Decision making using classical and quantum light

3bb96e0f5b6a84c489463499a0a38f17?s=47 Makoto Naruse
February 04, 2020

Decision making using classical and quantum light

Keynote presentation at SPIE Photonics West (Conference: AI and Optical Data Sciences) on Feb 4, 2020 at San Francisco.
https://spie.org/PWO/conferencedetails/ai-and-optical-data-sciences?SSO=1

3bb96e0f5b6a84c489463499a0a38f17?s=128

Makoto Naruse

February 04, 2020
Tweet

Transcript

  1. http://www.inter.ipc.i.u-tokyo.ac.jp/ Decision making using classical and quantum light Makoto Naruse1,

    Nicolas Chauvet1, Serge Huant2, Satoshi Sunada3, Atsushi Uchida4, and Hirokazu Hori5 1 The University of Tokyo, Tokyo, Japan 2 Université Grenoble Alpes, CNRS, Institut Néel, Grenoble, France 3 Kanazawa University, Saitama, Japan 4 Saitama University, Saitama, Japan 5 University of Yamanashi, Kofu, Japan
  2. http://www.inter.ipc.i.u-tokyo.ac.jp/ Decision making using classical and quantum light Outline 1.

    Introduction 2. Decision making based on single photons 3. Ultrafast decision making based on laser chaos 4. Social decision making by entangled photons 5. Application studies: Laser chaos WiFi 6. Summary Sci Rep 2015 Sci Rep 2017 Sci Rep 2018 Sci Rep 2019 Sci Rep 2019 Sci Rep 2020
  3. 3 ▪Decision making Conduct prompt and accurate action in dynamically

    changing uncertain environments Action Reward Decision-Making System Dynamically changing, uncertain environments
  4. http://www.inter.ipc.i.u-tokyo.ac.jp/ Sutton, et al. Reinforcement Learning, MIT Press Multi-Armed Bandit

    problem (MAB) - Simple but very difficult… a fundamental problem in reinforcement learning - A variety of computer algorithms (e.g., -greedy, UCB, etc.)
  5. http://www.inter.ipc.i.u-tokyo.ac.jp/ 5 Why? Decision Making (Reinforcement Learning) is foundation of

    important applications Decision Making by Photonics Single Photon Decision Maker (Sci. Rep. 2015) Decision making by Laser Chaos (Sci. Rep. 2017) Ultrafast 1GHz (100 Gsample/s) •Robotics •Fintech High Frequency Trading •AI •Data Center •Arbitration •Network •Social optimization •Everyday life Microsoft DeepMind (Unexpected) Impact.. Invited to Google 2016 Kimura award Docomo mobile science award
  6. http://www.inter.ipc.i.u-tokyo.ac.jp/ Single-photon decision maker Naruse, Huant, et al. Sci. Rep.

    2015 6 Autonomous adaptation to environmental uncertainty Utilize wave-particle duality of light quanta Two-armed bandit problem
  7. http://www.inter.ipc.i.u-tokyo.ac.jp/ Decision making using classical and quantum light Outline 1.

    Introduction 2. Decision making based on single photons 3. Ultrafast decision making based on laser chaos 4. Social decision making by entangled photons 5. Application studies: Laser chaos WiFi 6. Summary Sci Rep 2015 Sci Rep 2017 Sci Rep 2018 Sci Rep 2019 Sci Rep 2020
  8. http://www.inter.ipc.i.u-tokyo.ac.jp/ 8 What is laser chaos? (1) J. Ohtsubo, Semiconductor

    lasers: stability, instability and chaos (Springer, 2012). (2) A. Uchida, Optical communication with chaotic lasers (Wiley-VCH, 2012). Random number generator: A. Uchida, A. et al. Nature Photonics 2008. Reservoir computing: L. Larger, L. et al. Optics Express 2012. Applications
  9. http://www.inter.ipc.i.u-tokyo.ac.jp/ Ultrafast Decision Making by Laser Chaos M. Naruse, A.

    Uchida et al. Sci. Rep. 2017 REWARD 2-armed bandit problem 0
  10. http://www.inter.ipc.i.u-tokyo.ac.jp/ • LASER CHAOS achieved superior performance M. Naruse, A.

