A world with robots that reason or study remains only for sci-fi movies, but certainly it is already in that direction. For the first time in history, an interdisciplinary team of researchers created software that is capable of encoding different symbols independently from what you already know.
A group of scientists from the Massachusetts Institute of Technology ( MIT), University of New York (USA) and the University of Toronto (Canada) got a machine could imitate a basic human behavior. acquire knowledge
The scientific Brenden Lake, labor leader, told Science magazine that this made an algorithm that allows “shorten the computational process of learning new concepts and expand the application of machines to more creative tasks.”
Among the basic functions also is the possibility of self-programming and thus the machine can use the knowledge acquired beforehand, a form of proto-reading used to recognize and extract visual concepts simple.
The Bayesian learning software (BLP, Bayesian Program Learning) if it differs from the heretofore known basic software, because instead of recognizing standard patterns, you can learn generative models of real processes, allowing better use of data .
By the Turing test-the test of the ability of a machine to display a similar people- they could verify that the machine has the ability to learn to learn intelligent behavior.
“Before entering kindergarten, children already know how to recognize new concepts from a one example, and you can imagine things that have never seen. Are far from building as intelligent as a child machine yet, “said Joshua Tenenbaum, author of the study.
” This is the first time we have achieved a computer to be able to learn and use a large number of concepts from the real world, including simple visual concepts such as handwritten characters, “he said.
For example, they used their” knowledge “of the Latin alphabet to apply it and decode the Greek. Thus, the software recognized more than 1,600 types of handwritten characters with multiple writing systems, including alphabets invented or nonexistent.
The tests revealed that the BPL model had the same level of performance that people and far exceeded any deep learning result obtained so far.
No comments:
Post a Comment