Researchers at ETH Zurich in Switzerland say they have refined machine learning by deliberately misleading intelligent machines, enabling a new form of computerized data categorization.
The method pioneered by ETH Zurich professor Sebastian Huber not only permits categorization of any data, but also notes whether complex datasets contain categories at all.
The researchers applied the technique to a many-body system of interacting magnetic dipoles that never attains equilibrium. To pinpoint the boundary between systems that reach equilibrium and those that do not, the team took data from quantum systems to set up an arbitrary boundary, and then misled a neural network into thinking one group of data obtained equilibrium while the other did not.
The researchers demonstrated this data-sorting performance relies on the location of the boundary, and they determined the boundary's location based on the network's highest sorting performance.
From ETH Zurich
View Full Article
Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA
No entries found