UBER-HEB: Universal Biologically-inspired Environment for Research: Hierarchical Ersatz Brain

Decades of exponential advancement in computer capabilities have transformed the world, from toddlers’ playthings to the technologies of war and peace. Some further advances are constrained by the limits of von Neumann machines; in particular, they do not learn well. Aptima proposes to conduct a feasibility study to design “UBER-HEB,” a non-von Neumann hierarchical universal learning system. The work is grounded in the Ersatz Brain Project, a biologically-inspired architecture for cognitive computing, based on decades of research by Professor James Anderson of Brown University, a founder of neural networks. The team members have collaborated for a decade. The approach uses many sparsely connected modules inspired by cortical columns. Learning takes places via several integrated mechanisms: Hebbian, dynamic systems forming attractors, interference patterns of activations. It includes hierarchical structures and hierarchical properties also emerge naturally. It takes a biologically plausible and computationally powerful topographical approach to temporal encoding. We will structure the work around three test problems, for which we will develop running code. One of the problems concerns anomaly detection, which is of clear operational relevance. The feasibility assessment and resulting design will thus be based on both theoretical considerations and the results and analysis of the test problems.