Computers learn to look like people

Visual activity combined with the brain helps people recognize objects, differentiate people, different scenes, even visualize things that do not exist. Vicarious said it could simulate the brain to create software that has those capabilities.

Vicarious hoped to combine the activity of the human brain with the computer to create a visual observation system, repeating the observation activity of the human cortex. The idea of ​​a "computer neural network" - a program capable of simulating brain activity that makes connections between artificial neurons - has been spoken for over a decade. But Vicarious experts claim that they are bringing this idea to life with new steps based on methods developed at the present time, Technology Review said.

Project founder Dileep George, who was previously the technology director for the company Numenta, said that all developers of the "neuron link" are based on the "neocognitron" model that has been introduced. first released in 1980. Often these systems are taught to recognize visual images using random still images. According to George, Vicarious uses more complex architectures - computer training is done with video streams that change over time.

Vicarious hopes to develop and launch this system for sale within the next few years. Co-founder Scot Fenix ​​believes that the system can have many applications. For example, in health: New technology-equipped computers can analyze diagnostic images to determine if a patient has cancer. For example, a smartphone installed with this program may determine how many calories the plate of food in the photo contains."The visual-sensing computer system will recognize most types of human activities," he added.

Fenix ​​noted that Vicarious program has the ability to self-study by viewing the object images and forming criteria to identify common objects. That means that the system is smart enough to identify objects that information about it is incomplete. For example, it can recognize a hand even if the hand is covered by paint stains or partially obstructed by a watch.

Vicarious kept this technology secret, but their work still made investors excited. Last month, the company received $ 15 million in venture capital from a fund with the participation of Facebook co-founder Dustin Moskowitz.

According to Associate Professor Andrew Ng, head of Stanford University's Artificial Intelligence Laboratory, Vicarious's main problem may be inadequate computing power needed to accurately simulate cognitive processes. Andrew Ng used to participate in a Google project, where the software looked at millions of random screenshots of YouTube videos to detect which frames had cats. To ensure the program works, people have used 16,000 computers to form an artificial neural network system.