Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a...
Transcript of Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a...
![Page 1: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/1.jpg)
Raffaele [email protected]
Learning quantum physics from experiments with Bayesian inference and quantum computers
![Page 2: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/2.jpg)
![Page 3: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/3.jpg)
Outline
• Quantum Computers a (very) short introduction
• Quantum Hamiltonian Learning (QHL) -> Learning parameters of a model.
• From learning parameters to learning models…
• Wiebe et al., Hamiltonian Learning and Certification Using Quantum Resources. Phys. Rev. Lett. 112, (2014)
• Wang et al., Experimental quantum Hamiltonian learning - Nature Physics 1, 149 (2017)
• Santagati et al., Magnetic-field-learning using a single electronic spin in diamond… - Phys. Rev. X (2019)
• Gentile et al., Characterising quantum systems with Bayesian inference – manuscript in preparation (2019)
• Flynn et al., Exploring acyclic graphs for the study of quantum systems – manuscript in preparation (2019)
![Page 4: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/4.jpg)
Quantum physics
EINSTEIN, N. ROSEN and B. PODOLSKY, Phys. Rev. 47, 777 (1935) D. R. INGLIS, Rev. Mod. Phys. 33, 1 (1961)
Entanglement
Quantum mechanics although counterintuitive is correct in its predictions
Defines probability
Schrödinger’s cat
PROBABILISTIC MEASUREMENT OUTCOMES
Superposition
Quantum superposition of 2 states:
![Page 5: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/5.jpg)
Quantum Simulation
Quantum systems are hard to simulate.
If we study a system comprising N particles with two basis states and wewant to store their quantum state on a classical computer using an 8 bits (1 byte) resolution.
2 p 4 bytes
1 p 2 bytes
3 p 8 bytes
200 p ~ 1.6 1060 bytes
The total amount of memory required grows exponentially with the value of N.
Mem
ory
Number of particles
Total memory available on earth today is ~ 5.21 1021 bytes
![Page 6: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/6.jpg)
Quantum Simulation
Quantum systems are hard to simulate.
Feynman - Simulating Physics with Computers International Journal of Theoretical Physics, VoL 21, 6-7, 1982
Mem
ory
Number of particles
We need a new kind of computer to simulatequantum systems efficiently.
0
1
Bit Quantum bit
Measurement outcomes are probabilistic
Probability of 0
Probability of 1
![Page 7: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/7.jpg)
Quantum computers and what they are good at
Factoring Numbers (Shor’s algorithm) Quantum Machine Learning
Quantum Simulation Optimization
Exponential speed-upPoly and exponential speed-up
Exponential speed-ups Poly speed-ups
Ashley Montanaro Quantum Information 2, 15023, 2016
Quantum algorithm zoohttps://math.nist.gov/quantum/zoo/
![Page 8: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/8.jpg)
Quantum computing funding landscape
Topological• UK -> 400 million £
• EU -> 1.6 billion €
• GE -> 600 million €
And many more….
Investments inQuantum technologies
Other Solid state
Photonics
Superconducting
![Page 9: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/9.jpg)
Exponential scaling of
required classical
resources
31
Bayesian approach
(similar to Bayesian phase estim.)
Ø Choose experiment (i.e. evol. time ) given .
Ø Obtain data from the system.
Ø Use the quantum simulator to calculate the
likelihood functions .
Ø Update the prior distribution using Bayes’ rule
The information is encoded in a prior distribution
• Near-optimal choice: .
• Simulated Monte-Carlo approximation used to make the Bayesian update practical.
Quantum Hamiltonian LearningWhat is a Hamiltonian operator?
![Page 10: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/10.jpg)
Quantum System Model of the system (H)
with parameters
Quantum Hamiltonian Learning
QHL aims to find efficiently the set of parameters which best describe the dynamic of the system.
