Net2Vec: Quantifying and Explaining how Concepts are Encoded by Filters in Deep Neural Networks

TL;DR

This paper introduces Net2Vec, a framework that models semantic concepts as distributed embeddings over multiple neural network filters, revealing that concepts are often encoded collectively rather than by individual filters.

Contribution

Net2Vec provides a systematic method to visualize and quantify how multiple filters jointly encode semantic concepts in deep networks, moving beyond extremal filter response analysis.

Findings

Multiple filters are needed to encode a single concept.

Filters often encode multiple concepts rather than being concept-specific.

Filter embeddings better characterize the meaning and relationships of representations.

Abstract

In an effort to understand the meaning of the intermediate representations captured by deep networks, recent papers have tried to associate specific semantic concepts to individual neural network filter responses, where interesting correlations are often found, largely by focusing on extremal filter responses. In this paper, we show that this approach can favor easy-to-interpret cases that are not necessarily representative of the average behavior of a representation. A more realistic but harder-to-study hypothesis is that semantic representations are distributed, and thus filters must be studied in conjunction. In order to investigate this idea while enabling systematic visualization and quantification of multiple filter responses, we introduce the Net2Vec framework, in which semantic concepts are mapped to vectorial embeddings based on corresponding filter responses. By studying such embeddings, we are able to show that 1., in most cases, multiple filters are required to code for a concept, that 2., often filters are not concept specific and help encode multiple concepts, and that 3., compared to single filter activations, filter embeddings are able to better characterize the meaning of a representation and its relationship to other concepts.