Really interesting paper exploring adversarial inputs to ML models: https://arxiv.org/abs/1905.02175
They conclude:* It's a property of the input data, not the training* You can even train a model on non-robust features and obtain a model that works well on the original input data!
@wilfredh I'm not sure I'm drawing the same conclusion from their own examples...
@cwebber How so? What conclusion did you draw?
@cwebber @wilfredh I too overviewed thatUnfortunately ML is my weakest point,so I couldn't understand it deeply
Server run by the main developers of the project It is not focused on any particular niche interest - everyone is welcome as long as you follow our code of conduct!