Chapter
The Benefits of Combining Different Learning Methods
49:24 - 56:48 (07:23)
The speaker explains how there is no conflict between different learning methods, like self-supervised, reinforcement, supervised, imitation or active learning. Combining methods can help achieve better results in various tasks by requiring fewer training hours.
Clips
The ability to predict outcomes of a sequence of actions is based on a machine's predictive model of the world, making it easy for deterministic or quasi-deterministic games.
49:24 - 52:01 (02:37)
Summary
The ability to predict outcomes of a sequence of actions is based on a machine's predictive model of the world, making it easy for deterministic or quasi-deterministic games. However, training with least square without a prediction system may result in a blurry image of all possible future positions.
ChapterThe Benefits of Combining Different Learning Methods
EpisodeYann LeCun: Deep Learning, Convolutional Neural Networks, and Self-Supervised Learning
PodcastLex Fridman Podcast
Classical reinforcement learning takes about 80 hours of training to reach the level any human can reach in about 15 minutes.
52:01 - 56:48 (04:46)
Summary
Classical reinforcement learning takes about 80 hours of training to reach the level any human can reach in about 15 minutes. There is no opposition between self-supervised learning, reinforcement learning, and supervised learning, or imitation learning, or active learning to overcome this limitation.