    Uchida et al. Sci. Rep. 2017 WHY ?? By CHAOS By pseudorandom numbers Ultrafast Decision Making by Laser Chaos (cont) Correct decision ratio • Maximum performance is obtained when the sampling interval is 50 ps.
  11. http://www.inter.ipc.i.u-tokyo.ac.jp/ EXACT COINCIDENCE ! Decision making performance M. Naruse, A.

    Uchida et al. Sci. Rep. 2017 Time-domain property Time-domain dynamics in laser chaos provides certain positive effects!! Ultrafast Decision Making by Laser Chaos (cont)
  12. http://www.inter.ipc.i.u-tokyo.ac.jp/ MN, M. Berthel, et al. ACS Photonics 2016 Scalability

    of decision making 2-armed 4-armed 8-armed
  13. http://www.inter.ipc.i.u-tokyo.ac.jp/ Chaos-based scalable decision making 000 001 010 011 100

    101 110 110 Most Significant Bit (MSB) Second Most Significant bit Least Significant Bit (LSB) TH1 TH2,0 TH2,1 TH3,0,0 TH3,0,1 TH3,1,0 TH3,1,1 Naruse, Mihana, Uchida et al. Sci Rep 2018 Time-domain multiplexing
  14. http://www.inter.ipc.i.u-tokyo.ac.jp/ 1.16 a N  By CHAOS By CHAOS By

    CHAOS By pseudorandom numbers By pseudorandom numbers By pseudorandom numbers Naruse, Mihana, Uchida et al. Sci Rep 2018
  15. http://www.inter.ipc.i.u-tokyo.ac.jp/ Why chaos works well?  2 1 1 (

    ) ( ) ( ) T t ETMSD x t x t T            “Ensemble average of the Time-averaged Mean Square Displacement” (ETMSD) Random walker’s trajectory Diffusivity analysis Naruse, Uchida et al. Sci Rep 2018
  16. http://www.inter.ipc.i.u-tokyo.ac.jp/ Ring laser Waveguide GaAs/AlGaAs semiconductor ring laser Radius: 1

    mm On-chip photonic decision making using a ring laser Ring laser - Integrated photonics technology - Usually, deterministic operations (e.g. optical switch) [Principle] Spontaneous & probabilistic switching between CW and CCW oscillation Solving two-armed bandit problem CW CCW 16 2019 Chaotic time series is not the only resource for photonic decision making…
  17. http://www.inter.ipc.i.u-tokyo.ac.jp/ Decision making using laser network The emerging and unique

    attributes in network of lasers Leader-laggard synchronization Spontaneous and probabilistic switching between the leader and the laggard Solving two-armed bandit problem 17 • Synchronization among geographically different locations • Spontaneous symmetry breaking • Ultrafast dynamics Mihana, Uchida, et al. Optics Express 2019 POC of the utilization of laser synchronizations for intelligent functions
  18. http://www.inter.ipc.i.u-tokyo.ac.jp/ Decision making using classical and quantum light Outline 1.

    Introduction 2. Decision making based on single photons 3. Ultrafast decision making based on laser chaos 4. Social decision making by entangled photons 5. Application studies: Laser chaos WiFi 6. Summary Sci Rep 2015 Sci Rep 2017 Sci Rep 2018 Sci Rep 2019 Sci Rep 2020
  19. http://www.inter.ipc.i.u-tokyo.ac.jp/ 7 7 7 7 7 7 SLOT MACHINE A

    SLOT MACHINE B REWARD PROBABILITY 0.2 0.8 PLAYER 1 CONFLICT OF INTERESTS!! PLAYER 2 Competitive Multi-Armed Bandit problem (Competitive MAB)
  20. http://www.inter.ipc.i.u-tokyo.ac.jp/ 20

  21. http://www.inter.ipc.i.u-tokyo.ac.jp/ Social optimization and Equality [Social welfare] Conflict ! OR

