Computing in Holographic Representation

Abstract

This proposal is focused on brain inspired computing and seeks to answer (1) what kind of computing could explain the brain s powers, (2) how does it correspond to biological neural systems, and (3) how appropriate is it for building artificial systems? Over the past decade we have been building a mathematical framework for cognitive computing based on high dimensional holographic representation that we believe is well suited to the problem of sensorimotor integration. The computing architecture consists of an “inner machine” that is an eclectic pattern detector and generator that interacts with the outside world via sensor and motor specific peripheral processors. The sensor processors encode signals into high dimensional vectors, and the motor processors decode such vectors for the activators. This project will explore the utility of such an architecture by demonstrating a system that learns to choose correct action based on the combination of two or more senses, possibly sound, sight and acceleration. In terms of the correspondence to biological neural systems, we plan to explore connections to the cerebellum, whose neuroanatomical structure maps incredibly well onto the proposed architecture. We expect from this effort to produce neurophysiologically testable predictions of a functional model of cerebellar memory, along with guidelines for engineering systems using these principles.

Document Details

Document Type
DoD Grant Award
Publication Date
Jan 14, 2022
Source ID
FA95501910241

Entities

People

  • Bruno Olshausen

Organizations

  • Air Force Office of Scientific Research
  • United States Air Force
  • University of California Regents

Tags

Fields of Study

  • Computer science

Readers

  • Neural Network Machine Learning.
  • Neuroscience
  • Systems Analysis and Design