This repository allows for mixing and matching biologically plausible learning rules for vanilla RNNs, with a focus on tasks and constraints inspired by Brain Machine Interface (BMI) experiments in non-human primates.
The following repository was custom built for the NeurIPS paper Distinguishing Learning Rules with Brain Machine Interfaces (Portes, Schmid and Murray 2022).
Experiments for the NeurIPS 2022 submission are in the branch experiments-neurips-2022
Please contact j.portes@columbia.edu for more details.
The general structure of the simulation class, and probes/monitors in particular, are directly inspired by Owen Marschall's repo https://github.com/omarschall/vanilla-rtrl/
The implementation for the RFLO algorithm is inspired by Murray 2019 "Local online learning in recurrent networks with random feedback" (see https://github.com/murray-lab/rflo-learning for more details)