NCS-FO: Engineering Living Neural Networks for Learning

Lehigh University, Bethlehem PA

Investigators

Abstract

Recent developments in optogenetics, patterned optical stimulation, and high-speed optical detection enable simultaneous stimulation and recording of thousands of living neurons. Connected biological living neurons naturally exhibit the ability to perform computations and to learn. The proposed project will engineer living neural network to compute for learning task. Experimental testbed will be built to allow optical stimulation and detection. Algorithms will be developed to train the living neuron networks. The proposed testbed can be used by neuroscientists to verify network-level hypotheses. Insights learned from the proposed research can inspire other neuromorphic architectures based on solid state devices. Throughout this project, graduate students will be trained in computer engineering, bioengineering, and signal processing. Students will have the opportunity to work on interdisciplinary research in these fields. New courses based on the results from the proposed work will be introduced and new modules will be added to existing curriculum. The proposed outreach activities aim to attract interest to computer engineering and neural engineering. The goal of this project is to use optogenetic in vitro neural network to run learning applications. Living neural networks have spontaneous activities, which can interfere with precise modification of synaptic strength. This research will study how to stabilize the living neural network such that a Spike Time Dependent Plasticity (STDP)-based programming protocol can imprint the desired synaptic strengths onto a living neural network. This research will also investigate how to strategically design and apply an STDP-based protocol to maximize programming throughput and optimize convergence rate of the network states. On the algorithm side, the proposed research will study data representation and training algorithms that consider various constraints of the proposed wetware system. Learning algorithms will be designed to work on random neural networks of unknown topology. Observable details of neuron activities will be used to improve accuracy of learning tasks. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.