This library integrates selected neuroscience principles from Hierarchical Temporal Memory (HTM) into the pytorch deep learning platform. The current code aims to replicate how sparsity is enforced via Spatial Pooling, as defined in the paper How Could We Be So Dense? The Benefits of Using Highly Sparse Representations.
For detail on the neuroscience behind these theories, read Why Neurons Have Thousands of Synapses, A Theory of Sequence Memory in Neocortex. For a description of Spatial Pooling in isolation, read Spatial Pooling (BAMI).
nupic.torch is named after the original HTM library, the Numenta Platform for Intelligent Computing (NuPIC).
Interested in contributing?
To install from local source code:
python setup.py develop
Or using conda:
conda env create
To run all tests:
python setup.py test
We've created a few jupyter notebooks demonstrating how to use nupic.torch with standard datasets. You can find these notebooks in the examples/ directory or if you prefer you can open them in Google Colab and start experimenting.
For any installation issues, please search our forums (post questions there). Report bugs here.