I saw the HTM studio and I am totally impressed.
But I have a few questions:
Does HTM studio can only use one set of data and find the normal pattern in that data set?
Is there any other applications that can find the relationship between two data set?
Take automatic car for example, I get two data set, one is the distance between the front car and this car, and one is the speed of that car. So if there is a car near this car then this car should lower its speed. Does some application using HTM do that? (Learn from this two data sets and learn the behavior)
I could be wrong, but from what I know this does not really match up with what HTM does.
Hi, Thank you for reply.
Do you mean that HTM theory can only learn one certain set of data and find the certain pattern in the data set?
Can the location-based framework or any other framework proposed by Numenta do that (find the relationship of two data set)? Also, I saw that on the HTM-studio website, it shows that it can analyze the traffic pattern.
What I am interested in is using continuous learning to find out some certain pattern instead of traditional machine learning algorithm to feed some patches of data set
I think HTM applications (so far) are mainly auto-regressive – meaning a model is continuously fed data from a certain system, and it learns the sequential patterns to predict the future state of that system.
So it is not usually used to predict the value of variable Y from variable X, but rather to predict the next X value from the current X value (in the context of all prior X values). The system can also be multivariate (so the next time step of XY from all prior XY), though I think HTM studio only does anomaly detection on univariate data.
You can do forecasting (as opposed to anomaly detection) using NuPIC, using the ‘TemporalMultistep’ inference type. If the system is multivariate you can control which variable is forecasted using ‘predictedField’ in the config.