Episode 07: Yujia Huang, Caltech, on neuro-inspired generative models
Generally Intelligent
00:00
How to Make a Neural Network More Robust
For less artificial, I just feel neural networks are sometimes too stupid to be fooled. So, from a label of cats, it won't know what a cat would actually look like. If your label is a dog, you generate some image that will look a dog. You will know it is wrong, right? Because the image of a dog doesn't look what the image that I'm actually saying. That idea is really interesting. And I've always wondered better, maybe you have some thoughts on this. Like, how far can you take this idea?
Transcript
Play full episode
