Okay, it does sound like you have a pattern recognition task.

You can visualise the task as trying to match a pattern in your database to a set of pixels on the screen. That would be the lowest level at which you could perform this task. As you have indicated, you could alternatively compute a parameterisation of the pattern and try and match that to known parameterisations.

One important question: will your database of gestures be known before the player sits down to play the game? So that the AI is learning to recognise a specific gesture (basically the same as the voice recognition task. Present a word and learn how it is pronounced).
Or will it be the case that the player associates random gestures to states of the game and hence the correlations are learned by the system? Much like showing someone a picture of a chair and learning a label for it (and no one knows the word ''chair'' is an appropriate label for the object).

Tim

quote:

---

Original post by grahamr

Hey, Tim. How about writing up an article for GameDev? This is a very interesting topic, and I''d love to hear more about it.

---

This couldn''t really be contained in a single article. It might take 2 or even 3. I''ll give some thought to a specific modelling problem in games and formulate a solution using HMMs/DBNs. I would then need to write up everything you need to understand to get to a final working model. That''s a lot of work... I''ll consider it over the weekend.

Tim

quote:

---

Original post by Timkin
Okay, it does sound like you have a pattern recognition task.

---

Ah ! Where are simple words when I need them

Well yeah, pattern recognition. The reason I chose mouse input parametrization rather than picture recognition is because I think it''s more elegant (if only my explanations above weren''t so convoluted...)

Indeed, there is a major difference between having a limited set of gestures, which would be known at runtime; and being able to create your own gestures.
Thing is, I want the user to be able to create new gestures.
That''s where I was wondering how on Earth an ANN would accomodate that. And why I thought that "simple" pattern matching would be nicer...

Mmmmh.

youpla :-P

---

Sancte Isidore ora pro nobis !

Okay, this discussion is getting off-topic for this thread, so why don''t we start another to discuss neural nets and what they can/cannot do.

On the issue of me doing an article on Markov models... okay, but it''s not something I am going to rush. It''ll take me a week or two to get it done as I am quite busy.

Hope that''s okay.

Cheers,

Tim