OMASGAN: Out-of-Distribution Minimum Anomaly Score GAN for Sample Generation on the Boundary

Abstract

Generative models trained in an unsupervised manner may set high likelihood and low reconstruction loss to Out-of-Distribution (OoD) samples. This increases Type II errors and leads to missed anomalies, overall decreasing Anomaly Detection (AD) performance. In addition, AD models underperform due to the rarity of anomalies. To address these limitations, we propose the OoD Minimum Anomaly Score GAN (OMASGAN). OMASGAN generates, in a negative data augmentation manner, anomalous samples on the estimated distribution boundary. These samples are then used to refine an AD model, leading to more accurate estimation of the underlying data distribution including multimodal supports with disconnected modes. OMASGAN performs retraining by including the abnormal minimumanomaly-score OoD samples generated on the distribution boundary in a selfsupervised learning manner. For inference, for AD, we devise a discriminator which is trained with negative and positive samples either generated (negative or positive) or real (only positive). OMASGAN addresses the rarity of anomalies by generating strong and adversarial OoD samples on the distribution boundary using only normal class data, effectively addressing mode collapse. A key characteristic of our model is that it uses any f-divergence distribution metric in its variational representation, not requiring invertibility. OMASGAN does not use feature engineering and makes no assumptions about the data distribution. The evaluation of OMASGAN on image data using the leave-one-out methodology shows that it achieves an improvement of at least 0.24 and 0.07 points in AUROC on average on the MNIST and CIFAR-10 datasets, respectively, over other benchmark and state-of-the-art models for AD.

Improving the Fairness of the Min-Max Game in GANs Training

“…According to OMASGAN [9], NDA is not useful for all of the GAN backbones. Thus, we first apply the NDA on StyleGAN2 and find that NDA results in worse performance on StyleGAN2, as shown in Table 8.…”

Improving the Leaking of Augmentations in Data-Efficient GANs via Adaptive Negative Data Augmentation

“…NDA-GANs guide the discriminator to regard the out-of-distribution samples as "fake" instances to improve GANs training. A recent study, OMAS-GAN [7], shows that the performance of NDA varies on different datasets and backbones of GANs.…”

“…By adding NDA on different DE-GANs, the FID only achieves limited improvement or even deteriorates. This is because the performance of NDA varies on different datasets and backbones of GANs [7]. In this case, directly applying NDA in DE-GANs could produce part of the in-distribution samples on the 100-shot-Obama dataset with the Style-GAN2 backbone.…”

“…Given the optimal discriminator D * (T (x)), for the loss function of StyleGAN2, the goal of generator G is to minimize the V G (G, D * ) in Eq. (7). To analyze the convergence of V G (G, D * ) in Eq.…”