    MACHINE A MACHINE B PLAYER 1 PLAYER 2
  22. http://www.inter.ipc.i.u-tokyo.ac.jp/ Slot machine A Slot machine B |H> |V> PBS

    |H> |V> ENTANGLED PHOTONS +   1 , , 2 s i s i H V V H  Slot machine A Slot machine B PLAYER 1 PLAYER 2 ABSOLUTELY NO CONFLICT Architecture + Signal light Idler light
  23. http://www.inter.ipc.i.u-tokyo.ac.jp/ Slot machine A Slot machine B |H> |V> PBS

    |H> |V> CORRELATED PHOTON PAIR HV Slot machine A Slot machine B PLAYER 1 PLAYER 2 CONFLICTS CAN HAPPEN Polarization-orthogonal photon pair
  24. http://www.inter.ipc.i.u-tokyo.ac.jp/ Experimental setup SPDC Chuavet et al. Sci Rep 2019

  25. http://www.inter.ipc.i.u-tokyo.ac.jp/ CONFLICT ratio PLAYER 2 PLAYER 1 NO COLLABORATION Conflict

    ratio is VERY high! Accumulated Reward 7 7 7 Machine A 0.8 7 7 7 Machine B 0.2 7 7 7 Machine B 0.8 7 7 7 Machine A 0.2 Chuavet et al. Sci Rep 2019
  26. http://www.inter.ipc.i.u-tokyo.ac.jp/ PLAYER 2 PLAYER 1 ENTANGLED PHOTON PAIR Conflict ratio

    is very small! Total reward becomes very high! CONFLICT ratio Accumulated Reward Chuavet et al. Sci Rep 2019
  27. http://www.inter.ipc.i.u-tokyo.ac.jp/ PLAYER 2 PLAYER 1 CORRELATED PHOTON PAIR Conflict ratio

    still high! CONFLICT ratio Accumulated Reward Chuavet et al. Sci Rep 2019
  28. http://www.inter.ipc.i.u-tokyo.ac.jp/ PLAYER 2 PLAYER 1 28 DECEPTION

  29. http://www.inter.ipc.i.u-tokyo.ac.jp/ With correlated photons Chuavet et al. Sci Rep 2019

    Deception, success
  30. http://www.inter.ipc.i.u-tokyo.ac.jp/ 30 Yes, this way I would earn more …

    With entangled photons Chuavet et al. Sci Rep 2019 Deception, fail
  31. http://www.inter.ipc.i.u-tokyo.ac.jp/ 31 Featured at Behavioural & Social Sciences in Nature

    Research 2019
  32. http://www.inter.ipc.i.u-tokyo.ac.jp/ Decision making using classical and quantum light Outline 1.

    Introduction 2. Decision making based on single photons 3. Ultrafast decision making based on laser chaos 4. Social decision making by entangled photons 5. Application studies: Laser chaos WiFi 6. Summary Sci Rep 2015 Sci Rep 2017 Sci Rep 2018 Sci Rep 2019 Sci Rep 2020
  33. http://www.inter.ipc.i.u-tokyo.ac.jp/ 33 IEEE802.11a Takeuchi et al Sci Rep 2020 Higher

    throughput = Reward
  34. http://www.inter.ipc.i.u-tokyo.ac.jp/ 34 Dynamic channel selection Takeuchi et al Sci Rep

    2020
  35. http://www.inter.ipc.i.u-tokyo.ac.jp/ 35 Takeuchi et al Sci Rep 2020 Experimental results

  36. http://www.inter.ipc.i.u-tokyo.ac.jp/ 36 2019 Artificial data generation using chaos Via pseudorandom

    numbers Via laser chaos time series Chaos GAN
  37. http://www.inter.ipc.i.u-tokyo.ac.jp/ Category theory approach to photonic decision making MN, Hori,

    Huant, et al. Int. J. Information Technology and Decision Making, 2018 MN, Hori, et al. Philosophies 2017 Kubota, MN, Hori, et al. Philosophies 2017 MN, et al. PLOS ONE 2018 Casino Setting /2 Polarizer Setting /2 Decision RESULT (Win or Lose) Photon environment Machine environment P M F Q X Y Unveil complex interdependency involved in decision making Theoretical fundamentals
  38. Summary Decision making using classical and quantum light Thank you

    for your kind attention! 1. Decision making based on single photons 2. Ultrafast decision making based on laser chaos 3. Social decision making by entangled photons 4. Applications Sci Rep 2015 Sci Rep 2017, Sci Rep 2018 Sci Rep 2019 Sci Rep 2019 Sci Rep 2020