Adaptation of the emer/leabra (github.com/emer/leabra) hippocampus model for statistical learning
This repository contains an updated version of the Schapiro et al. (2017) statistical learning hippocampus model, originally written in C++ emergent (www.grey.colorado.edu/emergent/index.php/Main_Page). The model has been re-written for the Golang emergent (www.github.com/emer/emergent) framework and is based on the emer/leabra/examples/hip hip.go example.
- Install Golang.
- Install emergent by following the instructions at www.github.com/emer/emergent/wiki/Install.
- Download/Clone this repository.
cdto the repo folder and run the command
go build && hip-sl.
This should open the GUI view for the model. Please refer to emergent documentation for information on the GUI view.
Parameter differences between the hip.go example and hip-SL.go
All parameter changes are listed in the
Architecture differences between the C++ emergent version and hip-SL.go
- KWTA inhition has been replaced by FFFB inhibition. For more information see: www.grey.colorado.edu/CompCogNeuro/index.php/CCNBook/Networks
- Learning rates for MSP and TSP have been changed for better performance:
|Learning Rate||C++ model||Golang model|
- Golang emergent divides trials into four quarters of 25 cycles each, with each quarter serving as a different learning phase. The ActMid, ActM and ActP learning variables from C++ emergent are therefore recorded at different timepoints in a Golang emergent training trial vs a C++ emergent training trial. Cycle at which each variable is recorded for the two models:
Please see https://github.com/schapirolab/hip-sl/wiki/Results for full results.