Evidence

Fits a multi-dimensional LNP-Integrator model to behavioral data.

Model structure:

1. Feature extraction: time-varying stimulus is processed by LN models
2. Integration: output of each LN ('firing rate') is integrated to yield a feature value ('spike count')
3. Weighing: Feature values of multiple LN models are linearly combined to yield behavioral response value

Some tweaks to optimize performance:

• filter is represented in a raised-cosine basis
• nonlinearity is parameterized (sigmoidal)
• GPU implementation of the model for faster evaluation during fitting (using Matlab's GPU capabilities).

Demo code

load('demo/demo.mat')% loading stimulus and response
p.bee = Behave(stim, resp, ..);
pGa = GA(p);

This should produce the following Figure:

Code base used in:

Jan Clemens, Bernhard Ronacher
Feature extraction and combination underlying decision making during courtship in grasshoppers
2013, Journal of Neuroscience, 33(29):12136-12145 | pdf

Jan Clemens, Matthias Hennig
Computational principles underlying the recognition of acoustic signals in insects
2013, Journal of Computational Neuroscience, 35(1):75-85 | pdf