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?
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.
Per your comment, I wrote a Chrome extension that creates links to other HN submissions pointing to the same URL. Hovering over the links provides additional details.
This provides similar functionality as the 'past' link in a submission's subtitle, but shows the duplicates automatically.
I interpreted your original comment similarly to your description in your reply, so I don't believe I missed nuance (unless I missed it twice). My comment and question were intended to clarify the meaning and usage of the expression "ad hominem", independent of this thread's topic and nuance in your comment. Like I mentioned, I don't use that expression often.
> "It is not ad hominem to bring someone's motivation and prior record into consideration."
I don't speak latin and don't use that expression often. Are you saying it's not ad hominem because the motive is considered as opposed to attacked? I ask because I had thought that it is ad hominem "to bring someone's motivation and prior record into consideration".
Here's the definition on Wikipedia: "a logical fallacy in which an argument is rebutted by attacking the character, motive, or other attribute of the person making the argument, or persons associated with the argument, rather than attacking the substance of the argument itself".
https://en.wikipedia.org/wiki/Ad_hominem
I didn't read the link of this post, or the link in your comment, so hopefully I didn't miss something. I was just curious since I thought "ad hominem" describes what you were saying is not ad hominem.
> "This seems like an dangerous appeal to authority to me. Just because Linus said so doesn't make it true."
What is "this" that you're referring to? (e.g., posting the link to Hacker News, the post making the HN frontpage, a specific comment on this discussion, something in reply to Linus, etc.)
> "Please lead with the location of the position and include the keywords INTERN, REMOTE, or VISA if the corresponding sort of candidate is welcome."
Searching for "intern" matches postings with "internet", "international", etc. I've seen comments suggesting that a search for "remote" matches "no remote".
Here's a quick fix that uses a prefix, for consideration:
> "Please lead with the location of the position and include the keywords +INTERN, +REMOTE, or +VISA if..."
The TechCrunch article mentions a recent law:
"The state’s governor signed a law last October that banned the company from selling its vehicles directly to consumers."
> Join NOVA on a mathematical mystery tour—a provocative exploration of math's astonishing power across the centuries. We discover math's signature in the swirl of a nautilus shell, the whirlpool of a galaxy, and the spiral in the center of a sunflower. Math was essential to everything from the first wireless radio transmissions to the prediction and discovery of the Higgs boson and the successful landing of rovers on Mars. Astrophysicist and writer Mario Livio, along with a colorful cast of mathematicians, physicists, and engineers, follow math from Pythagoras to Einstein and beyond. It all leads to the ultimate riddle: Is math a human invention or the discovery of the language of the universe?
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?