Appendix CSI: Novelty Detection via Contrastive Learning on Distributionally Shifted Instances A Experimental details Training details

The learning rate starts at 0.1 and is dropped by a factor of 10 The detailed description of the augmentations are as follows: Inception crop. After the crop, cropped image are resized to the original image size. We apply color jitter with 80% of probability. Randomly apply a grayscale with 20% of probability. For unlabeled and labeled multi-class datasets, we train ResNet with CIFAR-10 and ImageNet-30.