Communications of the ACM
First Machine Learning Method Capable of Accurate Extrapolation
A new machine learning method uses observations made under safe conditions to accurately extrapolate all possible conditions shaped by the same physical dynamics.
Credit: Birgit Rieger/IST Austria
Researchers at the Institute of Science and Technology Austria and the Max Planck Institute (MPI) for Intelligent Systems in Germany have developed a machine learning method that uses observations made under safe conditions to accurately extrapolate all possible conditions shaped by the same physical dynamics.
MPI's Georg Martius says the method digests data from which it extracts the equations describing the underlying physics. The technique also yields more intuitive and understandable equations, and it ensures interpretability and optimization for physical situations because it is founded on a different type of framework than other machine-learning techniques.
Martius envisions this research being applied to a future scenario in which "[a] robot would experiment with different motions, then be able to use machine learning to uncover the equations that govern its body and movement, allowing it to avoid dangerous actions or situations."
From IST Austria
View Full Article
Abstracts Copyright © 2018 Information Inc., Bethesda, Maryland, USA
No entries found