Efficient Temporal Learning and Inference for Multiple Source Event-driven Neuromorphic Systems.

Abstract

In various disciplines, information about the same phenomenon can be acquired from differenttypes of sensors, each acquiring different types of information from the same observedphenomenon. The idea of multiplying sensors and modalities is the main characteristic of livingcreatures. These sensors are usually the result of evolution and are perfectly adapted to thesurvival of species in their environment. The design of a robust system implies redundancy,using multiple sensors is essential as a single modality rarely provides complete knowledge ofthe phenomenon of interest.Although neural systems effortlessly seem to be able to handle different type of sensors andseveral forms of data, artificial systems are still lacking a robust sensory fusion theory.The evolution of modern society increases the need to develop more robust artificial systemsthat can interact with the world. And although processing a single modality such as staticimages from the web and a variety of other modalities separately has been achieved, it iscurrently a challenge to merge different sources of information specifically when dealing withdynamic environments. Many of these questions, or challenges are common to multipledomains.

Document Details

Document Type
DoD Grant Award
Publication Date
Jul 20, 2020
Source ID
N629092012051

Entities

People

  • Ryad Benosman

Organizations

  • Office of Naval Research
  • United States Navy
  • Vision Institute

Tags

Readers

  • Distributed Systems and Data Platform Development
  • Educational Psychology
  • Neurological Diseases/Conditions/Disorders

Technology Areas

  • AI & ML