Mohseni et al. Phys. Rev. A 77, 032322 (2008)Hentschel and Sanders Phys. Rev. Lett. 104, 063603 (2010)Da Silva et al. Phys. Rev. Lett. 107, 210404 (2011)Wiebe et al. Phys. Rev. Lett. 112, 190501 (2014)Wiebe et al. Phys. Rev. A 89, 042314 (2014)
![Page 11: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/11.jpg)
1. From choose an
experiment (e.g. )
2. Perform experiment on
System and obtain
outcome E
Phys. Rev. Lett. 112, 190501 (2014)
3. Calculate likelihoods
using quantum simulator
4. Update the prior distribution
using Bayes’ rule:
Nature Physics 13, 551-555 (2017)
31
Bayesian approach
(similar to Bayesian phase estim.)
Ø Choose experiment (i.e. evol. time ) given .
Ø Obtain data from the system.
Ø Use the quantum simulator to calculate the
likelihood functions .
Ø Update the prior distribution using Bayes’ rule
The information is encoded in a prior distribution
• Near-optimal choice: .
• Simulated Monte-Carlo approximation used to make the Bayesian update practical.
Quantum Likelihood Estimation
![Page 12: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/12.jpg)
Experimental set-up
Nature Physics 13, 551-555 (2017)
Hamiltonian: Where we want to
estimate the frequency
of oscillations
c
![Page 13: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/13.jpg)
Experimental results
frequency inferred with QLE:
Rescaling
Nature Physics 13, 551-555 (2017)
![Page 14: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/14.jpg)
First part: QHL Take home message
• Finding the parameter of H in a scalable way
• We learned only parameters not models
• Can we use similar techniques to help us
improve models?
31
Bayesian approach
(similar to Bayesian phase estim.)
Ø Choose experiment (i.e. evol. time ) given .
Ø Obtain data from the system.
Ø Use the quantum simulator to calculate the
likelihood functions .
Ø Update the prior distribution using Bayes’ rule
The information is encoded in a prior distribution
• Near-optimal choice: .
• Simulated Monte-Carlo approximation used to make the Bayesian update practical.
• Gentile et al., Experimental Quantum Model Learning –manuscript in preparation (2019)
• Flynn et al., Exploring acyclic graphs for the study of quantum systems – manuscript in preparation (2019)
![Page 15: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/15.jpg)
Methods to model physical systems with ML
Crutchfield and McNamara, Complex Syst. 1, 417 (1987)Schmidt and Lipson, Science 324, 81 (2009)Torlai and Melko Phys Rev B 94, 165134 (2016)Bairey, Arad, and Lindner PRL 122, 020504 (2019)Iten, et al., arXiv preprint arXiv:1807.10300 (2018)
Devise an automatic strategy to generate and select models by performing a search on a tree for the Hamiltonian model which best approximate the system dynamics.
![Page 16: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/16.jpg)
QMLA prior
Quantum Model Learning
Explore and prune
for in DAG:
QHL update Consolidate
..
.
C. Granade, C. Ferrie, I. Hincks, S. Casagrande, T. Alexander, J. Gross, M. Kononenko, and Y. Sanders, Qinfer: Statistical Inference Software for Quantum Applications, Quantum 1, 5 (2017).
The protocol compares in pairs (Hi,Hj) all models within the
same layer using the Bayes factor
where
![Page 17: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/17.jpg)
Learning models of quantum system:NV centre in diamond
Electron spin interacting with a Carbon nuclear spin
We want to learn the basics of the
interaction between the spin in the NV centre in diamond and
the closest 13C nuclear spin through Hahn echo measurements.
Hyperfine gyromagnetic tensors
Electron gyromagnetic tensors
cc
![Page 18: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/18.jpg)
Results – NV-centre model learning
![Page 19: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/19.jpg)
Conclusions
Wang et al. Nature Physics 13, 551-555 (2017)Santagati et al. Phys. Rev. X 9, 021019 (2019)Gentile et al., man. in prep. (2019)Flynn et al., man. in prep. (2019)
• Quantum computers• Statistical inference (QHL) for parameters in H• Quantum model learning to learn the models of a quantum system• From the experiments we find models compatible with theoretical
developments.
