在2018年的時候，Koji Hashimoto教授發表了一篇期刊[1]，在期刊中他們用深度神經網絡(DNN)的結構來建構一個模型與AdS/CFT對偶性質做連結。在這篇論文中我們將以重建他們的模型為出發點，並討論在原模型下產生的諸多問題；接著在第三章節中，為了解決這些問題我們嘗試利用其他機械學習的模型來建構新的學習架構，在這個架構下我們期望能擺脫使用負面資料(negative data；因為我們發現這些資料並不能在實際面上被使用)，因此使用強化學習(RL)的方式並以其他近似函數來作配合(深度神經網絡(DNN)、神經微分方程(Neural ODE)、或其他近似函數)。然而從一直以來的結果中我們發現:在這個問題框架下會有複數對應解的問題，也因此我們在後記中在不同的兩個層面上討論對未來發展上的改進。

In 2018 [1], Koji Hashimoto had presented a deep-neural-like model to connect with AdS/CFT correspondence. We tried to reconstruct his model, and found some problems about uncertainty. Therefore, we attempted to use other learning models to solve these problems. The alternate models consist of using the concepts from reinforcement learning, Neural ODE, and Deep Neural Network. For our goal, we expect to keep from using the negative data in learning because the acquisition will come across problems in experimental. However, the result shows that the problem is actually about the uniqueness of solution, and we provide further discussion and improvement.

[1] Koji Hashimoto, Sotaro Sugishita, Akinori Tanaka, and Akio Tomiya. Deep learning and the AdS/CFT
correspondence. Phys. Rev. D, 98:046019, Aug 2018.
[2] Juan Maldacena. International Journal of Theoretical Physics, 38(4):1113:1133, 1999.
[3] Alfonso V. Ramallo. Introduction to the ads/cft correspondence, 2013.
[4] Russell D. Reed and Robert J. Marks. Neural Smithing: Supervised Learning in Feedforward Articial
Neural Networks. MIT Press, 1999.
[5] Viren Jain, Joseph F. Murray, Fabian Roth, Srinivas Turaga, Valentin Zhigulin, Kevin L. Briggman,
Moritz N. Helmstaedter, Winfried Denk, and H. Sebastian Seung. Supervised learning of image restoration with convolutional networks. In 2007 IEEE 11th International Conference on Computer Vision, pages 1-8, 2007.
[6] H. B. Barlow. Unsupervised learning. Neural Computation, 1:295-311, 1989.
[7] Richard S. Sutton and Andrew G. Barto. Reinforcement Learning: An Introduction. The MIT Press,
second edition, 2018.
[8] Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization, 2017.
[9] H. Robbins and S. Monro. A stochastic approximation method. Annals of Mathematical Statistics, 22:400-407, 1951.
[10] Marvin Minsky and Seymour Papert. Perceptrons: An Introduction to Computational Geometry. MIT Press, Cambridge, MA, USA, 1969.
[11] Vincent Francois-Lavet, Peter Henderson, Riashat Islam, Marc G. Bellemare, and Joelle Pineau. An
introduction to deep reinforcement learning. CoRR, abs/1811.12560, 2018.
[12] Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Andrei A. Rusu, Joel Veness, Marc G. Bellemare,
Alex Graves, Martin Riedmiller, Andreas K. Fidjeland, Georg Ostrovski, Stig Petersen, Charles Beattie,
Amir Sadik, Ioannis Antonoglou, Helen King, Dharshan Kumaran, Daan Wierstra, Shane Legg, and Demis
Hassabis. Human-level control through deep reinforcement learning. Nature, 518(7540):529-533, Feb 2015.
[13] Jianzhun Du, Joseph Futoma, and Finale Doshi-Velez. Model-based reinforcement learning for semi-markov
decision processes with neural odes. CoRR, abs/2006.16210, 2020.
[14] Tian Qi Chen, Yulia Rubanova, Jesse Bettencourt, and David Duvenaud. Neural ordinary dierential
equations. CoRR, abs/1806.07366, 2018.
[15] Patrick Kidger, James Morrill, James Foster, and Terry Lyons. Neural controlled dierential equations for
irregular time series, 2020.
[16] Koji Hashimoto. Ads/cft correspondence as a deep boltzmann machine. Physical Review D, 99(10), May
2019.
[17] Tetsuya Akutagawa, Koji Hashimoto, and Takayuki Sumimoto. Deep learning and ads/qcd. Physical
Review D, 102(2), Jul 2020.
[18] Koji Hashimoto, Hong-Ye Hu, and Yi-Zhuang You. Neural ode and holographic qcd, 2020.