SpinnCer

Project description

The cerebellum is the second major cortex of the brain and is connected to the cerebral cortex through a series of feed-forward and feed-back loops. The cerebellum is thought to implement a forward controller allowing the brain to learn the precise timing of correlated events and is critical for motor and cognitive control. These loops can speed-up brain computation, predict and correct errors, and optimize multiparametric procedures by trials and errors using multidimensional data-sets. Therefore, the cerebellum is likely to dramatically extend the computational capabilities of the whole brain.

Through HBP, the UNIPV group has developed the most advanced detailed models of the cerebellum ever and has started their implementation into neurorobots. A critical missing step is to transform these models into hardware accelerators to be connected to the cerebral cortex, in order to generate new computational architectures with unprecedented capabilities. These could be applied in turn to neurocomputation and neurorobotics. The Manchester group has developed an advanced neuromorphic platform, SPINNAKER, that allows the implementation of different network geometries, neuronal properties and synaptic plasticity rules. While a prototype Of the cerebral cortical circuit has been developed first, the cerebellar network has never been implemented in SPINNAKER.

SpinnCer aims at implementing the currently available cerebellar models written in pyNEST into PYNN and then in SPINNAKER, in order to run the first neuromorphic hardware simulations of the cerebrocortical-cerebellar loop. This work will exploit the Joint Platform system (in particular the Neuromorphic Platform and the Brain Simulation Platform) providing the first example ever of a brain loop transferring realistic neuronal properties into hardware reproducing a major brain loop rather than a single brain region.

Objectives

The expected outcome will be to simulate neuronal activity of the cerebellum through the use of brain inspired hardware and to evaluate possible neurorobotic applications. Through the use of the parallel Spinnaker platform a consistent acceleration is expected and this will allow to increase the number of the involved cells making the simulations more significant and realistic. The target of the project, in fact, relates to the realism of the models simulated: the more they are realistic, the greater will be the inferences that can be taken.

A more sophisticated use is expected for those interested in the simulation of continuous brain processing, e.g. those involved into the analysis of multidimensional time series in fMRl images and EEG signals.

Collaboration with HBP

The collaboration with the Spinnaker group will highlight the feasibility of a neuromorphic (hw-centered) implementation of the algorithms developed by the D'Angelo group on COTS processors. This will really open the possibility of designing a special purpose computer devoted to such a kind of scientific applications and the CCpS group features prestigious contacts and collaborations with companies specialized in supercomputer development.

Partnering Organisations

Publications

Giordana Florimbi, Emanuele Torti, Stefano Masoli, Egidio D’Angelo, Giovanni Danese and Francesco Leporati, “The Human Brain Project: Parallel Technologies for Biologically Accurate Simulation of Granule Cells”, Microprocessors and Microsystems, 2016 Elsevier ed., 47: 303-313, doi:10.1016/j.micpro.2016.05.015, http://www.sciencedirect.com/science/article/pii/S0141933116300515

Giordana Florimbi; Emanuele Torti; Stefano Masoli; Egidio D'Angelo; Giovanni Danese; Francesco Leporati, “Exploiting multi-core and many-core architectures for efficient simulation of biologically realistic models of Golgi cells”, Journal of Parallel and Distributed Processing, December 2018, Volume 126, April 2019, Pages 48-66, https://doi.org/10.1016/j.jpdc.2018.12.004.

Emanuele Torti, Giordana Florimbi, Marta Ticli, Stefano Masoli, Egidio D'Angelo and Francesco Leporati, “GPU parallelization of realistic Purkinje cells with complex morphology”, Proc. of IEEE Euromicro Conference on Parallel and Distributed Computing, Pavia, February 2019, Comp. Society Press ed., Los Alamitos (CA-USA), pp. 266-273. 10.1109/EMPDP.2019.8671581

E. D’Angelo, G. Danese, G. Florimbi, F. Leporati, A. Majani, S. Masoli, S. Solinas, E. Torti, “The Human Brain Project: High Performance Computing for Brain Cells Hw/Sw Simulation and Understanding”, Proc. of IEEE Euromicro Conference on Digital System Design, Funchal, August 2015, pp. 807-810, Comp. Society Press ed., Los Alamitos (CA-USA)

Giordana Florimbi, Emanuele Torti, Giovanni Danese and Francesco Leporati, “High performant simulations of cerebellar Golgi cells activity”, Proc. of IEEE Euromicro Conference on Parallel and Distributed Computing, St. Petersburg, March 2017, Comp. Society Press ed., Los Alamitos (CA-USA), pp. 527-534.. ISBN 978-1-5090-6058-0 .DOI 10.1109/PDP.2017.91

Key facts

Time frame: 2019 to 2020

Origin: HBP Voucher Programme