Monday 11 January 2016

A Learning Advance in Artificial Intelligence Rivals Human Abilities


Artificial Intelligence Surpassed Human Competences

Computer researchers had recently reported that artificial intelligence had surpassed human competences for a narrow set of vision related task. These developments are remarkable since the so-called machine vision methods are being a commonplace in various characteristics of life comprising of car-safety methods which tend to identify pedestrians and bicyclists and in video game controls, Internet search as well as factory robots.

 Researchers from the Massachusetts Institute of Technology, New York University together with the University of Toronto have recently reported a new kind of `one shot’ machine learning in the journal Science, wherein a computer vision program seemed to beat a group of humans in identifying handwritten characters founded on a single example. The program seems to have the ability of learning quickly the characters in a variety of languages as well as in generalizing from what it has learned.

The authors recommend that this ability is the same wherein humans tend to learn and understand perceptions. Bayesian Program Learning or B.P. L as the new approach is known is unlike the present machine learning technologies known as deep neural networks. Neural networks can be trained in recognizing human speech, identify objects in images or detect types of behaviour on being exposed to large sets of examples

Bayesian Approach

Though these networks may have been modelled on the behaviour of biological neurons, they have not yet learned the way human tend to do, in quickly acquiring new concepts. In comparison, the new software program defined in the Science article has the capabilities of recognizing handwritten characters on `seeing’ only a few or a single example.

The researchers had compared the capabilities of their Bayesian approach as well as other programming models utilising five separate learning tasks which involved a set of characters from a research set. This was known as Onmiglot which comprised of 1,623 handwritten characters sets from 50 languages.

 The images as well as the pen stokes that were need to create characters were taken. Joshua B, Tenenbaum, professor of cognitive science and computation at M.I.T. together with one of the authors of the Science paper had commented that `with all the progress in machine learning, it is amazing what one can do with lots of data and faster computers. But when one looks at children, it is amazing what they can learn from very little data and some come from prior knowledge and some is built in the brain’.

Imagenet Large Scale Visual Recognition Challenge

Moreover, the organizers of an annual academic machine vision competition also reported gains in lowering the error rate in software for locating and classifying objects in digital images. Alexander Berg, an assistant professor of computer science at the University of North Carolina, Chapel Hill had stated that he was amazed by the rate of progress in the field.

The competition which is known as the Imagenet Large Scale Visual Recognition Challenge pits the teams of researchers at government, academic as well as corporate laboratories against one another in designing programs in classifying as well as detecting objects. The same was won by a group of researchers at the Microsoft Research laboratory in Beijing, this year.

No comments:

Post a Comment

Note: only a member of this blog may post a comment.