In this workgroup, we will train spiking balanced random networks via supervised eligibility propagation (Bellec et al., 2020) in the NEST simulator (Sinha et al., 2023). One concrete idea would be to endow a network including spiking central pattern generators (CPGs) for locomotion to drive a robotic arm. Traditionally known as half-center oscillator architecture, this network model consists of two rhythm-generating populations that mutually inhibit each other to drive anti-phasic oscillating output. The populations in this spiking version of the Strohmer et al. (2022) model are random and balanced, and produce sparse firing. Applying plasticity would serve two purposes: tuning the parameters and training the network with feedback from a musculoskeletal model. Finally, we would compare the learning performance of this specific architecture to a balanced random network without the rhythm-generating mechanism to study the potential functional benefits of oscillations or to network models of higher brain regions.

Bellec G, Scherr F, Subramoney A, et al. (2020). A solution to the learning dilemma for recurrent networks of spiking neurons. Nat Commun 11, 3625. https://doi.org/10.1038/s41467-020-17236-y

Sinha A, de Schepper R, Pronold J, Mitchell J, Mørk H, Nagendra Babu P, et al. (2023). NEST 3.4. Zenodo. https://doi.org/10.5281/zenodo.6867800

Strohmer B, Mantziaris C, Kynigopoulos D, Manoonpong P, Larsen LB and Büschges A (2022). Network Architecture Producing Swing to Stance Transitions in an Insect Walking System. Front. Insect Sci. 2:818449. https://doi.org/10.3389/finsc.2022.818449