Build machines with the intellectual capabilities of a human being aún is on track, but a team of scientists from the University of Tortonto (Canada), the Moore-Sloan Data Science at the University of New York (USA ) and MIT have managed to emulate the ability of people to acquire new concepts from a unique example thanks to a particular algorithm. The study was published in the journal Science.
The algorithm allows the system to self-program and use previously acquired knowledge to recognize and draw simple visual concepts like humans do. And for a computer, this option requires hundreds or maybe thousands of continuous directions to the precision that so simply acquire humans. The study â € œacorta the computational process of learning new concepts and expands the application of the machines to work more creativasâ €, explains Brenden Lake, leader of the work.
To achieve this milestone tecnológico for the first time in history, the researchers created a Bayesian learning software (BLP, Bayesian Learning Program), in which the concepts are represented as informáticos básicos programs. Unlike programs standard pattern recognition, this algorithm learns € â € œmodelos generativosâ of real processes, increasing their efficiency through better use of the data.
In the same way This algorithm learns to learn, for example, using knowledge of the Latin alphabet to learn the letters of the Greek alphabet. To prove it, the software was applied to more than 1,600 types of handwritten characters with multiple writing systems, including alphabets invented or nonexistent, as well as with human trials to participants through similar tests known Turing Test.
The results of visual tests revealed that the Turing model type BPL got the same level of performance as human beings and overcame more than any deep learning results obtained to date machines.
â € œAntes entering the daycare, the kids ± os and can recognize new concepts from a single example, and can imagine new examples have ever seen. We are still far from building machines as smart as a child ± o, but this is the first time we have achieved a computer to be able to learn and use a large number of real-world concepts, including concepts such as simple visual written characters Manoah €, Joshua Tenenbaum explains study co-author.
Just spiders are ± ando the surface, but one day will arrive in which machines They will be able to learn from their environment like people who do.
No comments:
Post a Comment