simrun ❭ modular_reduced_model_inference

modular_reduced_model_inference¶

Create reduced models from synapse activation data.

This package implements strategies (see below) to create reduced models from synapse activation data. It is implemented so that it allows for a “modular” inference: fitting can be performed in parallel on different parts of the data. This was a necessary requirement when considering both the spatial and temporal dimension of the input snapse activation data.

A strategy defines a reduced model containing a set of free parameters \(\mathbf{x}\) that are optimized to match the input data. It also defines what data needs to be reproduced in the first place, and what constitutes a good match between the reduced model and the input data. For a list of available strategies, see strategy.

Example

Bast et al. [5] describes how to use a raised cosine basis to create a reduced model from synapse activation data. In this case, the input data are synapse activation patterns, and the target data are spike times. The strategy in this case is Strategy_spatiotemporalRaisedCosine, which defines a linear sum of raised cosine basis functions. These functions are to be multiplied with the input data to predict spike probabilities. The free parameters are then the weights of the linear combination of these basis functions.

Modules¶