dstein64

Adversarial images targeting an image classifier have been shown to be transferable to separately trained models (i.e., models with different weights or different architectures relative to the model for which the adversarial image was constructed to target).

I'm curious if the adversarial CA reprogramming techniques are similarly transferable. That is, do the adversarial CA and/or the adversarial perturbation matrix transfer to separate CAs (trained on the same task) with different weights or architectures than the original CA that was targeted?

Is this something you've explored or plan to?

I updated the the quote. Thanks!

> "Hofstadter's Law: It always takes longer than you expect, even when you take into account Hofstadter's Law"

- Gödel, Escher, Bach: An Eternal Golden Braid. 20th anniversary ed., 1999, p. 152.

https://en.wikipedia.org/wiki/Hofstadter%27s_law

Out of their Minds: The Lives and Discoveries of 15 Great Computer Scientists https://cs.nyu.edu/shasha/outofmind.html

I used a similar algorithm when developing a Chrome extension a few years ago: https://chrome.google.com/webstore/detail/auto-highlight/dnk...

Longer sentences had an inherent advantage, so I controlled for that by reducing sentence scores as a function of the sentence length.

If this question is directed at the author of the site, there is a comment section on the corresponding blog post: http://www.catonmat.net/blog/announcing-techurls/

Three factors I view as appealing, versus navigating to each site separately, include 1) the aggregation itself has value (e.g., to save time), 2) the primarily textual interface and 3) the search functionality.

For the title of this HN submission, the text is from the og:description tag in the page's html.

I'm not sure how the popular websites were selected.

Perhaps the author of the site can answer your question in the comments of the corresponding blog post. http://www.catonmat.net/blog/announcing-techurls/