Hi Hans! Welcome!
I will attempt to answer your questions until someone with more "long-standing" experience can reply. Yes. The HTM is capable of "recognizing" (read: Predicting) vast multiples of any type of patterns. After training, given a single input - it would "predict" the next input or inputs it expects. If you would like it to identify the overall pattern that the next predicted input is a part of, you would need to identify that externally or keep a mapping of group-->to-->participants. The same goes for the identification of the Sensor that would be active next (if you would like to retain that information) (Can someone confirm the above?) Does that answer your question? Regards, David On Tue, Feb 24, 2015 at 1:12 PM, Hans Peter <[email protected]> wrote: > Hello everyone, I'm new to the Nupic Community. I have a problem where I > would like to apply the CLA, however I'm not quite sure whether the CLA can > acutally do this. > > > I have several Sensors that give me time-dependent input, lets say > combination of letters: > > > Sensor1: A > > Sensor2: E > > Sensor3: L > > Sensor4: H > > . > > . > > . > > -----------------------------------------> time > > > There is a huge amount of these combinations and it is important to know > that there always comes exactly one letter from each sensor. Now I could > train all these combinations to the CLA and I would like to use the CLA on > input that looks like that. > > > Sensor1: A f > > Sensor2: E i > > Sensor3: m L > > Sensor4: H b > > . > > . > > . > > -----------------------------------------> time > > > As you can see there are now two letter combinations (A,E,H,L) and > (m,f,i,b) that have been trained both individually. Can the CLA recognize > these two patterns at once and tell me what patterns I have? > > > Ideally I would also like to deal with a scenario like this: > > > Sensor1: A f > > Sensor2: E > > Sensor3: m L > Sensor4: H b > > . > > . > > . > > -----------------------------------------> time > > > Here I have the two basic combinations (A,E,H,L) and (m,E,f,b). > > > In practice I have an Input signal where there come a lot of > combinations of letters and all these combinations overlap (in time) . My > goal is to decompose that Signal into „basic“ combinations that I have > taught the CLA. Is that something the algorithm could do? I realize that it > could recognize these letter combinations in time and space domain but > could it recognize more than one at once? > > In reality i dont have letters but similar (not identical) voltage peaks, > but i guess on the conceptual level that should not matter. > > > Greetings to you all! > -- *We find it hard to hear what another is saying because of how loudly "who one is", speaks...*
