Lately I've been focusing on boosting in my C++ implementation, and working toward a simple vision application. I've fallen behind on a big grad school research project so unfortunately I haven't had much time these days to contribute to OpenHTM. Itay expressed interest in comparing my C++ implementation with the OpenHTM implementation, to try to solve the remaining context forking problems in OpenHTM (which as I understand it is mostly a problem with boosting, to separate the spatial patterns of different inputs so that they are represented by different sets of active columns).
Anyway I've open-sourced my C++ implementation on GitHub, so that it's available if anyone wants to look at how I implemented boosting, or anything else. I wrote up a readme.txt file with instructions and discussions of how my implementation differs from OpenHTM (and from the white paper), and why. The repository is located here:
Very nice. Thanks for this. I think we (openHTM) should take a look at your boosting algorithm and consider implementing it.
Also we should consider implementing the technique of randomizing a range around MinSynapsesPerSegmentThreshold and using those randomized values for each column as Michael explains here:
Lately I've been focusing on boosting in my C++ implementation, and working toward a simple vision application. I've fallen behind on a big grad school research project so unfortunately I haven't had much time these days to contribute to OpenHTM. Itay expressed interest in comparing my C++ implementation with the OpenHTM implementation, to try to solve the remaining context forking problems in OpenHTM (which as I understand it is mostly a problem with boosting, to separate the spatial patterns of different inputs so that they are represented by different sets of active columns).
Anyway I've open-sourced my C++ implementation on GitHub, so that it's available if anyone wants to look at how I implemented boosting, or anything else. I wrote up a readme.txt file with instructions and discussions of how my implementation differs from OpenHTM (and from the white paper), and why. The repository is located here:
https://github.com/MichaelFerrier/HTMCLA
Let me know if there are any questions or if anything needs clarification.
Very nice. Thanks for this. I think we (openHTM) should take a look at your boosting algorithm and consider implementing it.
Also we should consider implementing the technique of randomizing a range around MinSynapsesPerSegmentThreshold and using those randomized values for each column as Michael explains here:
https://sourceforge.net/p/openhtm/discussion/team/thread/2d66fb21/