Episode 26: Sugandha Sharma, MIT, on biologically inspired neural architectures, how memories can be implemented, and control theory
Generally Intelligent
00:00
Are We Using Inductive Biases in Deep Learning?
I feel like somehow this entire process, the last two steps that take up like a bulk of the work are absent from the fields of deep modeling. And you know, it seems like there's a core part of your process, but not in deep learning. Yeah, definitely. We probably do need to put in more thought into how we are building these networks because we are kind of implicitly putting in a lot of inductive biases in the network. But sometimes they're not the right inductive biases. Well, that's the problem.
Transcript
Play full episode
