Communications of the ACM
The Neural Network That Remembers
Researchers at the University of California, San Diego have developed a recurrent neural network that writes its own contextually relevant reviews of beers.
Credit: John Hersey
Zachary C. Lipton and colleagues at the University of California, San Diego (UCSD) say they have developed a recurrent neural network that can mine patterns in reviews from articles on BeerAdvocate.com and write its own contextually relevant reviews.
The work is part of a growing body of research demonstrating the language-processing capabilities of the artificial-intelligence based tool.
Recurrent neural networks have achieved other feats, including captioning images, translating foreign languages, and answering email messages. The neural network developed at UCSD adds another set of connections between the artificial neurons that govern short-term memory.
However, there may be some fundamental limitations to the approach of teaching a machine to learn through imitating humans. One promising way forward might be reinforcement learning, in which a computer explores the possible actions it can take, guided by some sort of reward signal. Google DeepMind researchers used this approach in a system that can beat human players at 31 different video games.
From IEEE Spectrum
View Full Article
Abstracts Copyright © 2016 Information Inc., Bethesda, Maryland, USA
No entries found