A few months ago I was traveling in Japan and one of the things that excited me was to see live the future. Yes, in Tokyo Miraikan museum you discover the great technological advances that will govern the near future. One of the things I liked most was to see live display intelligent robot Asimo, Honda.
Just like that as a person walked out and performed several exercises that were answered immediately with a standing ovation. With the mouth we were all open. A robot was before our eyes running, jumping, scoring goals and interacting with the public. Amazing. In its official website you can know everything that is capable of Asimo.
How robots learn?
The technology applied to the development of intelligent machines are evolving rapidly and sometimes gives up a little scary to think how far they can get. How robots learn? Basically what is the time to train robots is applying machine learning techniques to get the robots themselves extract the information for themselves from a data provided. It is to learn as humans do. To achieve learn, learning various strategies in which once again the psychology is closely followed.
Machine learning strategies.
Surely Pavlov’s experiments with dogs will sound. This behavioral physiologist was a pioneer of classical conditioning and ideas are applied today to train robots using negative and positive stimuli. It seems simple. The algorithms implemented in the machine are designed to maximize the reward. When the robot receives instructions initials self-taught software improves the performance of a task in which the prize is to achieve a marked target (eg winning chess). Each time you learn a good move and avoid future errors.
2. Deep Learning.
Human neural networks are structured in different layers of which is collecting the information that feeds the next layer, and so on. Well, in deep learning it is to emulate the human brain on the machine creating artificial neural networks. The first layer of these ciberneuronas obtain the basic data and progressively will analyze in detail. For example, to recognize faces first recognize colors, shapes and then gradually more concrete details of each face. Like humans.
3. Decision Tree.
By using schemes with various options with their corresponding solutions it will solve a more complex problem within a specific scenario. It works like the typical manual will help discarding problems. We have all heard the “Reboot the router. Did this fix the problem? Yes / No” and from the response is linked to one or another scheme to solve the problem. The machine algorithm associates each situation to a reaction and applies logic to solve it with the best. And learn.
4. Optimization software.
In this case it is to train a software to resolve a task in the most efficient way possible regardless of the method you use. This is what the robots when they fall to learn to rise efficiently. Very similar to what a young child when he begins to psicomotrizmente develop, devotes much time to move and to “practice” the best way to move to gradually go using that acquired to improve and plan movements more psychomotor skills knowledge and speed .
>> Related article: Psychomotor Skills: learning phases
What’s next learn to make robots? Artificial intelligence is no longer just a thing of the future.