostics and recalibration by learning a single probability-probability map from calibration data. The key idea is to regress probability integral transform scores against $\mathbf{x}$. The estimated regression provides interpretable diagnostics of conditional coverage across the feature space. The same regression function morphs the misspecified PD to a re-calibrated PD for all $\mathbf{x}$. We benchmark our corrected prediction bands (a by-product of corrected PDs) against oracle bands and state-of-the-art predictive inference algorithms for synthetic data. We also provide results for two applications: (i) probabilistic nowcasting given sequences of satellite images, and (ii) conditional density estimation of galaxy distances given imaging data (so-called photometric redshift estimation). Our code is available as a Python package https://github.com/lee-group-cmu/Cal-PIT .

PAC-Bayes has recently re-emerged as an effective theory with which one can derive principled learning algorithms with tight performance guarantees. However, applications of PAC-Bayes to bandit problems are relatively rare, which is a great misfortune. Many decision-making problems in healthcare, finance and natural sciences can be modelled as bandit problems. In many of these applications, principled algorithms with strong performance guarantees would be very much appreciated. This survey provides an overview of PAC-Bayes bounds for bandit problems and an experimental comparison of these bounds. On the one hand, we found that PAC-Bayes bounds are a useful tool for designing offline bandit algorithms with performance guarantees. In our experiments, a PAC-Bayesian offline contextual bandit algorithm was able to learn randomised neural network polices with competitive expected reward and non-vacuous performance guarantees. On the other hand, the PAC-Bayesian online bandit algorithms that we tested had loose cumulative regret bounds. We conclude by discussing some topics for future work on PAC-Bayesian bandit algorithms.

There is a significant need for principled uncertainty reasoning in machine learning systems as they are increasingly deployed in safety-critical domains. A new approach with uncertainty-aware regression-based neural networks (NNs), based on learning evidential distributions for aleatoric and epistemic uncertainties, shows promise over traditional deterministic methods and typical Bayesian NNs, notably with the capabilities to disentangle aleatoric and epistemic uncertainties. Despite some empirical success of Deep Evidential Regression (DER), there are important gaps in the mathematical foundation that raise the question of why the proposed technique seemingly works. We detail the theoretical shortcomings and analyze the performance on synthetic and real-world data sets, showing that Deep Evidential Regression is a heuristic rather than an exact uncertainty quantification. We go on to discuss corrections and redefinitions of how aleatoric and epistemic uncertainties should be extracted from NNs.