Neural Dataset Generality

TL;DR

This paper introduces a metric to analyze and quantify the generality of datasets based on transfer learning performance, revealing that datasets with diverse atomic structures learn more general filters and that some classes are inherently more general.

Contribution

It proposes a novel method to measure dataset generality through transfer accuracy, providing insights into dataset and class-level filter transferability.

Findings

Datasets with diverse atomic structures learn more general filters.

Certain classes within datasets are more general than others.

The proposed metric reveals interesting relationships between datasets and their transferability.

Abstract

Often the filters learned by Convolutional Neural Networks (CNNs) from different datasets appear similar. This is prominent in the first few layers. This similarity of filters is being exploited for the purposes of transfer learning and some studies have been made to analyse such transferability of features. This is also being used as an initialization technique for different tasks in the same dataset or for the same task in similar datasets. Off-the-shelf CNN features have capitalized on this idea to promote their networks as best transferable and most general and are used in a cavalier manner in day-to-day computer vision tasks. It is curious that while the filters learned by these CNNs are related to the atomic structures of the images from which they are learnt, all datasets learn similar looking low-level filters. With the understanding that a dataset that contains many such atomic structures learn general filters and are therefore useful to initialize other networks with, we propose a way to analyse and quantify generality among datasets from their accuracies on transferred filters. We applied this metric on several popular character recognition, natural image and a medical image dataset, and arrived at some interesting conclusions. On further experimentation we also discovered that particular classes in a dataset themselves are more general than others.