![Page 20: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/20.jpg)
Backup slides
![Page 21: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/21.jpg)
Results – NV-centre model learning
![Page 22: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/22.jpg)
Other models performance
![Page 23: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/23.jpg)
Experiment measurements
c
We measure the traced-out state i.e. Measure only the electron and not theother interacting particles.
It is a bit like studying the property of an iceberg by looking only at the tip.
electron
nucleus
![Page 24: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/24.jpg)
What quantum computers are good at and what they are not?
Quantum computers don’t compute all the answers in parallel
Quantum computers are not faster than classical computers at every task
Quantum computers can give exponential and polynomial speed ups at specific tasks.
G Wendin 2017 Rep. Prog. Phys. 80 106001
https://www.smbc-comics.com/comic/the-talk-3
![Page 25: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/25.jpg)
Experiment – NV-centre model learning
![Page 26: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/26.jpg)
Experimental results: Variance
Variance:
SATURATION = Bad quantum simulator or bad model!
Wang et al. Nature Physics 13, 551-555 (2017)
![Page 27: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/27.jpg)
Model I:
Model II:
From QHL to comparing models
Bayes Factor:
Updated value (for model II, including
chirping) of Norm of covariance matrix:
Old Variance with model I:
Model I
Model II
0 10 20 30 40 5010- 6
10- 5
10- 4
10- 3
10- 2
10- 1
1
10- 6
10- 5
10- 4
10- 3
10- 2
10- 1
1
Experiment Number
2(
)
![Page 28: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/28.jpg)
From quantum simulation to qubits
A new theory of information based on quantum mechanics.
0
1
Bit Quantum bit
Measurement
Measurement outcomes are probabilistic
Probability of 0
Probability of 1
Quantum Computation and Quantum Information - M. Nielsen and I.L. Chuang - Cambridge
![Page 29: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/29.jpg)
Experiment measurements
c
electron
nucleus
Model1 Model2
BF(H2; H1) = 1010
![Page 30: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/30.jpg)
Estimating the number of spins
Gali, Fyta, and Kaxiras Ab initio supercell calculations on nitrogen-vacancy center in diamond:Electronic structure and hyperfine tensors. PHYSICAL REVIEW B 77, 155206 (2008)
Galli - Identification of individual 13C isotopes of nitrogen-vacancy center in diamond by combining thepolarization studies of nuclear spins and first-principles calculations. PHYSICAL REVIEW B 80, 241204 (2009)
![Page 31: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/31.jpg)
Sequential Monte Carlo
Phys. Rev. Lett. 112, 190501 (2014)
2p W
2p W
P(W
)P
(W)
![Page 32: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/32.jpg)
2 spins model – Expectation values
![Page 33: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/33.jpg)
Simulation – NV-centre model learning
Measuring the traced out state state
![Page 34: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/34.jpg)
QMLA prior
Model Learning
for in DAG:
QHL update
Learn from data
each model
Consolidate
The protocol compares in pairs (Hi,Hj) all models within the
same layer using the Bayes factor
where
Explore and prune
Generate new models till
n qubits allows or generate new layer.
.
.
![Page 35: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/35.jpg)
The problem is NP Hard… but
A Bayesian analysis of decision-tree learning has shown that near-optimal solutions can be found in polynomial time.
Muggleton, S. Page, D.: A learnability model of universal representations and its application to top-down induction of decision trees. In K. Furukawa, D. Michie, & S. Muggleton(eds.) Machine Intelligence 15, Oxford University Press, page 16 (1999)
![Page 36: Learning quantum physics from experiments with Bayesian ... · Outline •Quantum Computers a (very) short introduction •Quantum Hamiltonian Learning (QHL) -> Learning parameters](https://reader031.fdocuments.in/reader031/viewer/2022011910/5f8258e1e6c7c41441672e1d/html5/thumbnails/36.jpg)
Dynamics examples with BF and PGH times