Generative Low-Shot Network Expansion

TL;DR

This paper introduces a low-shot network expansion method that efficiently adds new classes to a pre-trained model with minimal data and memory, avoiding retraining the entire network.

Contribution

It proposes a hard distillation framework that expands networks with few examples, maintaining base performance and using a compact generative model for data representation.

Findings

Hard distillation effectively expands networks with few examples.

The method preserves base class performance.

Memory footprint is significantly reduced using a generative model.

Abstract

Conventional deep learning classifiers are static in the sense that they are trained on a predefined set of classes and learning to classify a novel class typically requires re-training. In this work, we address the problem of Low-Shot network expansion learning. We introduce a learning framework which enables expanding a pre-trained (base) deep network to classify novel classes when the number of examples for the novel classes is particularly small. We present a simple yet powerful hard distillation method where the base network is augmented with additional weights to classify the novel classes, while keeping the weights of the base network unchanged. We show that since only a small number of weights needs to be trained, the hard distillation excels in low-shot training scenarios. Furthermore, hard distillation avoids detriment to classification performance on the base classes. Finally, we show that low-shot network expansion can be done with a very small memory footprint by using a compact generative model of the base classes training data with only a negligible degradation relative to learning with the full training set.