An attempt at implementing the Numenta HTM model in java.
This is still very much a hobby/hacking project. I've gotten some functionality to work with an unprovided image data set. Eventually I'd like to flush out the core components and provide a highly parallelized implementation.
Hack away!
1/14/2015 - Recent updates:
- Created a gradle core project with subprojects:
- | core - base functionality including input channels and messaging from mmesh
- | sdr - core sparse, distributed concepts (original htm code)
- | memory - core temporal memory concepts (original mmesh code)
- Copied prototype code from the "mmesh" project to a "memory" subproject
- Moved a lot of code around, mostly down to "core"
- The original htm code now resides in the "sdr" subproject
Near-term hopeful changes:
- Refactor sdr to use the core's Identifier and messaging concepts
5/26/20 - UPDATE:
- The project has shifted a bit. What was a Java implementation of Numenta's HTM has slowly morphed into a homebrew AGI project.
- This is still a hacking project, I have a lot of code that's still local because much of it is old and needs refactoring.
- I'd probably recommend finding one of the Java Numenta HTM projects, consider this a general AI "playground".
- The goal is clear, however, AGI without limits, constraints and implementation worry. If AGI saves us or damns us, we probably deserve either outcome.