Forum for Science, Industry and Business

Sponsored by:     3M 
Search our Site:

 

First machine learning method capable of accurate extrapolation

13.07.2018

Scientists develop new machine learning method that can make robots safer.
New method provides simpler and more intuitive models of physical situations.

Understanding how a robot will react under different conditions is essential to guaranteeing its safe operation. But how do you know what will break a robot without actually damaging it?


A robot needs to learn about his body and the environment. It tries a few different motions and uses the algorithm. It can then predict what will happen with larger movements and at higher speeds.

IST Austria/Birgit Rieger

A new method developed by scientists at the Institute of Science and Technology Austria (IST Austria) and the Max Planck Institute for Intelligent Systems (MPI for Intelligent Systems) is the first machine learning method that can use observations made under safe conditions to make accurate predictions for all possible conditions governed by the same physical dynamics.

Especially designed for real-life situations, their method provides simple, interpretable descriptions of the underlying physics. The researchers will present their findings tomorrow at this year’s prestigious International Conference for Machine Learning (ICML).

In the past, machine learning was only capable of interpolating data—making predictions about situations that are “between” other, known situations. It was incapable of extrapolating—making predictions about situations outside of the known—because it learns to fit the known data as closely as possible locally, regardless of how it performs outside of these situations. In addition, collecting sufficient data for effective interpolation is both time- and resource-intensive, and requires data from extreme or dangerous situations.

But now, Georg Martius, former ISTFELLOW and IST Austria postdoc, and since 2017 a group leader at MPI for Intelligent Systems in Tübingen, Subham S. Sahoo, a PhD student also at MPI for Intelligent Systems, and Christoph Lampert, professor at IST Austria, developed a new machine learning method that addresses these problems, and is the first machine learning method to accurately extrapolate to unseen situations.

The key feature of the new method is that it strives to reveal the true dynamics of the situation: it takes in data and returns the equations that describe the underlying physics. “If you know those equations,” says Georg Martius, “then you can say what will happen in all situations, even if you haven’t seen them.” In other words, this is what allows the method to extrapolate reliably, making it unique among machine learning methods.

The team’s method sets itself apart in several other ways as well. First, the final approximations previously produced during machine learning were far too complex for a human to understand or work with. In the new method, the resulting equations are far simpler: “Our method’s equations are something you would see in a textbook—simple and intuitive,” says Christoph Lampert. The latter is another key difference: other machine learning methods give no insight into the relationship between conditions and results—and thus, no intuition on whether the model is even plausible.

“In every other area of research, we expect models that make physical sense, that tell us why,” adds Lampert. “This is what we should expect from machine learning, and what our method provides.” Finally, in order to guarantee interpretability and optimize for physical situations, the team based their learning method on a different type of framework. This new design is simpler than previous methods, which in practice means that less data is needed to give the same or even better results.

And it’s not all theory: “In my group, we’re actually working on developing a robot that uses this type of learning. In the future, the 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,” adds Martius. While robots are one active area of research, the method can be used with any type of data, from biological systems to X-ray transition energies, and can also be incorporated into larger machine learning networks.

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No 291734.

The research received funding from the ISTFELLOW program, a Marie Skłodowska-Curie COFUND grant co-funded by IST Austria and the European Union through the Horizon 2020 research and innovation programme. This program has since been succeeded by another COFUND grant, the ISTplus program, which is open for applications from qualified postdocs all over the world.

About IST Austria
The Institute of Science and Technology (IST Austria) is a PhD-granting research institution located in Klosterneuburg, 18 km from the center of Vienna, Austria. Inaugurated in 2009, the Institute is dedicated to basic research in the natural and mathematical sciences. IST Austria employs professors on a tenure-track system, postdoctoral fellows, and doctoral students. While dedicated to the principle of curiosity-driven research, the Institute owns the rights to all scientific discoveries and is committed to promote their use. The first president of IST Austria is Thomas A. Henzinger, a leading computer scientist and former professor at the University of California in Berkeley, USA, and the EPFL in Lausanne, Switzerland. The graduate school of IST Austria offers fully-funded PhD positions to highly qualified candidates with a bachelor's or master's degree in biology, neuroscience, mathematics, computer science, physics, and related areas. http://www.ist.ac.at

About MPI-IS
At the Max Planck Institute for Intelligent Systems we aim to understand the principles of Perception, Action and Learning in Intelligent Systems.
The Max-Planck-Institute for Intelligent Systems is located in two cities: Stuttgart and Tübingen. Research at the Stuttgart site of the Max Planck Institute for Intelligent Systems covers small-scale robotics, self-organization, haptic perception, bio-inspired systems, medical robotics, and physical intelligence. The Tübingen site of the institute concentrates on machine learning, computer vision, robotics, control, and the theory of intelligence.

