Defining a Lost Function in Deep Learning?

In the archive, prepent, we had just a few wesults. Nowin we got just more and more. The confidence can be compared to traditional, newer network. We also later, actually added halleni and wagner adversarial attacks and also the fast gradiant sign method. But apart from that, we have good documentation of new networks implementations. And i could just start with whatever had and then try once implemented. You can also get some of the standard architectures out there and do some empirical tests.