Intel envisions computers that learn

By Steve Malone

Posted on 9 Dec 2003 at 10:37

Researchers at Intel have unveiled software that claims to be able to allow computers to 'learn' from experience. By detecting how often something happened in the past, the programs are said to be able to allow a computer to predict future results.

The company has made the software available through its Open Source Machine Learning Library (OpenML).
The researchers have based OpenML on 'Bayesian' mathematics. This is based on the simple idea that the likelihood of future events can be based on how frequently they happened in the past. The more data that is collected the better the predictions. If the data changes, predictions will gradually change to reflect this.

If the software model works, the implications could be huge. OpenML could be applied to everything from genetic modelling to stock market trends to spotting and automating repetitive formatting tasks within a document. Not surprisingly, Intel says that OpenML has been made possible by the increasing power of modern processors to crunch huge amounts of data.

Intel researchers say they used OpenML to develop a 'lip-reading' program. Using a video camera to follow mouth movements it helped create a more accurate speech recognition program which might be of particular use in places with a lot of background noise.

The software consists of a set of functions that are aimed at helping developers create better machine learning applications. The open source software contains the C++ source code for all of the library's functions and a royalty-free redistribution license.

Intel made the announcement at the Neural Information Processing Systems Conference (NIPS2003).