Robust Nonparametric Hypothesis Testing to Understand Variability in Training Neural Networks
This addresses the issue of model interpretability and reliability for researchers and practitioners in machine learning, but it is incremental as it builds on existing work on variability.
The paper tackles the problem that neural networks with similar test accuracy may compute different functions, proposing a new measure of closeness based on robust hypothesis testing on pre-threshold outputs to better understand training variability.
Training a deep neural network (DNN) often involves stochastic optimization, which means each run will produce a different model. Several works suggest this variability is negligible when models have the same performance, which in the case of classification is test accuracy. However, models with similar test accuracy may not be computing the same function. We propose a new measure of closeness between classification models based on the output of the network before thresholding. Our measure is based on a robust hypothesis-testing framework and can be adapted to other quantities derived from trained models.