Variational Autoencoders and Performers in Deep Learning - Lecture 15

Explore key concepts in deep learning through this comprehensive lecture that covers two major topics: Performers (a variation of Transformers) and Variational Autoencoders (VAE). Begin with an examination of the "Rethinking Attention with Performers" paper, understanding how this architecture modifies traditional Transformer models. Then delve into the fundamentals of variational inference and master the computation of the Evidence Lower Bound (ELBO) for VAE, which forms the core focus of the latter portion of the lecture starting at the 24-minute mark. Gain valuable insights into these advanced deep learning concepts through clear explanations and detailed technical discussions.