Hacker News Reader

Comment by user5994461 | Hacker News Reader

There is a similar technique using markov chain. The author don't seem to be aware.

There are 6 mentions of the word Markov in the paper, not counting citations.

Has anyone proven that the recent spate of "apply neural networks to a text corpus" projects produce output any "better" than a markov chain with the same inputs would?

In speech recognition they fare additional 10-15% better than Markov models and are state-of-art.

Isn't that fundamentally different, though?

With password guessing, one wants to generate as many possible options as possible in order of most likely to least likely. If the generation method takes more than a microsecond it might already be faster to just go the brute force way.

With speech recognition one is also time-constrained, but much less so. If it takes 0.2 seconds to come up with the best match, the user is barely done pronouncing the next word.

Neural stuff is slower than Markov chains if I'm not mistaken, but while for speech recognition that might work great, it could be fatal for password guessing.

You could say that about practically anything.

what's the difference between "There is a similar technique using markov chain. The author don't seem to be aware." AND "There is a similar technique using markov chain! :) "

When you announce an improvement of 18-24% over the state of the art, what you think is the state of the art makes a lot of difference.

The authors state that HashCat uses Markov models and that they outperform it.

He doesn't get it. Someone else replied to his comment stating the same thing you did, i think he is selectively blind to the word "Markov", he missed them in the article and in the replies.

Could you link that?

Reply on news.ycombinator.com