Computers learn to see people

September 25, 2012 21:35

Human sight works in conjunction with the brain, because of this people can recognize objects, faces of other people, different places, and even to imagine non-existent things. Vicarious Company believes that it is able to simulate the brain and create programs that will also be capable of that.
Vicarious hopes to combine neuroscience and computers to create a visual perception system, repeated in its structure, the cerebral cortex. The idea of "computer neural network" — a program which is able to simulate the brain, creating a connection between Artificial neurons — was at the hearing for decades. But experts Vicarious say that they make it happen thanks to advanced operating time, based on the developed during this time methods, reports Technology Review.

Project co-founder Dileep George, who previously served as chief technology officer for the company Numenta, said that all the other developers "neural connections" based their designs on the model of "neocognitron", which was first proposed in 1980. Usually these systems are trained raspoznovat visual images using random static images. According to George, Vicarious uses a more complex architecture — students with a video stream that changes over time.
Vicarious hopes to develop and run the system on the market for several years. Phoenix co-founder Scott believes that the system may be a large number of applications. For example, in medicine: a computer equipped with the new technology can analyze diagnostic images to determine whether the patient has cancer. Another example: the program equipped smartphone can picture a plate of food to determine how many calories in it. "The computer system of visual perception converts almost all of the activities with which the people", — he added.
Phoenix notes that the program is capable of Vicarious learning, watching images of objects, and create common criteria for the identification of objects. This means that the system is smart enough to identify the object whose information is incomplete. For example, it recognizes that the hand — a hand, even if it is smeared in paint or part of it is hidden wristwatch. "
Vicarious technology keeps details secret, but its development is still aroused the interest of some investors. Last month, the company received $ 15 million from venture capital fund, which participates and Facebook co-founder Dustin Moskovitz.
As the head of the Stanford laboratory of the mind and an associate professor at Stanford University, Andrew Ng, the main problem may be Vicarious of computing power required for accurate modeling of neural processes. Ng participated in the project Google, in which the program was looking through millions of random screenshots videos from YouTube, determine which image has a cat. To make the program have been used 16,000 computers that have joined together in a kind of neural network.

Like this post? Please share to your friends: