# Quantum Optical Experiments Modeled by Long Short-Term Memory

^{1}

^{2}

^{3}

^{4}

^{5}

^{*}

^{†}

^{‡}

^{§}

## Abstract

**:**

## 1. Introduction

## 2. Methods

#### 2.1. Target Values

#### 2.2. Loss Function

#### 2.3. Network Architecture

## 3. Experiments

#### 3.1. Dataset

#### 3.2. Workflow

#### 3.3. Results

## 4. Outlook

## 5. Conclusions

## Author Contributions

## Funding

## Data Availability Statement

## Acknowledgments

## Conflicts of Interest

## References

- Esteva, A.; Kuprel, B.; Novoa, R.A.; Ko, J.; Swetter, S.M.; Blau, H.M.; Thrun, S. Dermatologist-level classification of skin cancer with deep neural networks. Nature
**2017**, 542, 115–118. [Google Scholar] [CrossRef] - Silver, D.; Schrittwieser, J.; Simonyan, K.; Antonoglou, I.; Huang, A.; Guez, A.; Hubert, T.; Baker, L.; Lai, M.; Bolton, A.; et al. Mastering the game of Go without human knowledge. Nature
**2017**, 550, 354–359. [Google Scholar] [CrossRef] - Hochreiter, S. Untersuchungen zu Dynamischen Neuronalen Netzen. Master’s Thesis, Technical University of Munich, Munich, Germany, 1991. [Google Scholar]
- Hochreiter, S.; Schmidhuber, J. Long short-term memory. Neural Comput.
**1997**, 9, 1735–1780. [Google Scholar] [CrossRef] - Sutskever, I.; Vinyals, O.; Le, Q.V. Sequence to sequence learning with neural networks. In Advances in Neural Information Processing Systems 27; Curran Associates, Inc.: San Jose, CA, USA, 2014; pp. 3104–3112. [Google Scholar]
- Shor, P.W. Scheme for reducing decoherence in quantum computer memory. Phys. Rev. A
**2000**, 52, R2493–R2496. [Google Scholar] [CrossRef] [PubMed] - Kaszlikowski, D.; Gnacínski, P.; Zukowski, M.; Miklaszewski, W.; Zeilinger, A. Violations of local realism by two entangled N-dimensional systems are stronger than for two qubits. Phys. Rev. Lett.
**2000**, 86, 4418–4421. [Google Scholar] [CrossRef] [PubMed][Green Version] - Krenn, M.; Malik, M.; Fickler, R.; Lapkiewicz, R.; Zeilinger, A. Automated Search for new Quantum Experiments. Phys. Rev. Lett.
**2016**, 116, 090405. [Google Scholar] [CrossRef] [PubMed][Green Version] - Malik, M.; Erhard, M.; Huber, M.; Krenn, M.; Fickler, R.; Zeilinger, A. Multi-photon entanglement in high dimensions. Nat. Photonics
**2016**, 10, 248–252. [Google Scholar] [CrossRef][Green Version] - Erhard, M.; Malik, M.; Krenn, M.; Zeilinger, A. Experimental GHZ entanglement beyond qubits. Nat. Photonics
**2018**, 12, 759–764. [Google Scholar] [CrossRef][Green Version] - Knott, P. A search algorithm for quantum state engineering and metrology. New J. Phys.
**2016**, 18, 073033. [Google Scholar] [CrossRef][Green Version] - Nichols, R.; Mineh, L.; Rubio, J.; Matthews, J.C.; Knott, P.A. Designing quantum experiments with a genetic algorithm. Quantum Sci. Technol.
**2019**, 4, 045012. [Google Scholar] [CrossRef][Green Version] - O’Driscoll, L.; Nichols, R.; Knott, P.A. A hybrid machine learning algorithm for designing quantum experiments. Quantum Mach. Intell.
**2019**, 1, 5–15. [Google Scholar] [CrossRef][Green Version] - Melnikov, A.A.; Nautrup, H.P.; Krenn, M.; Dunjko, V.; Tiersch, M.; Zeilinger, A.; Briegel, H.J. Active learning machine learns to create new quantum experiments. Proc. Natl. Acad. Sci. USA
**2018**, 115, 1221–1226. [Google Scholar] [CrossRef] [PubMed][Green Version] - Arrazola, J.M.; Bromley, T.R.; Izaac, J.; Myers, C.R.; Brádler, K.; Killoran, N. Machine learning method for state preparation and gate synthesis on photonic quantum computers. Quantum Sci. Technol.
**2019**, 4, 024004. [Google Scholar] [CrossRef][Green Version] - Zhan, X.; Wang, K.; Xiao, L.; Bian, Z.; Zhang, Y.; Sanders, B.C.; Zhang, C.; Xue, P. Experimental quantum cloning in a pseudo-unitary system. Phys. Rev. A
**2020**, 101, 010302. [Google Scholar] [CrossRef] - Krenn, M.; Kottmann, J.S.; Tischler, N.; Aspuru-Guzik, A. Conceptual understanding through efficient automated design of quantum optical experiments. Phys. Rev. X
**2021**, 11, 031044. [Google Scholar] [CrossRef] - Flam-Shepherd, D.; Wu, T.; Gu, X.; Cervera-Lierta, A.; Krenn, M.; Aspuru-Guzik, A. Learning Interpretable Representations of Entanglement in Quantum Optics Experiments using Deep Generative Models. arXiv
**2021**, arXiv:2109.02490. [Google Scholar] - Krenn, M.; Erhard, M.; Zeilinger, A. Computer-inspired quantum experiments. Nat. Rev. Phys.
**2020**, 2, 649–661. [Google Scholar] [CrossRef] - Mayr, A.; Klambauer, G.; Unterthiner, T.; Hochreiter, S. DeepTox: Toxicity Prediction using Deep Learning. Front. Environ. Sci.
**2016**, 3. [Google Scholar] [CrossRef][Green Version] - Mayr, A.; Klambauer, G.; Unterthiner, T.; Steijaert, M.; Wegner, J.K.; Ceulemans, H.; Clevert, D.A.; Hochreiter, S. Large-scale comparison of machine learning methods for drug target prediction on ChEMBL. Chem. Sci.
**2018**, 9, 5441–5451. [Google Scholar] [CrossRef][Green Version] - Yao, A.M.; Padgett, M.J. Orbital angular momentum: Origins, behavior and applications. Adv. Opt. Photonics
**2011**, 3, 161–204. [Google Scholar] [CrossRef][Green Version] - Erhard, M.; Fickler, R.; Krenn, M.; Zeilinger, A. Twisted photons: New quantum perspectives in high dimensions. Light. Sci. Appl.
**2018**, 7, 17146. [Google Scholar] [CrossRef] [PubMed][Green Version] - Huber, M.; de Vicente, J.I. Structure of multidimensional entanglement in multipartite systems. Phys. Rev. Lett.
**2013**, 110, 030501. [Google Scholar] [CrossRef] [PubMed][Green Version] - Huber, M.; Perarnau-Llobet, M.; de Vicente, J.I. Entropy vector formalism and the structure of multidimensional entanglement in multipartite systems. Phys. Rev. A
**2013**, 88, 042328. [Google Scholar] [CrossRef][Green Version] - Good, I.J. Rational Decisions. J. R. Stat. Soc. Ser. B
**1952**, 14, 107–114. [Google Scholar] [CrossRef] - Bishop, C.M. Pattern Recognition and Machine Learning, 5th ed.; Information Science and Statistics; Springer: Berlin/Heidelberg, Germany, 2007. [Google Scholar]
- Mikolov, T.; Chen, K.; Corrado, G.; Dean, J. Efficient estimation of word representations in vector space. arXiv
**2013**, arXiv:1301.3781. [Google Scholar] - Hastie, T.; Tibshirani, R.; Friedman, J.H. The Elements of Statistical Learning: Data Mining, Inference, and Prediction, 2nd ed.; Springer Series in Statistics; Springer: Berlin/Heidelberg, Germany, 2009. [Google Scholar] [CrossRef]
- Goodfellow, I.; Pouget-Abadie, J.; Mirza, M.; Xu, B.; Warde-Farley, D.; Ozair, S.; Courville, A.; Bengio, Y. Generative Adversarial Nets. In Advances in Neural Information Processing Systems 27; Curran Associates, Inc.: San Jose, CA, USA, 2014; pp. 2672–2680. [Google Scholar]
- Lowerre, B.T. The Harpy speech recognition system. Ph.D. Thesis, Carnegie Mellon University, Pittsburgh, PA, USA, 1976. [Google Scholar]
- Gulrajani, I.; Ahmed, F.; Arjovsky, M.; Dumoulin, V.; Courville, A.C. Improved Training of Wasserstein GANs. In Advances in Neural Information Processing Systems 30; Curran Associates, Inc.: San Jose, CA, USA, 2017; pp. 5767–5777. [Google Scholar]
- Yu, L.; Zhang, W.; Wang, J.; Yu, Y. SeqGAN: Sequence Generative Adversarial Nets with Policy Gradient. arXiv
**2016**, arXiv:1609.05473. [Google Scholar] - Fedus, W.; Goodfellow, I.; Dai, A.M. MaskGAN: Better Text Generation via Filling in the
_{------}. In Proceedings of the International Conference on Learning Representations, Vancouver, BC, Canada, 30 April–3 May 2018. [Google Scholar] - Bengio, S.; Vinyals, O.; Jaitly, N.; Shazeer, N. Scheduled Sampling for Sequence Prediction with Recurrent Neural Networks. In Advances in Neural Information Processing Systems 28; Curran Associates, Inc.: San Jose, CA, USA, 2015; pp. 1171–1179. [Google Scholar]
- Graves, A. Generating sequences with recurrent neural networks. arXiv
**2013**, arXiv:1308.0850. [Google Scholar] - Karpathy, A.; Li, F.-F. Deep visual-semantic alignments for generating image descriptions. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Boston, MA, USA, 7–15 June 2015; pp. 3128–3137. [Google Scholar]

