slides CapTechTalks Webinar May 2024 Alexander Perry.pptx

Slides from a webinar presented May 23, 2024 by Capitol Technology University and featuring faculty member Dr. Alexander Perry discussing hybrid quantum Machine Learning.

1. Presented by Dr. Alexander Perry May 23, 2024 Hybrid Quantum Machine Learning Utilizing Limited-Scale Quantum Computing

2. Agenda Bill Gibbs, Host About Capitol Technology University Session Pointers About the Presenter Presentation Q and A Upcoming Webinars Recording, Slides, Certificate

3. About Established in 1927, we are one of the few private Universities in the U.S. specifically dedicated to STEM-Based academic programs. The University offers degrees at the Associate, Bachelor, Master, and Doctoral levels 3

4. Nonprofit, Private & Accredited Capitol is a nonprofit, private accredited university located in Laurel, Maryland, USA Capitol Technology University is accredited by the Commission on Higher Education of the Middle States Association of Colleges and Schools The University is authorized by the State of Maryland to confer Associate’s (A.A.S.), Bachelor’s (B.S.), Master’s (M.S., M.B.A., M.Ed, M.Res., T.M.B.A, M.Phil.), and Doctoral (D.Sc., Ph.D., D.B.A., Ed.D.) degrees.

5. Capitol offers 16 accredited degrees from the Bachelor’s to Doctoral levels related to this webinar. For more information about degrees and certificates offered in related areas, visit CapTechU.edu.fields -of-study Join us for Master’s and Doctoral Virtual Information Sessions. Held monthly. To learn more: Email: [email protected] Phone: [phone]

6. Session Pointers We will answer questions at the conclusion of the presentation. At any time, you can post a question in the text chat and we will answer as many as we can. Microphones and webcams are not activated for participants. A link to the recording and to the slides will be sent to all registrants and available on our webinar web page. A participation certificate is available by request for both Live Session and On Demand viewers.

7. Dr. Alexander Perry Adjunct Professor at Capitol Data Scientist: Hybrid Quantum-Classical Machine Learning (HQML) Experience: Cyber, Data Science, AI/ML, Quantum Computing 30-year career as software engineer, system administrator, data scientist, technical director Doctor of Science (DSc) in Cybersecurity from Capitol Technology University

8. Presented by Dr. Alexander Perry May 23, 2024 Hybrid Quantum Machine Learning Utilizing Limited-Scale Quantum Computing

9. Hybrid Quantum Machine Learning Utilizing Limited-Scale Quantum Computing Dr. Alexander Perry Capitol Technology University May 23, 2024

10. Modified Heilmeier Catechism What is HQML via Limited-Scale Quantum Computing? Who cares? If you are successful, what difference will it make? What are you trying to do? How is it done today, and what are the limits of current practice? What is new in your approach and why do you think it will be successful? What are the risks? How much will it cost? How long will it take? What are the mid-term and final “exams” to check for success?

11. Quantum Computing The goal of quantum computation is not a single output but rather to create a sampling device of a probability distribution. A qubit is the computational unit in quantum computers. Quantum Superposition: The notion that tiny objects can exist in multiple places or states simultaneously—is a cornerstone of quantum physics. Knowing the quantum state of the system allows us to predict the outcomes of experiments. The Two Golden Rules of Quantum Mechanics: A particle can be in quantum superposition where it behaves as though it is in multiple states at once. When measured, the particle will be found in a single state.

12. Quantum Computing Implementation There are multiple ways to implement a quantum computer: 81,84

13. Quantum Machine Learning (QML) Quantum Machine Learning (QML) explores how to devise and implement quantum software that could enable machine learning on quantum computers (including noisy intermediate-scale quantum, or NISQ) that is faster than classical computers. Hybrid quantum machine learning (HQML) explores how to implement QML using quantum computers (including noisy intermediate-scale quantum, or NISQ ) in conjunction with classical computers to solve ML problems faster than classical computers.

14. Modified Heilmeier Catechism What is HQML via Limited-Scale Quantum Computing? Who cares? If you are successful, what difference will it make? What are you trying to do? How is it done today, and what are the limits of current practice? What is new in your approach and why do you think it will be successful? What are the risks? How much will it cost? How long will it take? What are the mid-term and final “exams” to check for success?

