I have implemented part of HTM in Rust (SpatialPooler and SDRClassifier), it should produce same output as the java/python versions because of the same random generator and same algo.
The performance is very similar to C++ and in some cases faster/slower by a bit.
MNIST is running with 95% accuracy
I’m planning to do TM and some Input Encoders later on, and got some more ideas to try, getting more accuracy is one of my goals too, any ideas?