Using Deep Learning to Make a Model Bigger

The qualities generally I think of them a little bit like a periodic table for for deep learning where you have fully connected layers which are layers that know about everything they know about the past. There's another type that's like a convolutional neural network which is very spatially oriented it cares about shapes and things like that. And then there's another type of layer which is based on attention so maybe you only want to focus on a specific or don't want tofocus on the whole image, etc. So from these basic components that's how these models are built but they're built in as much of an unappinionated way as researchers can manage and still get the models to learn

Transcript

Play full episode

The AI-powered Podcast Player

Save insights by tapping your headphones, chat with episodes, discover the best highlights - and more!

Get the app