Neuromorphic engineering seeks to provide efficient and effective solutions to challenging problems. By applying biologically inspired principles such as sparsity and computation via dynamical systems, neuromorphic approaches have demonstrated promising solutions to address various challenges. However, these approaches currently lack a “common language” which could allow for arbitrary algorithms and hardware to be integrated.
“Vector-symbolic architectures” (VSAs) provide a powerful computing framework which could potentially fill this role. VSAs can be used to construct and query lists, dictionaries, graphs, and more. We have also demonstrated that their operations can be used to construct a variety of neural networks. In this seminar, basic VSA operations – similarity, binding, and bundling – are introduced, and their application to carry out graph queries and construct spike-compatible neural networks is demonstrated. Lastly, we comment on VSAs’ relationship to neuromorphic hardware and use as a common primitive for these systems.