simrun ❭ modular_reduced_model_inference ❭ strategy

strategy¶

Strategies for creating reduced models.

Strategies are pipelines whose sole purpose is to define a cost function. Cost functions are functions \(f: D, x \rightarrow c\) that take data \(D\) and parameters \(x\) as input and return some cost \(c\). Solvers then optimize these cost functions to find the best parameters \(x\) for the given data \(D\).

This approach is purposefully kept very general, so that it can be used for a wide range of purposes.

One such example is given in Bast et al. [5]. Here, the strategy contains a set of raised cosine basis functions. These are weighed and superimposed to create spatiotemporal filters. Once multiplied with synaptic activation data, they provided a weighed input of synapse activations. This is then used to predict the spike probability based on the synaptic activation patterns. The parameters that are being optimized are the weights of the raised cosine basis functions.

Classes¶

Functions¶