Sam Smith

Constantin Cezar Petrescu, Sam Smith, Rafail Giavrimis et al.

Developers relax restrictions on a type to reuse methods with other types. While type casts are prevalent, in weakly typed languages such as C++, they are also extremely permissive. Assignments where a source expression is cast into a new type and assigned to a target variable of the new type, can lead to software bugs if performed without care. In this paper, we propose an information-theoretic approach to identify poor implementations of explicit cast operations. Our approach measures accord between the source expression and the target variable using conditional entropy. We collect casts from 34 components of the Chromium project, which collectively account for 27MLOC and random-uniformly sample this dataset to create a manually labelled dataset of 271 casts. Information-theoretic vetting of these 271 casts achieves a peak precision of 81% and a recall of 90%. We additionally present the findings of an in-depth investigation of notable explicit casts, two of which were fixed in recent releases of the Chromium project.

Ben Adlam, Jasper Snoek, Samuel L. Smith

Recent work has observed that one can outperform exact inference in Bayesian neural networks by tuning the "temperature" of the posterior on a validation set (the "cold posterior" effect). To help interpret this phenomenon, we argue that commonly used priors in Bayesian neural networks can significantly overestimate the aleatoric uncertainty in the labels on many classification datasets. This problem is particularly pronounced in academic benchmarks like MNIST or CIFAR, for which the quality of the labels is high. For the special case of Gaussian process regression, any positive temperature corresponds to a valid posterior under a modified prior, and tuning this temperature is directly analogous to empirical Bayes. On classification tasks, there is no direct equivalence between modifying the prior and tuning the temperature, however reducing the temperature can lead to models which better reflect our belief that one gains little information by relabeling existing examples in the training set. Therefore although cold posteriors do not always correspond to an exact inference procedure, we believe they may often better reflect our true prior beliefs.