Open chetan51 opened 10 years ago
:+1:
this will also align with the CLA whitepaper..
Note: The discussion about distance function in https://github.com/numenta/nupic/issues/837 is relevant for this task.
@chetan51 I think you had a paper that showed this distribution being useful in other machine learning techniques, can you link that here?
Experiment with initializing permanences using a gaussian distribution, with inputs close to the center of columns having high potential, and inputs farther away having low potential. This may give better results, especially when topology is enabled.