Comparing Spiking Neural Networks to Rate Models in Biological Computations

Victoria Parello*, Serhii Bahdasariants , and Sergiy Yakovenko
Department of Chemical and Biomedical Engineering and Department of Exercise Physiology, West Virginia University, Morgantown, WV 26506

Presentation No.: 97

Assigned Category (Presentation Format): Neuroscience (Poster Presentations)

Student’s Major: Biomedical Engineering

This research sought to compare the different approaches to neural networks and their ability to handle dynamic biological systems; specifically, central pattern generators of locomotion in humans. The brain and body are made of dynamic systems that change in response to themselves and external inputs. As a result, these systems are not easily explained or computed mathematically. One approach is by using neural networks, a computer simulation, which can represent dynamic systems, either through a systematic input of data (i.e., Artificial Neural Networks) or through a dynamic change of inputs over time (i.e., Spiking Neural Networks). Both artificial and spiking neural networks are advantageous when working with biological computations due to the similarity in structure. Noise was added to the extrinsic input data and the adaptability between the models was compared. We found that spiking neural networks were more biologically plausible and more adaptable to noise because of the synapse value that acts as a low-pass filter. Such simulations are important for the further understanding of how biological systems work.

Funding:

Program/mechanism supporting research/creative efforts: WVU's SURE program (Rita Rio & Michelle Richards-Babb)