Slow network that learns via backdrop, fast network that implements hebbian learning (gated by slow network)
Meta-learning network with a fast biological RNN (gated hebbian plasticity) coupled with a slow biologicial RNN (trained to adaptively perform a task with the help of the fast RNN). The slow network gates platicity (i.e. weight changes) in the fast network
Network performs 3back task and a WCST analog (see task.py for description of tasks) WCST analog currently trains on up to 6 rules
Can be used to study: (1) gating functions of plasticity in neuromodulatory systems (2) emergent function of networks with fast hebbian component
To run, call train.py Select task in train.py by setting task to 'WCST' or '3Back'