Wissenschaftliche Ansprechpartner:

Georg Martius: gmartius@tuebingen.mpg.de
Christoph Lampert: chl@ist.ac.at

Originalpublikation:

S. S. Sahoo, C. H. Lampert, and G. Martius: "Learning equations for extrapolation and control". In Proceedings of the 35th International Conference on Machine Learning, ICML 2018, Stockholm, Sweden, 2018. PMLR, 2018.
http://proceedings.mlr.press/v80/sahoo18a.html
Arxiv Preprint: https://arxiv.org/abs/1806.07259

Weitere Informationen:

https://icml.cc/ Link to conference
http://cvml.ist.ac.at/ Link to Prof. Lampert's research group

Dr. Elisabeth Guggenberger | idw - Informationsdienst Wissenschaft

Further reports about: Intelligent Systems Learning MPI Max Planck Institute machine learning

More articles from Information Technology:

nachricht Accelerating quantum technologies with materials processing at the atomic scale
15.05.2019 | University of Oxford

nachricht A step towards probabilistic computing
15.05.2019 | University of Konstanz

All articles from Information Technology >>>

The most recent press releases about innovation >>>

Die letzten 5 Focus-News des innovations-reports im Überblick:

Im Focus: The geometry of an electron determined for the first time

Physicists at the University of Basel are able to show for the first time how a single electron looks in an artificial atom. A newly developed method enables them to show the probability of an electron being present in a space. This allows improved control of electron spins, which could serve as the smallest information unit in a future quantum computer. The experiments were published in Physical Review Letters and the related theory in Physical Review B.

The spin of an electron is a promising candidate for use as the smallest information unit (qubit) of a quantum computer. Controlling and switching this spin or...

Im Focus: Self-repairing batteries

UTokyo engineers develop a way to create high-capacity long-life batteries

Engineers at the University of Tokyo continually pioneer new ways to improve battery technology. Professor Atsuo Yamada and his team recently developed a...

Im Focus: Quantum Cloud Computing with Self-Check

With a quantum coprocessor in the cloud, physicists from Innsbruck, Austria, open the door to the simulation of previously unsolvable problems in chemistry, materials research or high-energy physics. The research groups led by Rainer Blatt and Peter Zoller report in the journal Nature how they simulated particle physics phenomena on 20 quantum bits and how the quantum simulator self-verified the result for the first time.

Many scientists are currently working on investigating how quantum advantage can be exploited on hardware already available today. Three years ago, physicists...

Im Focus: Accelerating quantum technologies with materials processing at the atomic scale

'Quantum technologies' utilise the unique phenomena of quantum superposition and entanglement to encode and process information, with potentially profound benefits to a wide range of information technologies from communications to sensing and computing.

However a major challenge in developing these technologies is that the quantum phenomena are very fragile, and only a handful of physical systems have been...

Im Focus: A step towards probabilistic computing

Working group led by physicist Professor Ulrich Nowak at the University of Konstanz, in collaboration with a team of physicists from Johannes Gutenberg University Mainz, demonstrates how skyrmions can be used for the computer concepts of the future

When it comes to performing a calculation destined to arrive at an exact result, humans are hopelessly inferior to the computer. In other areas, humans are...

All Focus news of the innovation-report >>>

Anzeige

Anzeige

VideoLinks
Industry & Economy
Event News

SEMANTiCS 2019 brings together industry leaders and data scientists in Karlsruhe

29.04.2019 | Event News

Revered mathematicians and computer scientists converge with 200 young researchers in Heidelberg!

17.04.2019 | Event News

First dust conference in the Central Asian part of the earth’s dust belt

15.04.2019 | Event News

 
Latest News

Plumbene, graphene's latest cousin, realized on the 'nano water cube'

23.05.2019 | Materials Sciences

New flatland material: Physicists obtain quasi-2D gold

23.05.2019 | Materials Sciences

New Boost for ToCoTronics

23.05.2019 | Physics and Astronomy

VideoLinks
Science & Research
Overview of more VideoLinks >>>