Few-Shot Open-Set Recognition using Meta-Learning
About
The problem of open-set recognition is considered. While previous approaches only consider this problem in the context of large-scale classifier training, we seek a unified solution for this and the low-shot classification setting. It is argued that the classic softmax classifier is a poor solution for open-set recognition, since it tends to overfit on the training classes. Randomization is then proposed as a solution to this problem. This suggests the use of meta-learning techniques, commonly used for few-shot classification, for the solution of open-set recognition. A new oPen sEt mEta LEaRning (PEELER) algorithm is then introduced. This combines the random selection of a set of novel classes per episode, a loss that maximizes the posterior entropy for examples of those classes, and a new metric learning formulation based on the Mahalanobis distance. Experimental results show that PEELER achieves state of the art open set recognition performance for both few-shot and large-scale recognition. On CIFAR and miniImageNet, it achieves substantial gains in seen/unseen class detection AUROC for a given seen-class classification accuracy.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Domain Adaptive Few-Shot Open-Set Recognition | MiniImageNet to CUB | Accuracy30.71 | 26 | |
| Few-Shot Open-Set Recognition | miniImageNet (test) | Accuracy80.61 | 22 | |
| Domain Adaptive Few-Shot Open-Set Recognition | Office-Home Real-World to Clipart (test) | Accuracy (%)20.16 | 22 | |
| Domain Adaptive Few-Shot Open-Set Recognition | DomainNet Real to Clipart | Accuracy0.2252 | 22 | |
| Domain Adaptive Few-Shot Open-Set Recognition | DomainNet Real to Painting | Accuracy23.24 | 22 | |
| Domain Adaptive Few-Shot Open-Set Recognition | DomainNet Clipart to Painting | Accuracy24.82 | 22 | |
| Few-Shot Open-Set Recognition | tieredImageNet (test) | Accuracy84.1 | 20 | |
| Few-Shot Open-Set Recognition | tiered-ImageNet 5-shot | AUROC73.27 | 20 | |
| Few-Shot Open-Set Recognition | mini-ImageNet 1-shot | AUROC60.57 | 20 | |
| Few-Shot Open-Set Recognition | mini-ImageNet 5-shot | Accuracy80.61 | 20 |