This project was developed as part of the Computational NeuroEngineering Laboratory’s “Design, Analysis and Validation of Biologically Plausible Computational Models” research project, headed by Dr. Jose C. Principe. For this project, we used an Echo State Network, a type of Recurrent Neural Network, in conjunction with a Minimum Average Correlation Energy (MACE) filter in order to create a system that can identify neural action potentials (also known as “spikes”). Various experiments using real-world data were used to compare the performance of our ESN/MACE system against threshold and matched filter detectors to ascertain the capabilities of such a system in detecting neural action potentials. Our experiments demonstrated that the ESN/MACE can correctly identify spikes with lower false detection rates than established detection techniques due to the fact that our system captures the inherent variability and covariance information in spike shapes by training.
N. Dedual, M. C. Ozturk, J. C. Sanchez, and J. C. Principe, “An Associative Memory Readout in ESN for Neural Action Potential Detection,” in International Joint Conference on Neural Networks, Orlando, Florida, 2007.