The document discusses advances in continuous unsupervised training of deep learning architectures in biometric systems, focusing on methods such as semi-supervised and incremental learning. It highlights the significance of deep architectures, including Convolutional Neural Networks and Hierarchical Temporal Memory, and addresses the challenges, strategies, and solutions related to continuous learning, including problems like catastrophic forgetting. Additionally, it introduces the Core50 dataset for benchmarking and emphasizes the importance of temporal coherence in learning scenarios.