Mapping connectomes with crowd and machine intelligence
Sebastian Seung,
Massachusetts Institute of Technology
Tuesday, May 15th 4:30pm
Friend 008
A connectome is a map of a nervous system in the form of a directed
graph in which nodes represent neurons and edges represent synapses.
The ability to rapidly map connectomes could arguably revolutionize
neuroscience, much as genomics has impacted biology. However, the only
connectome known in its entirety is that of the roundworm C. elegans. A
mere 7000 connections between 300 neurons took over a dozen years of
labor to map in the 1970s and 80s. Fortunately, technological advances
are speeding up the mapping of connectomes. New multibeam scanning
electron microscopes will soon generate a petabyte of image data from a
cubic millimeter of brain tissue every two weeks. From such images, it
should be possible to map every connection between neurons in the
volume---in principle. Unfortunately, it could take up to a million
years for a single person to carry out this feat manually. Clearly,
our capacity to acquire "big data" from the brain has far outpaced our
ability to analyze it. My lab has been developing computational
technologies to deal with this data deluge. We have invented the first
machine learning methods based on genuine measures of image segmentation
performance, and have applied these to create artificial intelligence
(AI) for tracing the "wires" of the brain, the branches of neurons. We
have also developed methods of recruiting, training, and aggregating the
"wisdom of crowds" to work with the AI. Both machine and crowd
intelligence are harnessed by EyeWire, an online community of laypersons
who map connectomes by playing a game of coloring neural images.