This builds on a long line of research into the question of "what sort of adversarial robustness should we care about?", and we owe a large debt to @jmgilmer and @DanHendrycks for their work on this topic. 2/
Prikaži ovu nit
Most adversarial robustness work focuses on the L_inf and L_2 perturbation attacks (red box in the grid below). The new UAR metric lets researchers test their defenses on a much wider range of attacks and see which defenses give broad benefits and which do not. 1/ https://twitter.com/OpenAI/status/1164565192363008000 …pic.twitter.com/yN0ro515Bd
Ovo je potencijalno osjetljiv multimedijski sadržaj. Saznajte više
-
-
-
Also a reminder of my favorite definition of an adversarial example, via
@goodfellow_ian: "Adversarial examples are inputs to machine learning models that an attacker has intentionally designed to cause the model to make a mistake."https://openai.com/blog/adversarial-example-research/ …Prikaži ovu nit
Kraj razgovora
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.