Deep Learning Through the Lens of Example Difficulty

Robert John Nicholas Baldock, Hartmut Maennel, Behnam Neyshabur

Keywords: Deep Learning, Example Difficulty, Prediction Depth

Abstract: Existing work on understanding deep learning often employs measures that compress all data-dependent information into a few numbers. In this work, we adopt a perspective based on the role of individual examples. We introduce a measure of the computational difficulty of making a prediction for a given input: the (effective) prediction depth. Our extensive investigation reveals surprising yet simple relationships between the prediction depth of a given input and the model’s uncertainty, confidence, accuracy and speed of learning for that data point. We further categorize difficult examples into three interpretable groups, demonstrate how these groups are processed differently inside deep models and showcase how this understanding allows us to improve prediction accuracy. Insights from our study lead to a coherent view of a number of separately reported phenomena in the literature: early layers generalize while later layers memorize; early layers converge faster and networks learn easy data and simple functions first.

Code Of Conduct: I certify that all co-authors of this work have read and commit to adhering to the NeurIPS Statement on Ethics, Fairness, Inclusivity, and Code of Conduct.

TL;DR: We study the number of hidden layers used inside a deep model to determine the output for each given input, and find that this quantity connects several aspects of deep learning.

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/arxiv:2106.09647/code)