Communications of the ACM
AI Computer Program Transforms Typed Text Into Handwriting
Credit: The Guardian
A new computer program developed by the University of Toronto's Alex Graves applies his work on recurrent neural networks (RNNs) to convert typed text into organic-like handwriting. The program also enables the user to adjust the bias setting to shape the legibility and style of the handwriting it generates.
Graves' program and similar long short-term memory RNNs combine individual data points into complex sequences based on probability predictions. The RNN samples from its own output, feeding each data point back into the algorithm as a fresh input. Many neural networks that depend on simpler learning via trial and error can only produce the same outcome patterns once any choice or string of choices the network makes is rated as successful. However, RNNs seldom generate the same solution more than once, because they synthesize data in an intricate input and output system that analyzes data and produces results point-by-point.
Graves implies RNNs are superior to other artificial intelligence models, as the RNN system is more capable of modeling complex, multivariate data.
From Techly
View Full Article
Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA
No entries found