![Page 1: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/1.jpg)
Machine Learning and Quantum Technology
German SinucoDepartment of Physics and Astronomy
University of Sussex
Sussex Data Science MeetUpBrighton, 27th June 2019
IBM
MPL - Erlangen
![Page 2: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/2.jpg)
Experimental research in:✔ Trapped ions✔ Cold atomic gases✔ Trapped electrons✔ Non-linear Photonics✔ Terahertz technology
Theoretical studies:✔ Quantum optics✔ Quantum metrology
➢ 10 academic staff, (Lecturers +)➢ ~10+ Postdocts➢ ~30 PhD students➢ ? Master students
Atomic, Molecular and Optical Physics at Sussex
www.sussex.ac.uk/amo/
Atomic, Molecular, and Optical (AMO) research at Sussex University is devoted to the study of fundamental physics and quantum effects using the techniques of atomic and laser physics.
![Page 3: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/3.jpg)
✔ What is Quantum technology?
✔ Machine Learning interlude
✔ How is ML helping us to develop QS/QT?
Plan
![Page 4: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/4.jpg)
“Quantum technology is an emerging field of physics and engineering, which is about creating practical applications … based on properties of quantum mechanics ...” Wikipedia
What is quantum technology?
![Page 5: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/5.jpg)
“Quantum technology is an emerging field of physics and engineering, which is about creating practical applications … based on properties of quantum mechanics ...” Wikipedia
What is quantum technology?
iℏ∂t Ψ=H ΨSpatio-temporal evolution of the wave-function ~ probability distribution
Max Planck1858 - 1947x
![Page 6: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/6.jpg)
INTEL
✔ UK national program in QT: £270 millions 2014-2019
✔ Europe's Quantum Flagship initiative: 1billion for the next 10 years (July 2017)
✔ Quantum information science and tech. in Japan: 1billion in the last 15 years
✔ Quantum Canada: 1billion, last decade alone.
✔ Quantum Computing: Google, Microsoft, IBM, Intel, D-Wave, Riggetti, QuTech, …
✔ Bosh, Total, AirBus, Facebook, ...
Quantum technology
![Page 7: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/7.jpg)
INTEL
✔ UK national program in QT: £270 millions 2014-2019
✔ Europe's Quantum Flagship initiative: 1billion for the next 10 years (July 2017)
✔ Quantum information science and tech. in Japan: 1billion in the last 15 years
✔ Quantum Canada: 1billion, last decade alone.
✔ Quantum Computing: Google, Microsoft, IBM, Intel, D-Wave, Riggetti, QuTech, …
✔ Bosh, Total, AirBus, Facebook, ...
Quantum technology
![Page 8: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/8.jpg)
Devices that rely in our understanding of quantum mechanics: Quantum 1.0
Sizewell A & B Suffolk, England.
Nuclear power stations
MRI scanner Laser
Information technology
Devices/technology exploits the quantum behaviour of very large ensembles of subsystems (~ 1023)
What is quantum technology?
![Page 9: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/9.jpg)
What is quantum technology?
Quantum 2.0 . Generation of devices that exploit controlling individual systems and their interactions.
Phys. Rev. Applied 6, 064007
Supercond. Processor (IBM)
|0 ⟩ ,|1 ⟩ , D=2Qubit:
![Page 10: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/10.jpg)
✔ Many technical challenges: ➢ low-noise generators➢ improve the yield of fabrication processes➢ Slow process to characterise/tuning devices
✔ Quantum devices have very large number of configurations, and full numerical studies are impossible.
✔ Quantum devices are highly sensitive to environmental noise.
Challenges to develop quantum technology
Env. noise
![Page 11: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/11.jpg)
✔ Many technical challenges: ➢ low-noise generators➢ improve the yield of fabrication processes➢ Slow process to characterise/tuning devices
✔ Quantum devices have very large number of configurations, and full numerical studies are impossible.
✔ Quantum devices are highly sensitive to environmental noise.
To be continued ...
Challenges to develop quantum technology
Env. noise
![Page 12: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/12.jpg)
✔ What is Quantum technology?
✔ Machine Learning interlude
✔ How is ML helping us to develop QT?