15. The Big Picture

16. Strategic Goal

17. Government and Industry Investments May 09, 2024: DOE Announces $60-70M Quantum Information Science Funding Opportunity Aug 30, 2023: DOE Announces $24M for Research on Quantum Networks Aug 24, 2023: "The administration has requested $75 million for a new account focused on near-term applications of quantum information science.“ Aug 17, 2023: NIST Issues Congressionally Mandated Report on Emerging Tech Areas Aug 16, 2023: NSF Invests $38M to Advance Quantum Information Science and Engineering Aug 15, 2023: AFRL opens Extreme Computing centre for quantum computing research Jul 27, 2023: DOE Announces $11.7 Million for Research on Quantum Computing Jul 12, 2023: Truist and IBM Collaborate on Emerging Technology Innovation and Quantum Computing Jun 22, 2023: Expansion of National Quantum Initiative Pitched to Science Committee

18. Quantum Potential Quantum computing makes use of intrinsically quantum properties such as entanglement and superposition to design algorithms that are faster than classical ones for some class of problems. They offer computational speed-up, that provably no classical system could ever exhibit. Some approaches are based on a parameterized quantum circuit (PQC, discussed in detail later), using neural network-inspired algorithms to train them.

19. Modified Heilmeier Catechism What is HQML via Limited-Scale Quantum Computing? Who cares? If you are successful, what difference will it make? What are you trying to do? How is it done today, and what are the limits of current practice? What is new in your approach and why do you think it will be successful? What are the risks? How much will it cost? How long will it take? What are the mid-term and final “exams” to check for success?

20. Hybrid Quantum Machine Learning (HQML) High-level depiction of hybrid algorithms used for machine learning. Explore implementing a hybrid quantum machine learning (HQML) prototype using noisy intermediate scale quantum (NISQ) computers (a type of LSQC) in conjunction with classical computers to solve machine learning problems faster than classical computers.

21. Outcome, Not Technology, Focused Goal: Use Data Classification via Machine Learning as a way to learn quantum thinking. Method: Variational Quantum Kernel-Based Classification (VQC): Operates through using a variational quantum circuit to classify a training set in direct analogy to conventional SVMs. 119 NISQ (a type of LSQC) computers via Parameterized Quantum C ircuits (PQCs): PQCs offer a concrete way to implement algorithms in the NISQ era . 2,102 IBM’s Qiskit will be the quantum simulator of choice for prototyping. Currently a de-facto community standard. Offers a rich set of quantum computing examples. Offers backends that can run simulator code multiple NISQ devices.

22. Parameterized Quantum Circuit A parameterized quantum circuit (PQC) is a type of ansatz (educated guess or starting point). The core idea is based Variational Quantum Eigensolver (VQE). The goal of a VQE is to find the ground state (expected value of a quantum measurement in this case) of a Hamiltonian H by minimizing the parameters of a PQC given by with regards to an objective function that represents the energy of a given Hamiltonian (classification of data in this case).  

23. HQML Parameterized Quantum Circuit

24. Modified Heilmeier Catechism What is HQML via Limited-Scale Quantum Computing? Who cares? If you are successful, what difference will it make? What are you trying to do? How is it done today, and what are the limits of current practice? What is new in your approach and why do you think it will be successful? What are the risks? How much will it cost? How long will it take? What are the mid-term and final “exams” to check for success?

25. Classical ML Classification Today Done via: Linear regression (univariate and multivariate) Support vector machine (SVM) for support vector classification (SVC). Deep Neural Networks (Deep Learning) Others… Limitations: Speed of processing the data at scale Certain categories problems are intractable for classical computers in : Encryption and Cybersecurity Financial Services Drug Research and Development

26. Modified Heilmeier Catechism What is HQML via Limited-Scale Quantum Computing? Who cares? If you are successful, what difference will it make? What are you trying to do? How is it done today, and what are the limits of current practice? What is new in your approach and why do you think it will be successful? What are the risks? How much will it cost? How long will it take? What are the mid-term and final “exams” to check for success?

