No abstract

No abstract

Bayesian formulations of deep learning have been shown to have compelling theoretical properties and offer practical functional benefits, such as improved predictive uncertainty quantification and model selection. The Laplace approximation (LA) is a classic, and arguably the simplest family of approximations for the intractable posteriors of deep neural networks. Yet, despite its simplicity, the LA is not as popular as alternatives like variational Bayes or deep ensembles. This may be due to assumptions that the LA is expensive due to the involved Hessian computation, that it is difficult to implement, or that it yields inferior results. In this work we show that these are misconceptions: we (i) review the range of variants of the LA including versions with minimal cost overhead; (ii) introduce laplace, an easy-to-use software library for PyTorch offering user-friendly access to all major flavors of the LA; and (iii) demonstrate through extensive experiments that the LA is competitive with more popular alternatives in terms of performance, while excelling in terms of computational cost. We hope that this work will serve as a catalyst to a wider adoption of the LA in practical deep learning, including in domains where Bayesian approaches are not typically considered at the moment. Our laplace library is available at: https://github.com/AlexImmer/Laplace.

We address the problem of predicting aggregate vote outcomes (e.g., national) from partial outcomes (e.g., regional) that are revealed sequentially. We combine matrix factorization techniques and generalized linear models (GLMs) to obtain a flexible, efficient, and accurate algorithm. This algorithm works in two stages: First, it learns representations of the regions from high-dimensional historical data. Second, it uses these representations to fit a GLM to the partially observed results and to predict unobserved results. We show experimentally that our algorithm is able to accurately predict the outcomes of Swiss referenda, U.S. presidential elections, and German legislative elections. We also explore the regional representations in terms of ideological and cultural patterns. Finally, we deploy an online Web platform (www.predikon.ch) to provide realtime vote predictions in Switzerland and a data visualization tool to explore voting behavior. A by-product is a dataset of sequential vote results for 330 referenda and 2196 Swiss municipalities.

Pre-trained contextual representations have led to dramatic performance improvements on a range of downstream tasks. Such performance improvements have motivated researchers to quantify and understand the linguistic information encoded in these representations. In general, researchers quantify the amount of linguistic information through probing, an endeavor which consists of training a supervised model to predict a linguistic property directly from the contextual representations. Unfortunately, this definition of probing has been subject to extensive criticism in the literature, and has been observed to lead to paradoxical and counterintuitive results. In the theoretical portion of this paper, we take the position that the goal of probing ought to be measuring the amount of inductive bias that the representations encode on a specific task. We further describe a Bayesian framework that operationalizes this goal and allows us to quantify the representations' inductive bias. In the empirical portion of the paper, we apply our framework to a variety of NLP tasks. Our results suggest that our proposed framework alleviates many previous problems found in probing. Moreover, we are able to offer concrete evidence that-for some tasks-fastText can offer a better inductive bias than BERT. 1