A group of scientists has developed a new computer model that mimics the ability of humans to learn new concepts, “one small step” more in the field of artificial intelligence, according to a study published today in the journal Science.
“What we want is to try to reduce the gap between learning abilities of humans and machines (…). And discover why humans are so good at in generalizing concepts, “said Joshua Tenenbaum, one of those responsible for the investigation, the Department of Cognitive Sciences at the Massachusetts Institute of Technology (MIT) in the US.
According to the study, the main virtue of human beings is their “speed” and “diversity” when learning new concepts and apply them in new situations.
“computers find it hard to generalize from individual samples,” said Brenden Lake, New York University and senior author on a conference call to present the study.
The researchers focused on learning handwritten characters from different scripts and developed an algorithm that allow generalizations from a few examples.
“The computer does not have a program that applies to every situation, but rather a complex program of various learning programs, to suit every circumstance,” he said Tenenbaum in same conference.
When comparing the capacity of these computers face when learning tasks, including generating from examples of characters seen only on a few occasions, with other computers and loved human, they found how outperformed their peers and matched to humans.
In many cases, the results of human cognitive and this new model, called “Bayesian Learning Program,” were “virtually indistinguishable”.
“In artificial intelligence no major findings. There are a number of good ideas that work. This is another one, it’s a small step,” said Lake.
For the researcher, proving “our work is that the principles of composition, and learn to understand causality will be critical to advance the capabilities of the machines.”
No comments:
Post a Comment