27. What’s New This diagram gives a brief overview of the Variational Quantum Classification protocol.

28. Why it will work: HQML in Feature Hilbert spaces The basic idea of quantum computing is surprisingly similar to that of kernel methods in machine learning, namely to efficiently perform computations in an intractably large Hilbert space. We interpret the process of encoding inputs in a quantum state as a nonlinear feature map that maps data to quantum Hilbert space. PQCs can form Gaussian Kernels that can be used to derive adaptive learning rates for gradient ascent. Even at low circuit depth, some classes of PQCs can generate highly non-trivial outputs. PQCs may offer a concrete way to implement QML algorithms on NISQ devices.

29. Kernel Functions The “kernel trick” maps input data into a higher dimensional space, making it easier to solve non-linearly separable problems. Mathematically, a kernel function can be defined as: where is the kernel function, and are -dimensional inputs, is a map from n-dimension to -dimension space and denotes the inner product. When considering finite data, a kernel function can be represented as a matrix:  

30. Kernel Methods for Machine Learning

31. Modified Heilmeier Catechism What is HQML via Limited-Scale Quantum Computing? Who cares? If you are successful, what difference will it make? What are you trying to do? How is it done today, and what are the limits of current practice? What is new in your approach and why do you think it will be successful? What are the risks? How much will it cost? How long will it take? What are the mid-term and final “exams” to check for success?

32. Risks: Potentially Expensive Failure Hardware challenges: Quantum Decoherence : In quantum information processing, the term decoherence is often used loosely to describe any kind of noise that can affect/collapse quantum particles to a classical state, as if it’s being measured, and eliminate the quantum behavior of particles. Algorithmic Challenges: Supervised HQML training often requires extensive amounts of time. HQML suffers from the barren plateau problem.

33. Modified Heilmeier Catechism What is HQML via Limited-Scale Quantum Computing? Who cares? If you are successful, what difference will it make? What are you trying to do? How is it done today, and what are the limits of current practice? What is new in your approach and why do you think it will be successful? What are the risks? How much will it cost? How long will it take? What are the mid-term and final “exams” to check for success?

34. Extremely Expensive Commercial usage of existing NISQ systems can easily reach into the $100,000 to +$1,000,000 range. Vendors offer researchers credits and/or free usage of their smaller NISQ systems (often with time limits).

35. Modified Heilmeier Catechism What is HQML via Limited-Scale Quantum Computing? Who cares? If you are successful, what difference will it make? What are you trying to do? How is it done today, and what are the limits of current practice? What is new in your approach and why do you think it will be successful? What are the risks? How much will it cost? How long will it take? What are the mid-term and final “exams” to check for success?

36. Timeframes General Purpose Quantum Computers with millions of logical qubits are 10-15 years away (at best). NISQ systems with up to 1000 raw qubits and 48 logical qubits exist: Dec 4, 2023: IBM releases first-ever 1,000-qubit quantum chip Dec 7, 2023: Logical quantum processor based on reconfigurable atom arrays

37. Modified Heilmeier Catechism What is HQML via Limited-Scale Quantum Computing? Who cares? If you are successful, what difference will it make? What are you trying to do? How is it done today, and what are the limits of current practice? What is new in your approach and why do you think it will be successful? What are the risks? How much will it cost? How long will it take? What are the mid-term and final “exams” to check for success?

38. Success Checkpoints If HQML is going to work, we should see successful prototypes in the next 3-5 years. In 5-10 years, we should see productions applications of HQMLs if the technology is successful.

39. Questions Thank you Questions?

40. Upcoming webinar www.captechu.edu/webinar-series Defining the DoD Roadmap to Digital Supremacy by Effectively Adopting Digital Transformation June 20 Dr. Donovan Wright

41. captechu.edu/webinars-and-podcasts

42. Recording, Slides & Certificate Links to the slides and recording will be sent to all registrants. Watch for an email A Certificate of Completion is available upon request to both live session and On Demand viewers Simply reply to the email

43. Thanks for Joining Us! Thank You! This concludes today’s webinar Watch for a follow up email that contains: How to get a Participation Certificate (Available by request for both Live Session and On Demand viewers) Link to the webinar recording and slides