The uninteresting answer to nearly any "is it possible" question is "yes" (though there are a fair number of no answers out there too for various questions).
But that really doesn't answer the question that wants to be asked. Instead "How would you design..." is a better one (assume the answer is 'yes').
This then brings into the question of "what do you consider an advertisement" and "what do you believe it would be trained on?" Banner image size? Domain source of data?
All in all, I believe the straight forward (not ML) ad blockers do a sufficient job in most cases.
In the advertisement blocking arms race, turning to AI or ML as an approach doesn't really do much besides turn over the battle plans to an... well... idiot. One that you can't understand and can be tricked in ways that we cannot fathom. Trying to debug an issue of "why is this advertisement getting through" in an ML world isn't an easy one to dig into.
2 comments
[ 3.6 ms ] story [ 7.2 ms ] threadBut that really doesn't answer the question that wants to be asked. Instead "How would you design..." is a better one (assume the answer is 'yes').
This then brings into the question of "what do you consider an advertisement" and "what do you believe it would be trained on?" Banner image size? Domain source of data?
Since its rather hard to figure out what is going on behind machine learning, the classifier could be attacked with adversarial approaches ( https://blog.openai.com/adversarial-example-research/ ).
All in all, I believe the straight forward (not ML) ad blockers do a sufficient job in most cases.
In the advertisement blocking arms race, turning to AI or ML as an approach doesn't really do much besides turn over the battle plans to an... well... idiot. One that you can't understand and can be tricked in ways that we cannot fathom. Trying to debug an issue of "why is this advertisement getting through" in an ML world isn't an easy one to dig into.