ICLR 2020: Yann LeCun and Energy-Based Models
Machine Learning Street Talk (MLST)
00:00
Innovations in GAN Training
This chapter explores the use of self-supervised discriminators in training Generative Adversarial Networks (GANs), emphasizing the integration of contrastive self-supervised learning. It discusses challenges like overfitting, adversarial examples, and the potential for improved manifold formation through advanced model architectures. The chapter also highlights the significance of probabilistic approaches and the use of latent variable models in enhancing predictive capabilities and understanding dynamic scenarios.
Transcript
Play full episode
