UMASS Superior AI Study

Abstract

The objective of this research program is to bring more biological realism to the field of artificial intelligence (AI). To do this we have a multipronged approach involving four distinct tasks. Each task will, by itself, contribute to the overall goal. The individual tasks are each high risk and present their own technical challenges. However, since each is essentially different, failure of one will not spell failure of the overall goal. The goal of the first task is to combine human connectome data with spiking neuron models to create more biologically realistic neural networks. This would of necessity not be a conventional layered network. The second task is to combine fMRI and other data from, for example the Allen Brain Atlas, with spiking neuron models to give us our goal of greater biological realism. The third task will focus on energy utilization in human brains. This task consists of two orthogonal approaches. In one approach we will again use MIR-related data (e.g. Allen Brain Atlas) and data we already have collected, to study the energy utilization. The second orthogonal approach will use gene expression data from the Allen Brain Atlas and National Center for Biological Information combined with protein-protein interaction data from BioGrid, for studying the energy utilization and energy differences associated with brain development, brain diseases, and brains that did not develop properly (e.g. synaptic deficiencies). These two orthogonal approaches may provide insight into more efficient computational architectures, which may help improve the overall goal of biologically realistic AI. Finally, the fourth task will focus on building cortical array architecture analog hardware, studying its phase-space, and developing an “algebra” for design of new hardware systems. The main challenge, and the high-risk element in this task is the development of a mathematical language for system design. Edward Rietman, PhD

Document Details

Document Type
DoD Grant Award
Publication Date
Sep 13, 2016
Source ID
HR00111610006

Entities

People

  • Edward Rietman

Organizations

  • Defense Advanced Research Projects Agency
  • University of Massachusetts

Tags

Readers

  • Instructional Design and Training Evaluation.
  • Neural Network Machine Learning.
  • Neuroscience

Technology Areas

  • AI & ML
  • AI & ML - DoD AI Strategy
  • Space