URSABench: Comprehensive Benchmarking of Approximate Bayesian Inference Methods for Deep Neural Networks
This work provides a tool for researchers to systematically compare methods, but it is incremental as it builds on existing benchmarking efforts without introducing new inference techniques.
The authors tackled the problem of evaluating approximate Bayesian inference methods for deep neural networks by developing URSABench, a benchmark focusing on uncertainty, robustness, scalability, and accuracy for classification tasks.
While deep learning methods continue to improve in predictive accuracy on a wide range of application domains, significant issues remain with other aspects of their performance including their ability to quantify uncertainty and their robustness. Recent advances in approximate Bayesian inference hold significant promise for addressing these concerns, but the computational scalability of these methods can be problematic when applied to large-scale models. In this paper, we describe initial work on the development ofURSABench(the Uncertainty, Robustness, Scalability, and Accu-racy Benchmark), an open-source suite of bench-marking tools for comprehensive assessment of approximate Bayesian inference methods with a focus on deep learning-based classification tasks