[intelligent robot]

Ever since I first read On Intelligence I have been awed and inspired by it. I eagerly anticipated HTMs and was thrilled when NuPIC came out. I also worked on some small memory-prediction-framework programs of my own, and I reached an interesting problem.
I realized that spatial pattern recognition was, in fact, a very simple problem. You just need an array of 'neurons', each responding to a single pattern, and a smooth mechanism for some patterns to 'strengthen' over time and others to 'weaken'.
The problem of temporal sequence recognition, especially forward prediction, confused me. I thought about it, day and night, for over two months. I reached no satisfactory conclusions. Finally, reaching the conclusion that it required some knowledge that I did not possess, I let it go for a while.

As an undergraduate student in mathematics, I continued working on some mathematical structures that had little relevance to AI. I called these structures 'Memory Transformations', because they could provide an (almost) unique representation of the past history of a given vector function (over an infinite time scale) in a single vector. I worked on the theoretical basis for these transformations for some time.
When I changed from continuous functions to discrete functions, however, was when the revelation hit. I could use a combination of neuron-like logic (for example, fuzziness and emergence) and discrete memory transformations to do forward prediction.

Then, I read on Numenta's web page that forward prediction had not yet been implemented in NuPIC and that Numenta was looking for people to assist them in this. I was thrilled.
If anyone from Numenta is reading this, I believe I have the solution to this problem and I am willing to share it if you're interested. I would like very much to see the success of HTMs and I believe I can help here.