Improving Accuracy of Nonparametric Transfer Learning via Vector Segmentation

TL;DR

This paper proposes a method to improve nonparametric transfer learning accuracy by segmenting deep neural network features, leveraging the property that relevant information may be concentrated in specific feature segments.

Contribution

It introduces a novel approach of feature segmentation to enhance nonparametric transfer learning, supported by experiments on multiple datasets and neural network extractors.

Findings

Segmenting features improves accuracy in certain distributions.

Deep neural network features have properties exploitable by segmentation.

Method is validated on vision and audio datasets.

Abstract

Transfer learning using deep neural networks as feature extractors has become increasingly popular over the past few years. It allows to obtain state-of-the-art accuracy on datasets too small to train a deep neural network on its own, and it provides cutting edge descriptors that, combined with nonparametric learning methods, allow rapid and flexible deployment of performing solutions in computationally restricted settings. In this paper, we are interested in showing that the features extracted using deep neural networks have specific properties which can be used to improve accuracy of downstream nonparametric learning methods. Namely, we demonstrate that for some distributions where information is embedded in a few coordinates, segmenting feature vectors can lead to better accuracy. We show how this model can be applied to real datasets by performing experiments using three mainstream deep neural network feature extractors and four databases, in vision and audio.