Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

REINFORCEMENT LEARNING

Solving optimization tasks in condensed matter

Nature Machine Intelligence volume 2, pages 557–558 (2020)Cite this article

Subjects

Finding states of matter with properties that are just right is a main challenge from metallurgy to quantum computing. A data-driven optimization approach based on gaming strategies could help.

As an example of an optimization challenge, consider the famous travelling salesperson problem: a salesperson has a number of cities to visit, which are randomly placed on a map. The question is in which order to visit the cities to minimize the travel time. It is hard to know if the best solution has been found as this requires comparing all the possible paths. This problem is closely related to finding ground states of spins on an atomic lattice where it is possible to find good candidate states with low energy, but checking that a given state has indeed the lowest possible energy requires comparing all the options. Artificial intelligence and machine learning have both been part of statistical physics and computer science for decades. Relatively recently, the technology has progressed into the mainstream and we can now take practical advantage of these abstract models. One method that has been particularly successful is reinforcement learning, where an agent learns to formulate a policy for future actions based on input data and a reward defined by the user. This method is similar to how a human learns to play a board game: at first the human may try to randomly move the pieces on the board around, but they will soon realize which moves are good and which moves are disadvantageous. They then build on the good moves to make the next moves and so on. Reinforcement learning with artificial neural networks is designed to mimic this learning process. A prominent example of such algorithms is AlphaGo2, where the algorithm famously learned to outperform human players in the game of Go.

This is a preview of subscription content, access via your institution

Access options

Rent or buy this article

Get just this article for as long as you need it

$39.95

Learn more

Prices may be subject to local taxes which are calculated during checkout

Fig. 1: Machine learning.

References

  1. Mills, K., Ronagh, P. & Tamblyn, I. Nat. Mach. Intell. https://doi.org/10.1038/s42256-020-0226-x (2020).

  2. Silver, D. et al. Nature 550, 354–359 (2017).

    Article  Google Scholar 

  3. Carrasquilla, J. & Melko, R. G. Nat. Phys. 13, 431–434 (2017).

    Article  Google Scholar 

  4. Zhang, Y. et al. Nature 570, 484–490 (2019).

    Article  Google Scholar 

  5. Bukov, M. et al. Phys. Rev. X 8, 031086 (2018).

    Google Scholar 

  6. Sweke, R., Kesselring, M. S., van Nieuwenburg, E. P. L. & Eisert, J. Preprint at https://arxiv.org/abs/1810.07207 (2018).

  7. Brockman, G. et al. Preprint at https://arxiv.org/abs/1606.01540 (2016).

Download references

Author information

Authors and Affiliations

  1. Institute for Theoretical Physics, ETH Zurich, Zurich, Switzerland

    Eliska Greplova

  2. Kavli Institute of Nanoscience, Delft University of Technology, Delft, the Netherlands

    Eliska Greplova

Authors
  1. Eliska Greplova
    View author publications

    You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to Eliska Greplova.

Rights and permissions

Reprints and Permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Greplova, E. Solving optimization tasks in condensed matter. Nat Mach Intell 2, 557–558 (2020). https://doi.org/10.1038/s42256-020-00240-8

Download citation

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1038/s42256-020-00240-8

Search

Advanced search

Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing