I think this architecture might be more robust than ConvNets, did you analyze the robustness of it?
-
-
-
Really interesting question - we found a relation between human comprehensible capsules and adversarial robustness (https://arxiv.org/pdf/1812.09707.pdf …) so probably "geometric capsules" are more robust without training the network with ERM under attack :D
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.