#86 - Prof. YANN LECUN and Dr. RANDALL BALESTRIERO - SSL, Data Augmentation, Reward isn't enough [NEURIPS2022]
Machine Learning Street Talk (MLST)
00:00
Advancements in Self-Supervised Learning
This chapter explores the intersection of self-supervised learning and data augmentation, emphasizing their role in improving supervised learning tasks. The discussion covers spectral properties of similarity matrices, the efficiency of different embedding architectures, and the challenges of modeling at multiple levels of abstraction. Additionally, the speakers highlight the potential of enhancing self-supervised representations through various techniques while advocating for more efficient alternatives to reinforcement learning.
Transcript
Play full episode
