New Use Case: Cerebro-cerebellar loops

[alberto-antonietti]

New Use Case

Aspect	Detail
Summary	This use-case shows a modeling effort of whole brain functionalities by leveraging on the existing cerebellar models. We model and simulate the forward and inverse controller functions of the cerebrocortical-cerebellar loops in a closed-loop setting, emulating error-based learning and action prediction. The cerebellum model, made of single-point neurons with non-linear dynamics), leverages on the scaffold developed in Task 6.2.2 (see Use-case #529) and is embedded with long-term synaptic plasticity rules for implicit learning. The cerebellum models are split into two modules, one operating in feed-forward mode (receiving motor command, i.e. the efference copy, and generating a sensory error prediction) and the other operating in inverse mode (receiving the sensory information of the desired target and generating compensatory motor commands). The model is used to simulate a specific sensorimotor task in closed-loop. We emulate a pointing task perturbed by prismatic glasses. The cerebellar modules allow to achieve complementary processing able to predict a sensory discrepancy and compensatory motor commands. Therefore, the cerebellum works as a general-purpose massive predictive machine in whole-brain modular systems.
Usecase Group	Brain Area Circuit In Silico Experiments
Expert	Alberto Antonietti
Scientific User
Deadline	March 2020
Target audience	Developers, Computational Neuroscientists
Target interface	Notebook
HPC Requirements	OpenMP
Dependencies	libraries needed are compiled if not present
Nominal runtime	2 hours

List of features

This use case: 1) Prepares the environment with all needed libraries and Python packages:

• NEST 2.18.0
• NEST additional modules: cereb-nest and extra-cereb-nest
• cerebellum-scaffold library 2) Creates the cerebro-cerebellar network and calibrates it (i.e., the output of the spiking neural network has to be transformed into a meaningful variable - angular deviation)

3) Creates the cerebro-cerebellar network and generates two simulations

• A first simulation with a static model of the cerebellar modules (i.e., the plasticity is not active). As a result, the error remains high, around 25°.
  • A second simulation with a plastic model of the cerebellar modules (i.e., the plasticity is active). As a result, the error decreases trial after trial, thanks to cerebellar adaptation.

Acceptance Criteria

The three notebooks have to run, the user can perform the calibration and then the two simulations.

Extra Requirements

System

The notebooks have been tested in CollabV1 and it uses OpenMP (the 16 threads available for Notebooks)

Performance

Preparation of the environment: 30 minutes Calibration: 30 minutes Simulation: 1 hour

[alberto-antonietti]

The use-case in now in PROD in the BSP.

[alberto-antonietti]

PR created to add this UC in the BSP guidebook.

https://github.com/lbologna/hbp-sp6-guidebook/pull/77

[lbologna]

Hello, I accepted the PR and did some minor changes/typos correction:

https://lbologna.github.io/hbp-sp6-guidebook/online_usecases/brain_area_circuit_in_silico_experiments/cerebrocerebellar/cerebrocerebellar_closed_loop_experiment/cerebrocerebellar_closed_loop_experiment.html

@alex4200 , @alberto-antonietti could you have a look before pushing to prod? Thanks.

[alberto-antonietti]

Thanks Luca!

I pushed a few minor changes, I am good with the rest.

[alberto-antonietti]

The documentation of this UC is now in the guidebook. Shall we close this issue?

[alex4200]

@alberto-antonietti @lbologna I found a small typo for this documentation, in the section "Output":

[alex4200]

Fix on production, closing ticket