Plan
![Page 13: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/13.jpg)
Machine Learning craze motivated by:
✔ Big data sets from the internet flow✔ Massiv`e computing resources✔ Good algorithms✔ Success of AlphaGo
Machine Learning
Machine learning: refers to a set of statistical tools for analysis of large data sets
➢ It can extract features of large data set.➢ Generate new data.
![Page 14: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/14.jpg)
Machine Learning3 Flavours
1I1
I2
IN
::
::
I3
1h1
1
h1
2
h1
m
::
h1
3 ...
h1
1 1h1
1
h2
2
h1
m
::
h2
3
h2
1 1h1
1
hn
2
hn
m
::
hn
3
hn
1
Z
Supervised training
Unsupervised training
Physical system
Auxiliar system
Reinforced learning
CNN (Ω): N N ⇒N
![Page 15: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/15.jpg)
Machine Learning3 Flavours
1I1
I2
IN
::
::
I3
1h1
1
h1
2
h1
m
::
h1
3 ...
h1
1 1h1
1
h2
2
h1
m
::
h2
3
h2
1 1h1
1
hn
2
hn
m
::
hn
3
hn
1
Z
Supervised training
Unsupervised training
Physical system
Aux. system
Reinforced learning
CNN (Ω): N N ⇒N
The ML system (network, Agent) defines a non-linear function that depends on the data and on a large set of parameters
Training: the process of adjusting the parameters to obtain optimal representation of the data.
This results in an algorithm that:
✔ produces compressed representation of the data.
✔ Can generate new data that reflect the properties of the training one.
The ML system (network, Agent) defines a non-linear function that depends on the data and on a large set of parameters
Training: the process of adjusting the parameters to obtain optimal representation of the data.
This results in an algorithm that:
✔ produces compressed representation of the data.
✔ Can generate new data that reflect the properties of the training one.
![Page 16: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/16.jpg)
✔ What is Quantum technology?
✔ Machine Learning interlude
✔ How is ML helping us to develop QT?
Plan
![Page 17: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/17.jpg)
Wave-function of the system:
|0 ⟩ ,|1 ⟩ , D=2
Compact representation of quantum states
|Ψ ⟩=∑β=1
2N
Cβ ,Ψ|β ⟩≡ (C1,Ψ
C2,Ψ
C3,Ψ
...C2N ,Ψ
)
![Page 18: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/18.jpg)
|0 0 ⟩ ,|10 ⟩ ,|01 ⟩ ,|11 ⟩ , D=22
Wave-function of the system:
Compact representation of quantum states
|Ψ ⟩=∑β=1
2N
Cβ ,Ψ|β ⟩≡ (C1,Ψ
C2,Ψ
C3,Ψ
...C2N ,Ψ
)
![Page 19: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/19.jpg)
|0 00 0 ⟩ ,|10 0 0 ⟩ , ...|1111 ⟩ D=24
Wave-function of the system:
Compact representation of quantum states
|Ψ ⟩=∑β=1
2N
Cβ ,Ψ|β ⟩≡ (C1,Ψ
C2,Ψ
C3,Ψ
...C2N ,Ψ
)
![Page 20: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/20.jpg)
|00 00 ⟩ ,|10 00 ⟩ , ...|1111 ⟩ D=24
|Ψ ⟩=∑β=1
2N
Cβ ,Ψ|β ⟩
Φ(β ,Ω)=∑[hk ]
e∑k
a kσkz+∑
k '
bk ' hk '+∑kk '
W kk ' hk 'σkz
Cβ ,Ψ=Φ(β ,Ω)
Variational parametrisation
Ω=(ak ,hk ' ,W kk ')
Training ==
tweak to minimise the energy
Compact representation of quantum states
Carleo and Troyer, Science 355, 602 (2017)
![Page 21: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/21.jpg)
|00 00 ⟩ ,|10 00 ⟩ , ...|1111 ⟩ D=24
|Ψ ⟩=∑β=1
2N
Cβ ,Ψ|β ⟩
Φ(β ,Ω)=∑[hk ]
e∑k
a kσkz+∑
k '
bk ' hk '+∑kk '
W kk ' hk 'σkz
Cβ ,Ψ=Φ(β ,Ω)
Variational parametrisation
Ω=(ak ,hk ' ,W kk ')
Training ==
tweak to minimise the energy
Compact representation of quantum states
Accuracy of the ground state energyCarleo and Troyer, Science 355, 602 (2017)
![Page 22: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/22.jpg)
Compact representation of quantum statesReconstruction/Certification
Nieuwenburg, et al arXiv 1904.08441 (2019)
![Page 23: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/23.jpg)
Goal: To protect arbitrary quantum states against deleterious
effects of environmental noise during a period of time T
|Ψ ⟩=∑β=1
2N
Cβ ,Ψ|β ⟩
|⟨Ψ(0)|Ψ(T )⟩|2=e−ΓT
Reduction of effect of noiseError correction
|Ψ(0) ⟩= (αβ0...0
) Noise modifies the qubit state.
