# Understanding trained CNNs by indexing neuron selectivity

## TL;DR

This paper introduces a framework for analyzing CNN neurons by their selectivity to features and classes, aiding interpretability of neural representations in vision models.

## Contribution

It proposes a novel method to index neurons based on selectivity properties, enabling systematic understanding of feature and class representation in CNNs.

## Key findings

- Identified color-selective neurons like red-mushroom in Conv4
- Found class-selective neurons such as dog-face in Conv5
- Established a methodology for deriving various selectivity properties

## Abstract

The impressive performance of Convolutional Neural Networks (CNNs) when solving different vision problems is shadowed by their black-box nature and our consequent lack of understanding of the representations they build and how these representations are organized. To help understanding these issues, we propose to describe the activity of individual neurons by their Neuron Feature visualization and quantify their inherent selectivity with two specific properties. We explore selectivity indexes for: an image feature (color); and an image label (class membership). Our contribution is a framework to seek or classify neurons by indexing on these selectivity properties. It helps to find color selective neurons, such as a red-mushroom neuron in layer Conv4 or class selective neurons such as dog-face neurons in layer Conv5 in VGG-M, and establishes a methodology to derive other selectivity properties. Indexing on neuron selectivity can statistically draw how features and classes are represented through layers in a moment when the size of trained nets is growing and automatic tools to index neurons can be helpful.