Targeted machine learning is too much trouble? Turing Learning Learned by Observing Only

Researchers at the University of Sheffield in the United Kingdom published their research results in the Swarm Intelligence journal recently. Now that the machine does not require human guidance, it can learn the workings of natural or artificial systems only through observation. This may greatly promote the development of technologies such as machine prediction of human behavior.

The experiment in the article draws inspiration from the famous Turing test. That is, the subjects and the machines in the two rooms are talking to each other. If the subjects can't tell who is the machine and who is the person, they can judge the machine. Passed the test, with the intelligence of humans.

Dr. Roderich Gross of the Department of Automation, Control and Systems Engineering at the University of Sheffield said: “Our research used Turing tests to reveal how a given system works. We first placed a group of supervised robots in order to find them in motion. Laws, we placed another group of learning robots under supervision. We recorded the movements of all these robots and then showed the movement data to the subjects.”

“But unlike the Turing test, our subjects are not humans, but are computer programs that can learn on their own initiative. Their task is to distinguish the two robots. If they correctly distinguish the motion data of two robots, they will In contrast, the 'learners' who imitate the first group of robots will score."

Dr. Roderich Gross called this method “Turing Learning,” with the advantage that humans no longer need to tell the machine what to look for.

“For example, if you want a robot to paint like Picasso, traditional machine learning algorithms will score points based on the similarities between robot and Picasso paintings. But someone must tell the algorithm first what factors should be considered to be similar. Turing learning does not require such prior knowledge. If the subject thinks that the robot painting is original, the robot will score. Turing learning can learn how to judge and draw at the same time.”

Dr. Roderich Gross believes that Turing Learning will promote the development of science and technology. He said: "Scientists can use it to discover the laws of natural or artificial systems, especially those that cannot be easily categorized by similarity indicators."

"Like computer games, you can use Turing Learning to match the reality. Game characters can observe and acquire the character characteristics of human players. They will not simply copy the observed behavior, but they will reveal what makes human players different. ."

This finding can also be used to create an algorithm for detecting perverted behavior, which can also be useful for livestock health monitoring and preventive maintenance of machines, cars, and airplanes. Turing Learning can also be used for security applications such as lie detection or online authentication.

Up to now, Dr. Roderich Gross and his team have tested Turing Learning only on cluster robots, and they plan to use it next to reveal how animal colonies, such as schools of fish or bees, work. This allows people to better understand what factors influence the behavior of these animals and ultimately apply them in the policy development of protecting these animals.

Via sheffield

Recommended reading:

Philosophy of Intelligence: How to determine if a machine is artificial intelligence?

In the face of the difficult Turing test, the chat robot began to expose itself

Safety Light Curtain

Safety Light Curtain,Safety Curtain,Laser Safety Light Curtain,Safety Optic Light Curtain,Security Light Curtain,Press Brake Safety Light Curtains

Jining KeLi Photoelectronic Industrial Co.,Ltd , https://www.sdkelien.com