Highlights
This is the official release of ngc-learn version 3.0.0. Major updates/upgrades made to its backend (ngc-sim-lib) and all neuroscience components/tools have been refactored/updated to be compliant with new format/changes (all unit-tests passed). In addition, the ngc-museum has been revised to reflect new format/backend. In addition, new tools, models, and components are included in this stable release, including standard dynamic synapses, a mini-suite of mixture models for density analysis, and new models in the museum including the harmonium and RL-centric SNN.
The following key/recent updates have been made (this list is non-exhaustive):
- (major) upgrade of framework/components to integrate w/ upgrade ngc-sim-lib - small speedup as a result of the changes
- mini-suite of in-built mixture models now in "utils.density"; these include full Gaussian mixture model (GMM), Bernoulli mixture model (BMM), and exponential mixture model (EMM) (Note: tutorial for GMM included in docs)
- new harmonium exhibit and walkthrough tutorial
- new RL-SNN exhibit and walkthrough tutorial
- new dynamic synapses supported - exponential, double-exponential, and alpha synapses
- convenience integrate-and-fire component included
- bernoulli error-cell component included
- new Nesterov's accelerated gradient (NAG) update rule included in in-built optimize tools
- Rao&Ballard PC model now featured as the exhibit model in the "pc_recon" location of museum
- ngc-lava/lava support has been removed from the v3 release
- docs have been reorganized / cleaned up to be v3-compliant (and tested offline); new intro docs written to reflect new ngc-sim-lib and ngc-learn formatting
- minor patches / bug fixes throughout, corrections to typo's/errors in docs