Randomness of Low-Layer Parameters Determines Confusing Samples in Terms of Interaction Representations of a DNN

TL;DR

This paper reveals that the complexity of low-layer interactions in a DNN influences its confusing samples, highlighting the importance of low-layer parameters in generalization and representation power.

Contribution

It demonstrates that low-layer parameters primarily determine confusing samples, extending the lottery ticket hypothesis and deepening understanding of DNN generalization.

Findings

Low-layer parameters significantly influence confusing samples.

Different low-layer parameters lead to distinct confusing sample sets.

Interaction complexity correlates with the DNN's generalization ability.

Abstract

In this paper, we find that the complexity of interactions encoded by a deep neural network (DNN) can explain its generalization power. We also discover that the confusing samples of a DNN, which are represented by non-generalizable interactions, are determined by its low-layer parameters. In comparison, other factors, such as high-layer parameters and network architecture, have much less impact on the composition of confusing samples. Two DNNs with different low-layer parameters usually have fully different sets of confusing samples, even though they have similar performance. This finding extends the understanding of the lottery ticket hypothesis, and well explains distinctive representation power of different DNNs.