Episode 17: Andrew Lampinen, DeepMind, on symbolic behavior, mental time travel, and insights from psychology
Generally Intelligent
00:00
Understanding Deep Learning Rethinking Generalization
We were able to say, indeed, linar networks, how well they generalized from noisy data. And we argued that when a linear network learns from noisy data, it will learn all of the signal before it learns any of the pure noise. So you have kind f like a nice story why if you do often while stopping with some validation data, you should get good generalization.
Transcript
Play full episode
