Transfer Learning in CNNs Using Filter-Trees

TL;DR

This paper introduces Bank of Filter-Trees (BFT), a transfer learning method for CNNs that reuses subnetworks from multiple source models to improve learning efficiency without fine-tuning.

Contribution

It proposes a novel transfer learning approach using filter-trees as subnetworks, enabling efficient knowledge transfer from multiple CNNs without weight fine-tuning.

Findings

BFT achieves performance comparable to training from scratch.

Transfer occurs at a subnetwork level from multiple sources.

No fine-tuning required for transferred weights.

Abstract

Convolutional Neural Networks (CNNs) are very effective for many pattern recognition tasks. However, training deep CNNs needs extensive computation and large training data. In this paper we propose Bank of Filter-Trees (BFT) as a trans- fer learning mechanism for improving efficiency of learning CNNs. A filter-tree corresponding to a filter in k^{th} convolu- tional layer of a CNN is a subnetwork consisting of the filter along with all its connections to filters in all preceding layers. An ensemble of such filter-trees created from the k^{th} layers of many CNNs learnt on different but related tasks, forms the BFT. To learn a new CNN, we sample from the BFT to select a set of filter trees. This fixes the target net up to the k th layer and only the remaining network would be learnt using train- ing data of new task. Through simulations we demonstrate the effectiveness of this idea of BFT. This method constitutes a novel transfer learning technique where transfer is at a sub- network level; transfer can be effected from multiple source networks; and, with no finetuning of the transferred weights, the performance achieved is on par with networks that are trained from scratch.