As we all know humans can't generate truly random numbers. So, I was wondering if anyone has developed an application that uses ANN to predict what number a human will guess next. If so, has anyone succeeded in getting a high success rate? I want to create my own version but I wanna see if there are some already made first to see if its a worthwhile effort or if its been proven impossible.

I haven't done much with neural networks myself, but I have seen someone make a neural network to play scissors-paper-rock, which I guess is a similar application.

Searching for them found this link, but the Java applet version of the game seemed to be down (at least for me).

thanks thats exactly what I was looking for... unfortunately its down for me as well... anyone know any similar ones?

Check this one :

http://www.essentially.net/rsp/index.jsp

Check this one :

http://www.essentially.net/rsp/index.jsp

Why a neural network? Other algorithms do exist that would actually work. It's difficult to imagine what advantages a neural network would have in a problem like this where the goal is to quickly recognize and adapt to changing patterns.

because the project curriculum requires the use of an ANN.

Another example of holding a hammer and looking for a screw to bash in. Why do people insist on starting with a tool and looking for a problem to apply it to?

Anyway, enough of my crusade...

If you've been given the task of using an ANN for a prediction task, then you'll have more success if you use a recurrent architecture. However, you'll also face more difficulties in training it.

Which ANN architecture are you required to use and what are the constraints on the application that you may apply it to?

Cheers,

Timkin

IMHO, an ANN having the last 4 or 5 precedent "moves" of both players as input could work just fine.

But then, of course, a little tweaking one could surely improve the success rate.