AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
Machine Learning as an Interpolation Problem
The crux of the problem is that we're trying to approximate a function that has specially unique behavior in more than about 20 or so dimensions. In machine learning, the test set will never fully characterize the problem they were interested in. The whole endeavor of machine earning is defining the right inductive biases and leaving whatever you don't know to the data. And an physically learning to focus your models on the data tha u are actually seen.