Interesting Article
Interesting, but it links to a mere abstract in some pre-proceedings, and the results do not match those reported in the article you linked. However, I have no doubt that this can be done using different supervised learning methods, when the machine is both trained and tested on the same kind of games (e.g. OTB rated games) and could be applied to online chess. However, because only a small fraction of online chess players have a lot of games that can be traced to their "real" identity (you likely being one of these) this should have small value for marketing and could more likely result in someone trying to identify some super GM's last account (if the code is released) than in some privacy nightmare (we have plenty of those without chess).
Would be cool to analyse your games with it and it tells you which top player (present or past) has the most similar style as you.
I would for sure study this person a lot afterwards :)
Thx for sharing!
I would for sure study this person a lot afterwards :)
Thx for sharing!
@jeffsonadam The full paper is at that link you need to click on the PDF icon to get it (openreview.net/pdf?id=9RFFgpQAOzk).
@fgermuth That's what we were trying to do, but the current version doesn't put similar people close together so we wrote about the privacy issues.
@fgermuth That's what we were trying to do, but the current version doesn't put similar people close together so we wrote about the privacy issues.
@reidmcy said in #5:
> @jeffsonadam The full paper is at that link you need to click on the PDF icon to get it (openreview.net/pdf?id=9RFFgpQAOzk).
>
> @fgermuth That's what we were trying to do, but the current version doesn't put similar people close together so we wrote about the privacy issues.
Interesting choice of application. Behavior identification using neural networks goes back a long way: decades. Years ago I worked for a company that applied it to the financial industry with a great deal of financial success. I'm not at all surprised by the results.
How much did your being a chess player factor into choosing chess behavior?
> @jeffsonadam The full paper is at that link you need to click on the PDF icon to get it (openreview.net/pdf?id=9RFFgpQAOzk).
>
> @fgermuth That's what we were trying to do, but the current version doesn't put similar people close together so we wrote about the privacy issues.
Interesting choice of application. Behavior identification using neural networks goes back a long way: decades. Years ago I worked for a company that applied it to the financial industry with a great deal of financial success. I'm not at all surprised by the results.
How much did your being a chess player factor into choosing chess behavior?
@Algernon12 Our group has done a lot of work with chess and the Lichess database. This was building off of previous work (arxiv.org/abs/2008.10086) that showed players could be identified with only 100s of games, but it required pre-training a model with 20,000+ games. So we were hoping their would be useful insights possible looking at just a few games.
Using Machine/Deep Learning for identification is a well developed task. We adapted a method from speaker verification (GE2E) for this. What's new about our paper is we show that _just_ the behavior is enough, which we the Behavioral Stylometry problem.
Using Machine/Deep Learning for identification is a well developed task. We adapted a method from speaker verification (GE2E) for this. What's new about our paper is we show that _just_ the behavior is enough, which we the Behavioral Stylometry problem.
hmmm interesting.
Do not know what to think.....
First did you also ad time control in your observations. It can add to a clearer picture. Also what patterns/moves played in time trouble.
Did you ask Lichess permission??????
To use Lichess to identify the players is not in the spirit of their conduct. . Lichess is proud of being a 1 cookie website. We users are not the product.
Do not know what to think.....
First did you also ad time control in your observations. It can add to a clearer picture. Also what patterns/moves played in time trouble.
Did you ask Lichess permission??????
To use Lichess to identify the players is not in the spirit of their conduct. . Lichess is proud of being a 1 cookie website. We users are not the product.
p.s. The comment that adding the top 1500 to the player pool did not matter made me smile.
@reidmcy said in #7:
> @Algernon12 Our group has done a lot of work with chess and the Lichess database. This was building off of previous work (arxiv.org/abs/2008.10086) that showed players could be identified with only 100s of games, but it required pre-training a model with 20,000+ games. So we were hoping their would be useful insights possible looking at just a few games.
>
> Using Machine/Deep Learning for identification is a well developed task. We adapted a method from speaker verification (GE2E) for this. What's new about our paper is we show that _just_ the behavior is enough, which we the Behavioral Stylometry problem.
Can you rephrase the last sentence? It seems to have gotten geebed up.
> @Algernon12 Our group has done a lot of work with chess and the Lichess database. This was building off of previous work (arxiv.org/abs/2008.10086) that showed players could be identified with only 100s of games, but it required pre-training a model with 20,000+ games. So we were hoping their would be useful insights possible looking at just a few games.
>
> Using Machine/Deep Learning for identification is a well developed task. We adapted a method from speaker verification (GE2E) for this. What's new about our paper is we show that _just_ the behavior is enough, which we the Behavioral Stylometry problem.
Can you rephrase the last sentence? It seems to have gotten geebed up.
This topic has been archived and can no longer be replied to.