This repository provides a tutorial on how to model neural adaptation. It complements the primer on "Neural adaptation" by Jan Benda published in Current Biology, 2021. Feel free to use and distribute the scripts and figures for teaching. The code, figures, and descriptions are provided under the GNU General Public License v3.0.

For each of the topics listed below, there is a folder TOPIC/ containing the following files:

• README.md: a tutorial on the topic and how to code the models.
• TOPIC.py: standalone python script containing the functions explained in the README.md file that can be run as a demo.
• TOPICplots.py: python script using the functions in TOPIC.py to generate the figures needed for the README.md file.
• TOPIC-*.png: figure files for README.md generated by TOPICplots.py.

For running the demo script, change into the directory of the topic and run the script. For example, to run the demo for the adapting leaky integrate-and-fire model in 'lifac/' do:

cd lifac
python3 lifac.py

Or open the 'lifac/lifac.py' script in your favorite IDE and run it from there.

Cite this repository via the associated manuscript:

Benda, Jan (2021): Neural adaptation. Current Biology 31(3), R110-R116. doi:10.1016/j.cub.2020.11.054

The python scripts run in python version 3, using the following packages:

• numpy
• scipy >= 1.2.0
• matplotlib >= 2.2.0

The leaky integrate-and-fire model is a simple model of a spiking neuron. Augmented with a generic adaptation current it reproduces many features of intrinsically adapting neurons. Read more in lifac/.

Spike-frequency adaptation is a phenomenon observed on the level of, well, spike frequencies. Modeling adaptation on the level of spike freuqencies is thus a natural choice. Read more in sfa/.

Spike-frequency adaptation basically adds a high-pass filter to the neuron's input-output function. This filter operation interacts with the non-linear f-I curves of the neuron. Read more in filter/.

Subtractive adaptation is perfectly suited to make the neuron's response invariant with respect to the mean of the stimulus. Invariance to the stimulus variance, however, requires thresholding to extract the amplitude modulation and divisive adaptation. Read more in meanvariance/.

Adaptation in parallel pathways leads to stimulus-specific adaptation. Read more in ssa/.

Absolute stimulus intensity is ambiguously encoded by an adapting neuron. Nonetheless, matched intrinsic adaptation allows down-stream neurons to robustly encode absolute stimulus intensity. Read more in ambiguities/.

Efficient codes are both temporally and spatially sparse. Intrinsic adaptation together with lateral inhibition generate such sparse codes. Read more in sparse/.

You are welcome to improve the code and the explanations. Or even add another chapter.

Fork the repository, work on your suggestions and make a pull request.

For minor issues, e.g. a reference you want me to add, or fixing little quirks, please open an issue.