11
NoMad
4y

For this new project I'm looking into a neural network the same way neuroscientists look into brain and ngl, it's so fucking cool.

Comments
  • 1
    Explain the project plz
  • 1
    @codecrow no 😛
  • 1
    @codecrow kidding. It's actually a representational similarity analysis of a neural network.
  • 2
    Are you trying to correlate the weights with what the layer or network actually does? That's a really cool exercise.
  • 2
    @RememberMe ya-bingo! Sample-activations vs target-matrices. People have done this type of thing before and I think it's pretty cool.
  • 2
    @RememberMe cooler than that, is that you can literally see how it clusters them (sample/target) in different layers. And how it adjusts those clusters. How it moves from "feature" to "oh, I guess the feature means it's the [target]"
Add Comment