Scientists Working to Get Machines to See Things the Same as Humans


With the help of the federal government, three teams composed of neuroscientists and computer scientists are attempting to wire machines to perform visual identification the same way as the brain. The Intelligence Advanced Research Projects Activity (IARPA) is hoping to increase machine learning by reconstructing pieces of the brain and reverse engineering the algorithms and computations it undergoes. By summer 2017 algorithms based on the teams’ findings will be given an example of an object and then be required to pick out several examples of it among thousands of images in a database. The entire project is slated to last a total of five years.

The researchers are focusing on the part of the cortex that processes vision, a sensory system. The area of the brain has been extensively studied and computer scientists have tried to emulate it many times. To date, no scientists have attempted to reconstruct a complete wiring diagram of a small cube of brain to this extent. Each team is using the theory known as analysis-by-synthesis, meaning the brain makes predictions about immediate future happenings and then reconciles those predictions with what it sees. Each team is exploring a different possibility since the brain might execute analysis-by-synthesis in a number of ways.

—More info:


Leave a reply