Adversarial Training of Variational Auto-encoders for Continual Zero-shot Learning(A-CZSL)

Most of the existing artificial neural networks(ANNs) fail to learn\ncontinually due to catastrophic forgetting, while humans can do the same by\nmaintaining previous tasks' performances. Although storing all the previous\ndata can alleviate the problem, it takes a large memory, infeasible in\nreal-world utilization. We propose a continual zero-shot learning model(A-CZSL)\nthat is more suitable in real-case scenarios to address the issue that can\nlearn sequentially and distinguish classes the model has not seen during\ntraining. Further, to enhance the reliability, we develop A-CZSL for a single\nhead continual learning setting where task identity is revealed during the\ntraining process but not during the testing. We present a hybrid network that\nconsists of a shared VAE module to hold information of all tasks and\ntask-specific private VAE modules for each task. The model's size grows with\neach task to prevent catastrophic forgetting of task-specific skills, and it\nincludes a replay approach to preserve shared skills. We demonstrate our hybrid\nmodel outperforms the baselines and is effective on several datasets, i.e.,\nCUB, AWA1, AWA2, and aPY. We show our method is superior in class sequentially\nlearning with ZSL(Zero-Shot Learning) and GZSL(Generalized Zero-Shot Learning).\n