Titlebook: Geometry of Deep Learning; A Signal Processing Jong Chul Ye Textbook 2022 The Editor(s) (if applicable) and The Author(s), under exclusive

虚弱的神经

Deep Learning Optimizationtworks are extremely non-convex and not even smooth. This non-convexity and non-smoothness make the optimization unaffordable to analyze, and the main concern was whether popular gradient-based approaches might fall into local minimizers.

resuscitation

Generalization Capability of Deep Learningset, this situation being notorious for overfitting from the point of view of classical statistical learning theory. However, empirical results have shown that a deep neural network generalizes well at the test phase, resulting in high performance for the unseen data.

Vulnerable

1612-3956 to understand the working mechanism of deep learning from high-dimensional geometry is offered. Then, different forms of generative models like GAN, VAE, normalizing flows, optimal transport, and so on are desc978-981-16-6048-1978-981-16-6046-7Series ISSN 1612-3956 Series E-ISSN 2198-3283

Maximizer