Communications of the ACM
Active Learning Model for Computer Predictions
University of Wisconsin-Madison professor Rob Nowak has been exploring an active learning model of computing, in which the machine receives all of the data up front. At first, the data has no labels and the machine makes very poor predictions, improving as a human expert supplies labels for some of the data.
Nowak and his student Kevin Jamieson applied this principle to an iOS app that can predict which craft beers a user will prefer. The similarities between data points were based on flavor, color, taste, and other characteristics defined by the terms used to describe beers in reviews. The app can use that data to find the closest match for beers the user might like. In addition, finer point comparisons offer the algorithm more reliable data to improve its categorizations and predictions over time.
Nowak says their process enables computers to process data much faster, because they require less human help to categorize the data. He notes the algorithm's efficiency makes a bigger difference as data sets get larger and human labor cannot keep pace.
From University of Wisconsin-Madison
View Full Article
Abstracts Copyright © 2013 Information Inc., Bethesda, Maryland, USA
No entries found