Rethinking Two Consensuses of the Transferability in Deep Learning

TL;DR

This paper investigates the transferability of pre-trained deep neural networks across diverse datasets, confirming previous beliefs and revealing new factors like data amount and diversity affecting transferability.

Contribution

It introduces a method to measure transferability and broadens understanding beyond natural images, highlighting new factors influencing transferability.

Findings

Larger domain gap reduces transferability.

More data and diversity can hinder transferability.

Lower layers are not always the most transferable.

Abstract

Deep transfer learning (DTL) has formed a long-term quest toward enabling deep neural networks (DNNs) to reuse historical experiences as efficiently as humans. This ability is named knowledge transferability. A commonly used paradigm for DTL is firstly learning general knowledge (pre-training) and then reusing (fine-tuning) them for a specific target task. There are two consensuses of transferability of pre-trained DNNs: (1) a larger domain gap between pre-training and downstream data brings lower transferability; (2) the transferability gradually decreases from lower layers (near input) to higher layers (near output). However, these consensuses were basically drawn from the experiments based on natural images, which limits their scope of application. This work aims to study and complement them from a broader perspective by proposing a method to measure the transferability of pre-trained DNN parameters. Our experiments on twelve diverse image classification datasets get similar conclusions to the previous consensuses. More importantly, two new findings are presented, i.e., (1) in addition to the domain gap, a larger data amount and huge dataset diversity of downstream target task also prohibit the transferability; (2) although the lower layers learn basic image features, they are usually not the most transferable layers due to their domain sensitivity.