
Adversarial Machine Learning
The Data Exchange with Ben Lorica
00:00
I'm Not Trying to Be Creative and Tell You Where the Exploits Can Exist
M l is significantly ulmerable to adversarial exploitation across the spectrum. Theprob mens, it is so vulnerable that if you are deploying it in the mainso for instance, and i'm not trying to. And we can circumvent and penetrate through and albat the fences with significant degree of certainty or accuracy. It's a worrying fact, and I'm purposely not wanting to talk about, is that models, especially some of the deep learning models, tend to be opaque. They're not robust. You an easily perturb an example and make it misbehave. So how do we understand how to protected? But the first problem is what the end to end learning represent representation
Transcript
Play full episode