MATLAB code for a network filter model, consisting of 'filter-and-fire' neurons with both recurrent and feed-forward filters, that performs close to optimal tracking of its input.
When using this code, please cite the following preprint:
Zeldenrust, F., Gutkin, B., & Denève, S. (2019). Efficient and robust coding in heterogeneous recurrent networks. BioRxiv. https://doi.org/10.1101/804864
This MATLAB toolbox creates and runs the filter network for efficient coding described in the paper above. A few examples scripts are given:
* In 'Example_run_model' it is shown how to create representing filters for each neuron, and then run the network. It first creates a set of basis-kernels (make_basisfunctions). Next, it creates a set of feed-forward filters as a random combination of these basisfunctions, and creates the corresponding recurrent filters for optimal coding (generate_filters). The network can now track any input signal (run_model).
* In 'Example_run_network_paper' the different networks as used in the paper (parameter 'net' sets it to homogeneous, heterogeneous and type 1 & type 2) are pre-defined. A network is then created (make_kernels_network) and run.
* The following scripts will run the simulations for the figures of the paper:
* Figure 3: run_model_if_sta_prc
* Figure 4: run_model_redundancy
* Figure 5: run_model_noise
* Figure 6: run_model_correlations
see: Example_run_model and Example_run_network_paper.
Scientific Coordinator: Fleur Zeldenrust
This model was originally developed in: Matlab
The code for this model is hosted on GitHub: https://github.com/fleurzeldenrust/Efficient-coding-in-a-spiking-predictive-coding-network