Episode 07: Yujia Huang, Caltech, on neuro-inspired generative models
Generally Intelligent
00:00
The Different Types of Robustness
The two main categories that has benchmark is now adversarial and corruption. It's much harder to model it mathematically or rigorously, because people have some models to generate those data with snow or with some digital or all sorts of effects you can have. But in general, it's just very hard to measure the perturbations in high-dimensional space and image space. The L2 norm is not a very good way to measure the distance between images, right? Although it's a very naturally we use L2 distance to measure how far away are two people, for example,. but it's not avery good metric in image space.
Transcript
Play full episode
