Matching tutor to student: rules and mechanisms for efficient two-stage learning in neural circuits

Authors: Tiberiu Tesileanu, Bence Olveczky, Vijay Balasubramanian

This repository contains the code and data for our paper on efficient two-stage learning in songbirds (and beyond).

Most of the detailed simulation code is contained in the files simulation.py and basic_defs.py. There are tests checking that the code works properly, contained in tests.py. It's a good idea to run this script first after downloading the code to make sure that everything is set up correctly.

The code uses Python 2.7, and it requires a relatively recent installation of iPython (including the jupyter notebook), matplotlib, numpy, scipy, and seaborn. The optimization code uses CMA-ES optimization routines that can be downloaded from https://www.lri.fr/~hansen/cmaes_inmatlab.html.

The code responsible for generating the results and making the plots from the paper is contained in iPython notebooks rate_based_simulations.ipynb for the rate-based model and spiking_simulations.ipynb for the spiking model. The spiking model makes use of the parameters obtained from the optimization procedure described in the Methods section of our paper; these parameters are available in default_params.pkl. The code makes use of helpers.py, which contains various functions that are useful for visualizing the results of the simulations.

To perform the parameter optimization, use the experiment_matcher.ipynb notebook. The data used for the optimization is contained in the data folder -- we thank Timothy Otchy for parts of this data. Note that due to the stochastic nature of the learning simulations and of the CMA-ES optimization algorithm, the result will change every time this code is run. This means that you will not get exactly the same parameters as in default_params.pkl upon running this code.

plasticity_plot.ipynb is a short notebook that was used to make the plot of the plasticity curve for our rule when alpha = beta = 1.

Results and figures are saved in the save and figs folders, respectively, by the iPython notebooks. run_once.py, run_reinf_rate_optim.py, run_reinf_tscale.py, and run_tscale_batch.py are scripts that can be used to generate the job scripts necessary to run the time-consuming spiking simulations on a cluster. These assume a system based on qsub and bash. summarize.py can be used to 'summarize' the results from multiple batch runs by keeping only information about the learning curve and deleting the (very space-consuming) information about intermediate states of the learning process.

Contact

If you have any issues or questions regarding the code, please use the issue tracker, or write to Tiberiu Tesileanu at ttesileanu@gmail.com.