Explore advanced deep learning concepts in this comprehensive lecture covering discriminative recurrent sparse auto-encoders, group sparsity, world models for autonomous control, and generative adversarial networks (GANs). Delve into the combination of sparse coding with discriminative training, learn about structuring networks with recurrent autoencoders, and understand how group sparsity can extract invariant features. Examine the neural network architecture and training schema for world models, compare them to reinforcement learning, and investigate GANs from an energy-based model perspective using the contrastive method. Gain insights from renowned speaker Yann LeCun in this nearly two-hour session, part of a larger course on practical deep learning.
Group Sparsity, World Model, and Generative Adversarial Networks
Overview
Syllabus
– Week 9 – Lecture
– Discriminative Recurrent Sparse Auto-Encoder and Group Sparsity
– AE With Group Sparsity: Questions and Clarification
– Convolutional RELU with Group Sparsity
– Learning World Models for Autonomous Control
– Reinforcement Learning
– Generative Adversarial Network
Taught by
Alfredo Canziani
