Deep Learning Through the Lens of Example Difficulty
This work provides insights into how deep models process data, offering a coherent view of phenomena like early layers generalizing and later layers memorizing, which is incremental but useful for researchers in machine learning interpretability.
The paper tackles the problem of understanding deep learning by analyzing individual examples through a measure called prediction depth, revealing relationships between prediction depth and model uncertainty, confidence, accuracy, and learning speed, and shows how this can improve prediction accuracy.
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.