Generative Low-Shot Network Expansion
This work addresses the challenge of incremental learning for deep networks in low-data scenarios, which is incremental but practical for applications requiring dynamic class addition.
The paper tackles the problem of expanding a pre-trained deep network to classify novel classes with very few examples, achieving this through a hard distillation method that adds minimal new weights and avoids performance degradation on base classes.
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.