Tuesday, February 10, 2009

Gonna get rough out there

In the last year or so, we've been quite quiet around this blog. Too much work is one reason. Another one is that we've had some hopefully real progress, with some new ideas that aren't yet written down. From the top of my mind:

(i) A massively multidimensional representation system that might be able to account for any domain; from Copycat to Bongard problems and Decision-support systems and god knows what else;

(ii) A redefinition of temperature to a distributed and massively parallel model;

(iii) A model of "hedonic" reinforcement learning, or something like that. We still don't have a name for it.

We are now working on the following problems:

(iv) How to create a slipnet from scratch? How to create nodes from experience? How to find out the distances to other nodes? How to change distances on-the-fly, as Copycat does when "opposite" goes nuclear? How and when to create links? When do links become nodes?

The funny thing is, we think we can solve these. But wait, there's more:

(v) how to find what actions can be done with an object? In other words, how to find what is an object's type on-the-fly, during runtime? The representations we are being driven towards to are so general that they do not contain type information.


(vi) how to make new codelets on-the-fly, during runtime?

Ladies and gentlemen, please return to your seats and keep your seat belts securely fastened. Please make sure your seat backs and tray tables are in their full upright position.