From Neuroscience to Computer Science: Multiple Parallel Networks and Big Data Science
Abstract
Abstract Our ability to analyze and make use of big data, an industry predicted to reach $50 Billion by 2017 (Wikibon), is increasingly falling behind our capabilities to generate and collect it. The brain achieves super fast retrieval, and highly efficient decision making by utilizing multiple, parallel neural circuits of differing specialties and properties; its capacity to utilize vast quantities of data is unmatched by current artificial systems. Understanding the brain’s data processing architecture and algorithms is a central issue of neuroscience, neural networks, and big data science. Yet, the fundamental question of how the brain organizes information, parallelizes processing into circuits and draws cohesive semantics out of the activity of multiple, non-similar, parallel modules has yet to be answered; and this gap in our understanding has severely hampered the research community in transferring neuroscientific methods to break through the fast-processing bottle-neck still existing in big data processing. We will approach this fundamental problem from a neuroscientific standpoint by identifying and describing mathematically the brain’s algorithms and supporting structures, the motifs of parallelization and information representation, and the connectors required for a unified system. We will characterize modules and combinators by studying adaptivity, utilizing our state-of–the-art mathematical and computational modeling, our recent neuro-informatics methods, as well as neural simulations. My research, at the nexus of neuroscience, theoretical computation and neural networks, as well as my experience with the practical side of computer science and information retrieval systems, has afforded me insight into potential, biologically sourced, and as yet, untapped methods for combining diverse neural networks into a cohesive and productive information system. Neural networks have been proposed as a solution, yet current networks do not take sufficient advantage from brain motifs of structure and adaptivity. Our approach may provide a breakthrough in neuroscience, neural networks, intelligent big data analysis, and adaptive AI. Big Data management is increasingly vital to the military, government, industry, and medicine; the underlying science is poised and waiting for a next major step.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 12, 2016
- Source ID
- N000141512126
Entities
People
- Hava T Siegelmann
Organizations
- Office of Naval Research
- United States Navy
- University of Massachusetts