The engineers over at Google have been busy with another interesting project in the elusive X lab. After self-driving cars, they now managed to develop a computer system that can learn without human supervision. By connecting over 16.000 processors, they created a network powerful enough to simulate some features of the human brain. Google developers then managed to create an algorithm that turned the powerful computing grid into a 'self-learning machine'. While that may be the first step in order to develop scary machines that are set to take over the world, it actually helps us to understand our own brains better.
Cats
In order to test whether their powerful computing grid was actually able to learn something on its own, the Google engineers gave it a task: try to recognize cats in a collection of 10 million pictures derived from YouTube videos. However, the network was not provided with any information that 'explained' what defines a cat. The system was supposed to figure that out by itself. And it worked: Google noted that their algorithm had recognized around 20.000 cats after being fed with random YouTube images.
Neurons
The ability to learn what certain objects represent is also hypothesized to exist in our brains. In the visual cortex, where images are processed, neurons are thought to be trained to recognize specific objects. Repetition helps us to learn and get better at recognizing things. While we do sometimes learn with supervision, for example our parents, neurons are supposed to be capable of learning autonomously as well.
Outlook
By simulating the way the brain works, we can do more sophisticated experiments that tell us something about how we function. It is easy to see the similarities between our brain and the algorithm Google developed for their computer network. Nevertheless, their system, despite having 16.000 processors and about a billion connections, is rather limited when compared to our own visual cortex. That also tells us that simulating the whole brain will be an even bigger ordeal. Developing a self-learning machine is still pretty interesting though.
Cats
In order to test whether their powerful computing grid was actually able to learn something on its own, the Google engineers gave it a task: try to recognize cats in a collection of 10 million pictures derived from YouTube videos. However, the network was not provided with any information that 'explained' what defines a cat. The system was supposed to figure that out by itself. And it worked: Google noted that their algorithm had recognized around 20.000 cats after being fed with random YouTube images.
Neurons
The ability to learn what certain objects represent is also hypothesized to exist in our brains. In the visual cortex, where images are processed, neurons are thought to be trained to recognize specific objects. Repetition helps us to learn and get better at recognizing things. While we do sometimes learn with supervision, for example our parents, neurons are supposed to be capable of learning autonomously as well.
Outlook
By simulating the way the brain works, we can do more sophisticated experiments that tell us something about how we function. It is easy to see the similarities between our brain and the algorithm Google developed for their computer network. Nevertheless, their system, despite having 16.000 processors and about a billion connections, is rather limited when compared to our own visual cortex. That also tells us that simulating the whole brain will be an even bigger ordeal. Developing a self-learning machine is still pretty interesting though.
 |
| Cats on YouTube are popular. |
No comments:
Post a Comment