SPORE (Synaptic Plasticity with Online Reinforcement learning) is a NEST module that features a collection of neuron and synapse models and infrastructure targeting reward-based learning algorithms.
The source code of the project is available form here: https://github.com/IGITUGraz/spore-nest-module
The main features of the module are:
- It introduces a new class of synapse models that is guaranteed to be updated on a regular time grid (DiligentConnectorModel).
- It introduces TracingNode, a new class of NEST nodes that can store a history of a pice-wise constant real-valued time series (e.g. the membrane potential) for a fixed time window. We call these histories traces. Connections may use these traces to update their state, e.g. to implement a reward mechanism or for learning rules that depend on the somatic potential. This is in many cases more suitable to implementations of synapse models that were developed for reward-based learning.
- It introduces a MUSIC proxy that allows to receive traces from a MUSIC port (RewardInProxy).
- See also
- DiligentConnectorModel, TracingNode, RewardInProxy