Energy in Learning
Ishan discusses the innovative concept of using energy functions to unify various machine learning models, such as GANs and VAEs. By framing relationships between data points in terms of energy, similarities are represented as low energy states, while dissimilarities correspond to high energy peaks. This approach not only enhances understanding of contrastive learning but also highlights its applicability in both supervised and self-supervised contexts.In this clip
From this podcast
Lex Fridman Podcast
Ishan Misra: Self-Supervised Deep Learning in Computer Vision | Lex Fridman Podcast #206
Related Questions