The U.S. Intelligence Advanced Research Projects Activity will invest $100 million in the Machine Intelligence from Cortical Networks (MICrONS) program, a project to reverse-engineer a section of the brain, study its computational mechanisms, and feed those insights into the improvement of machine-learning and artificial-intelligence algorithms.
The goal is to enhance artificial neural networks so they get better at pattern recognition in cluttered environments and generalization.
Three teams will employ distinctive methods to map out the neural pathways in a cubic millimeter of a rat's cortex as it engages in visual perception and learning tasks, and then determine how to usefully apply the information to machine-learning algorithms. Their resulting theories will be distilled into internal models they will test against the reverse-engineered brain data. MICrONS intends to use these models to make machines more automatic, especially in terms of simplifying and accelerating object recognition.
MICrONS program manager Jacob Vogelstein believes extracting information from the brain at the computational level can bring the algorithms closer to brain-like performance. "We hope to achieve...better generalization, better capacity for abstraction, better use of sparse data," he says.
The project's challenges include dealing with the massive dataset generated by brain measurements, and mining a vast array of images with segmentation using more refined computer-vision techniques.
From Scientific American
View Full Article
Abstracts Copyright © 2016 Information Inc., Bethesda, Maryland, USA