|Ψ(t ) ⟩= (~α~β~η...~θ
)
![Page 24: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/24.jpg)
|Ψ ⟩=∑β=1
2N
Cβ ,Ψ|β ⟩
Reduction of effect of noiseError correction
|Ψ(0) ⟩ ,|Ψ(T ) ⟩
|Ψ(0) ⟩= (αβ0...0
) Noise + actions
|Ψ(t ) ⟩= (~α~β~η...~θ
)Contain the same information
![Page 25: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/25.jpg)
Reduction of effect of noiseError correction with RL
Reinforced learning
MPI – Earlangen, Fosel, et al PRX 8, 031084 (2018)
![Page 26: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/26.jpg)
Reduction of effect of noiseError correction
Goal: Protect quantum information for a given period of time
MPI – Earlangen, Fosel, et al PRX 8, 031084 (2018)
![Page 27: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/27.jpg)
Reduction of effect of noiseError correction
State aware network:We know the multiqubit state
This procedure can be applied to any model
MPI – Earlangen, Fosel, et al PRX 8, 031084 (2018)
|Ψ ⟩=∑β=1
2N
Cβ ,Ψ|β ⟩
![Page 28: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/28.jpg)
Produce training examplesTo train a second NN
The event aware network becomes a controller that decides on gate sequences depending on measurements
State aware network Event aware network
Reduction of effect of noiseError correction
MPI – Earlangen, Fosel, et al PRX 8, 031084 (2018)
![Page 29: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/29.jpg)
● ML achieve better performances than established numerical tools to study/design Quantum Devices
● Studying how the learning process can give insights about quantum physics.
● Speeding the process of tuning QD
● Estimation model parameters
● Error correction schemes
● …..
Review
● Dunjko and Breigel, Machine Learning & AI in the quantum domain, Rep. Prog. Phys 81, 074001 (2018)
Learning material
● Florian Maquart website, https://machine-learning-for-physicists.org/
ML is a hot-topic in quantum science:
![Page 30: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/30.jpg)
ML plus Quantum Science
Thanks for listening!
![Page 31: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/31.jpg)
Characterisation of quantum devices
Current maps
Ares' group (Oxford) arXiv 1810.10042 (2018)
➢ 4D parameter space➢ It takes long to explore it in full and we
would like to speed up the process
![Page 32: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/32.jpg)
Characterisation of quantum devices
Ares' group (Oxford) arXiv 1810.10042 (2018)
Use experimental data to reconstruct full resolution current maps and select next combination of parameters to measure.
Data Reconstructed current maps
Next set of measurements is selected identifying the regions with the largest average current gradient
![Page 33: New Machine Learning and Quantum Technology · 2020. 3. 11. · Machine Learning craze motivated by: Big data sets from the internet flow Massiv`e computing resources Good algorithms](https://reader033.vdocuments.net/reader033/viewer/2022060600/60541a2994b6705180797e55/html5/thumbnails/33.jpg)
1. Encoder: CNN to build a compact representation of the training examples (Latent set Z).2. Decoder: Feed with the latent set Z + low resolution sample. Trained to reduced the
difference between the training current map and one reconstruction.
Characterisation of quantum devicesTraining an autoencoder
Ares' group (Oxford) arXiv 1810.10042 (2018)