Visualizing Transfer Learning

TL;DR

This paper visualizes neuron activity during transfer learning in deep image recognition networks, revealing insights into adaptation speed, neuron reuse, feature scales, and small data behavior, supported by a new large-scale dataset.

Contribution

It introduces a novel visualization approach for neurons during transfer learning and provides a large dataset for analyzing transfer learning dynamics.

Findings

Neurons show rapid adaptation during transfer learning.

Neuron reuse varies across layers and features.

Transfer learning behavior differs with small data sets.

Abstract

We provide visualizations of individual neurons of a deep image recognition network during the temporal process of transfer learning. These visualizations qualitatively demonstrate various novel properties of the transfer learning process regarding the speed and characteristics of adaptation, neuron reuse, spatial scale of the represented image features, and behavior of transfer learning to small data. We publish the large-scale dataset that we have created for the purposes of this analysis.