**Figure 1.**Sequence processing model for a many-to-one mapping. The target value $\widehat{y}$ can be either an estimate for ${y}_{\mathrm{E}}$ (entanglement classification) or ${y}_{\mathrm{SRV}}$ (SRV regression).

**Figure 2.**Negative and positive samples in the data set as a function of the leading Schmidt rank n.

**Figure 3.**Workflow. We split the entire data by their leading Schmidt rank n. All samples with $n\ge 9$ constitute the extrapolation set, which we use to explore the out-of-distribution capabilities of our model. For the remaining samples (i.e., $n<9$), we make a random test split at a ratio of $1/4$. The test set is used to estimate the conventional generalization error of our model. We use the training set to perform cluster cross validation.

**Figure 4.**True negative rate (TNR), true positive rate (TPR), rediscovery ratio of the LSTM network using cluster cross validation for different folds 0–8. True negative rates are high for all validation folds. All metrics are good for the extrapolation set 9–12, demonstrating that the models perform well on data beyond the training set distribution, covering only leading Schmidt rank numbers 0–8. Error bars represent $95\phantom{\rule{0.166667em}{0ex}}\%$ binomial proportion confidence intervals.

**Figure 5.**(

**a**) True negative rate, (

**b**) true positive rate, (

**c**) rediscovery ratio, and (

**d**) precision for the extrapolation set 9–12 for varying sigmoid threshold $\tau $ and SRV radius r. For too restrictive parameter choices ($\tau \to 1$ and $r\to 0.5$), the TNR and precision approach the value 1, while TPR and rediscovery ratio approach 0, such that no interesting new setups would be identified. For too loose choices (small $\tau $, large r), too few negative samples would be rejected, such that the advantage over random search becomes negligible, reflected in smaller precision values. Hence, there is a trade-off between rediscovery ratio (diversity of discoveries) and precision (speed of discoveries). For a large variety of $\tau $ and r, the models perform satisfyingly well, allowing a decent compromise between TNR and TPR. This is also reflected by a value of 0.64 for the mean average precision, where the mean is taken over $r=0.5$ to $r=7$ with a step size of 0.1, and the average precision is over $\tau =1$ to $\tau =0$, with a step size of 0.01 for each value of r.

**Table 1.**Cluster cross validation folds (0–8) and extrapolation set (9–12) characterized by leading Schmidt rank n. Samples with $n=0$ and samples with $n=1$ are combined and then split into two folds (0,1) at random.

0,1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9–12 |

0,1 |

**Table 2.**Conventional in-distribution training and test errors. The test set consists of 20 % of the data. Performance on predicting the entanglement is measured using the BCE loss, TNR, and TPR. Performance on predicting the SRV is measured using the SRV loss according to Equation (2), SRV accuracy, and the mean distance between true SRV and predicted SRV.

Training | Test | |
---|---|---|

BCE loss | 10.2 | 10.4 |

TNR | 0.9271 ± 2.4 $\times {10}^{-4}$ | 0.9261 ± 3.8 $\times {10}^{-4}$ |

TPR | 0.9469 ± 4.1 $\times {10}^{-4}$ | 0.9427 ± 6.5 $\times {10}^{-4}$ |

SRV loss | 2.247 | 2.24 |

SRV accuracy | 0.9382 | 0.938 |

SRV mean distance | 1.3943 | 1.4 |

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |

© 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Adler, T.; Erhard, M.; Krenn, M.; Brandstetter, J.; Kofler, J.; Hochreiter, S.
Quantum Optical Experiments Modeled by Long Short-Term Memory. *Photonics* **2021**, *8*, 535.
https://doi.org/10.3390/photonics8120535

**AMA Style**

Adler T, Erhard M, Krenn M, Brandstetter J, Kofler J, Hochreiter S.
Quantum Optical Experiments Modeled by Long Short-Term Memory. *Photonics*. 2021; 8(12):535.
https://doi.org/10.3390/photonics8120535

**Chicago/Turabian Style**

Adler, Thomas, Manuel Erhard, Mario Krenn, Johannes Brandstetter, Johannes Kofler, and Sepp Hochreiter.
2021. "Quantum Optical Experiments Modeled by Long Short-Term Memory" *Photonics* 8, no. 12: 535.
https://doi.org/10.3390/photonics8120535