Quantum and Neural Networks

Martin Gärttner and Markus Oberthaler

Tuesdays, 16:15-18:00; INF 227 (KIP), SR 2.402. [LSF] 

All talks have been assigned. If you are interested in coming to some of the talks, you can register for the practice group (bottom of this page) in order to recieve emails concerning the preparatory reading.

Machine learning techniques, especially the ones building on deep neural networks, have been proven extremely useful for tasks like recognizing objects in images, finding clusters in high dimensional datasets, or optimizing stock market investment strategies. In recent years, these techniques have also found multiple applications in various areas of physics research. This seminar will give an overview of these recent developments with a strong bias towards topics in quantum physics. We will explore both the basics of the applied machine learning techniques and the background of the physics topics they have been applied to.

General literature:

• Physics today: Machine learning meets quantum physics
• A high-bias, low-variance introduction to Machine Learning for physicists, https://arxiv.org/abs/1803.08823 (also has demo python codes!)
• Machine learning and the physical sciences, https://arxiv.org/abs/1903.10563

List of topics:

A. Supervised learning:

1. [23.4. Adrian Braemer, slides] Introduction to supervised learning: Feed forward neural networks, CNNs and how to train them. Literature: A high-bias, low-variance introduction to Machine Learning for physicists, https://arxiv.org/abs/1803.08823, especially chapters IX and X.
2. [30.4. Daniel Kirchhoff, slides] Carrasquilla and Melko: Machine learning phases of matter, Nature Physics volume 13, pages 431–434 (2017), see also: Machine learning quantum phases of matter beyond the fermion sign problem, Scientific Reports 7, 8823 (2017)
3. [7.5. Nils Rörup slides] Discovering physical concepts with neural networks. Raban Iten, Tony Metger, Henrik Wilming, Lidia del Rio, Renato Renner. arXiv:1807.10300 [quant-ph]
4. [14.5. Moritz Reh slides and Patrick Jentsch slides] arXiv:1811.12425 [cond-mat.quant-gas], Classifying Snapshots of the Doped Hubbard Model with Machine Learning. Annabelle Bohrdt, Christie S. Chiu, Geoffrey Ji, Muqing Xu, Daniel Greif, Markus Greiner, Eugene Demler, Fabian Grusdt, Michael Knap

B. Unsupervised leaning:

1. [21.5. Felix Lübbe slides] Introduction to unsupervised learning and generative models: From restricted Boltzmann machines to more advanced models. Literature: A high-bias, low-variance introduction to Machine Learning for physicists, https://arxiv.org/abs/1803.08823, especially chapters XVI and XVII. Restricted Boltzmann Machines: Introduction and Review, https://arxiv.org/abs/1806.07066
2. [28.5. Alexander Wagner slides] Carleo and Troyer: Solving the quantum many-body problem with artificial neural networks. Science  10 Feb 2017, Vol. 355, Issue 6325, pp. 602-606
3. [4.6. Philipp Schultzen slides] Carrasquilla et al.: Reconstructing quantum states with generative models, arXiv:1810.10584
4. [11.6. Stephen Schaumann slides] Experimental Realization of a Quantum Autoencoder: The Compression of Qutrits via Machine Learning. Alex Pepper, Nora Tischler, and Geoff J. Pryde. Phys. Rev. Lett. 122, 060501 – Published 11 February 2019

C. Reinforcement learning:

1. [25.6. Nicholas Kiefer slides] Introduction to reinforcement learning. Literature: An Introduction to Deep Reinforcement Learning, https://arxiv.org/abs/1811.12560 Algorithms for Reinforcement Learning, https://sites.ualberta.ca/~szepesva/papers/RLAlgsInMDPs.pdf
2. [2.7. Robert Klassert slides] Reinforcement Learning in Different Phases of Quantum Control. Marin Bukov, Alexandre G. R. Day, Dries Sels, Phillip Weinberg, Anatoli Polkovnikov, and Pankaj Mehta, Phys. Rev. X 8, 031086
3. [9.7. Raphael Stock slides] Toward an AI Physicist for Unsupervised Learning, Tailin Wu, Max Tegmark, https://arxiv.org/abs/1810.10525
4. [16.7. Christoph Smaczny] Active learning machine learns to create new quantum experiments. Alexey A. Melnikov, Hendrik Poulsen Nautrup, Mario Krenn, Vedran Dunjko, Markus Tiersch, Anton Zeilinger, and Hans J. Briegel. PNAS February 6, 2018 115 (6) 1221-1226

We divided the topics into three blocks according to the three machine learning paradigms of supervised, unsupervised and reinforcement learning. Each block starts with a presentation that gives an overview of the machine learning techniques used in this block, which will be less directly physics related. However, we expect the speaker within each block to coordinate such that the relevant techniques for the other presentations are covered and redundancy is avoided. For the introductory talks it might be nice to also show demos of computer codes.

Each presenter should arrange a meeting with one of us at least 2 weeks prior to the talk to clarify the contents. In the week prior to the presentation a rehearsal talk will be scheduled.

Requirements for obtaining 6 CPs for the seminar: Oral presentation during one session of the seminar, as well as a written report/documentation on the topic of the talk. The report may contain up to 2000 words text plus figures if appropriate and should be handed in two weeks after the presentation at the latest.