[PPT] - Can an Artificial-Intelligence Win a Nobel Prize? Can an PowerPoint Presentation

SLIDE 1

Can an Artificial-Intelligence Win a Nobel Prize?

GTC 2017

SLIDE 2

Can an Artificial-Intelligence Win a Nobel Prize?

Andre  Luiten Anton van   den Hengel John  Bastian Ian  Petersen Carlos  Kuhn Cairon  Quinlivan Kyle  Hardman Mahassen  Sooriyabandara Gordon  McDonald Patrick  Everitt Perumbil  Manju

Paul  Wigley

Nick  Robins Joe  Hope

Michael  Hush

SLIDE 3

CAN AN ARTIFICIAL- INTELLIGENCE WIN A NOBEL PRIZE?

SLIDE 4

CAN AN ARTIFICIAL- INTELLIGENCE WIN A NOBEL PRIZE?

NO

SLIDE 5

CAN MACHINE LEARNING BE USED TO OPTIMIZE AN ULTRA- COLD ATOM EXPERIMENT AND REDISCOVER A RESULT THAT PREVIOUSLY WON A NOBEL PRIZE?

SLIDE 6

CAN MACHINE LEARNING BE USED TO OPTIMIZE AN ULTRA- COLD ATOM EXPERIMENT AND REDISCOVER A RESULT THAT PREVIOUSLY WON A NOBEL PRIZE? PRETTY MUCH

SLIDE 7

Overview

Machine Learning  Algorithm Automated  Optimization Evaporation 

f a BEC

X C(X)

X (a) (b) C(X ) X (a) (b) t V C(X )

P B Wigley et al. Scientific Reports 6 25890 (2016)

Can an Artificial-Intelligence Win a Nobel Prize?

SLIDE 8

Overview

Machine Learning  Algorithm Automated  Optimization Evaporation 

f a BEC

X C(X)

X (a) (b) C(X ) X (a) (b) t V C(X )

Physics Control Computer  Science P B Wigley et al. Scientific Reports 6 25890 (2016)

Can an Artificial-Intelligence Win a Nobel Prize?

SLIDE 9

Overview

Machine Learning  Algorithm Automated  Optimization Evaporation 

f a BEC

X C(X)

X (a) (b) C(X ) X (a) (b) t V C(X )

Can an Artificial-Intelligence Win a Nobel Prize?

SLIDE 10

What is a Bose Einstein Condensate (BEC)?

Can an Artificial-Intelligence Win a Nobel Prize?

Absolute Zero

SLIDE 11

What is a Bose Einstein Condensate (BEC)?

Can an Artificial-Intelligence Win a Nobel Prize?

Absolute Zero nK mK kK μK K 293K 2.73K 1.9K 100 nK

SLIDE 12

Nobel Prize 2001

Can an Artificial-Intelligence Win a Nobel Prize?

▸ BEC proposed in 1924 ▸ BEC created in 1995 ▸ Nobel prize awarded in 2001

Eric  Cornell Wolfgang  Ketterle Carl  Wieman Satyendra   Nath Bose Albert   Einstein

SLIDE 13

Precision measurement with a BEC

▸ Atoms are sensitive to gravity

and magnetic fields.

▸ Geoscience. ▸ BECs are a coherent, narrow

linewidth source for atomic interferometers.

BEC  Source Measure Phase  Change S S Szigeti et al. NJP 14 023009

Can an Artificial-Intelligence Win a Nobel Prize?

SLIDE 14

Precision measurement with a BEC

▸ Atoms are sensitive to gravity

and magnetic fields.

▸ Gravitation precision: 

10-9 Δg/g

▸ Magnetic field gradient

precision: 8 pT/m

▸ First interferometer to

measure both.

K S Hardman et al. Phys. Rev. Lett. 117, 138501 (2016)

Can an Artificial-Intelligence Win a Nobel Prize?

SLIDE 15

Evaporative cooling to create a BEC

E E E V ρ ρ ρ

Can an Artificial-Intelligence Win a Nobel Prize?

SLIDE 16

Evaporation ramps

X C(X ) (b) t V

V(t) t V(t)

▸ Ergodic dynamics,  

two-body s-wave interactions, no other loss rates,   => Exponential ramps optimal. 

Can an Artificial-Intelligence Win a Nobel Prize?

SLIDE 17

Evaporation ramps

X C(X ) (b) t V

V(t) t V(t)

▸ Ergodic dynamics,  

two-body s-wave interactions, no other loss rates,   => Exponential ramps optimal. 

?

Can an Artificial-Intelligence Win a Nobel Prize?

SLIDE 18

Overview

Machine Learning  Algorithm Automated  Optimization Evaporation 

f a BEC

X C(X)

X (a) (b) C(X ) X (a) (b) t V C(X )

Can an Artificial-Intelligence Win a Nobel Prize?

SLIDE 19

Evaporation as an optimization problem

▸ We can parametrize the ramps:

Can an Artificial-Intelligence Win a Nobel Prize?

SLIDE 20

Evaporation as an optimization problem

▸ We can parametrize the ramps: ▸ 3 ramps, common = 16 parameters

Can an Artificial-Intelligence Win a Nobel Prize?

SLIDE 21

Evaporation as an optimization problem

▸ Condensate number difficult with few measurements

▸ Use width of image above a threshold. ▸ Cost => . Uncertainty from 2 measurements

Thermal  state Condensed  state

Can an Artificial-Intelligence Win a Nobel Prize?

SLIDE 22

Overview

Machine Learning  Algorithm Automated  Optimization Evaporation 

f a BEC

X C(X)

X (a) (b) C(X ) X (a) (b) t V C(X )

How do we pick  what X to test next?

Can an Artificial-Intelligence Win a Nobel Prize?

SLIDE 23

Previous automated optimization experiments

▸ Quantum chemistry ▸ R S Judson et al. PRL 68, 1500–1503 (1992) ▸ Cold ion quantum computing ▸ Kelly et al. PRL 112, 240504 (2014) ▸ Cold atoms ▸ I Geisel et al. APL 102, 214105 (2013)

Can an Artificial-Intelligence Win a Nobel Prize?

SLIDE 24

Previous used automated optimization algorithms

▸ Brute force search ▸ 16 parameters, to 10% 

each experiment 1 min  total time ~ 1017 s

▸ Nelder-Mead ▸ Caught in local minima ▸ Genetic ▸ Chooses new points randomly ▸ What’s missing?

X1 X2 C(X)

Can an Artificial-Intelligence Win a Nobel Prize?

SLIDE 25

Overview

Machine Learning  Algorithm Automated  Optimization Evaporation 

f a BEC

X C(X)

X (a) (b) C(X ) X (a) (b) t V C(X )

Can an Artificial-Intelligence Win a Nobel Prize?

SLIDE 26

Machine learning: The problem

▸ Regression

Can an Artificial-Intelligence Win a Nobel Prize

SLIDE 27

Machine learning: Gaussian process fit

▸ Kriging: Geoscience ▸ Assumes data is samples from

a set of gaussian process.

▸ Produces estimate of the mean

and uncertainty.

▸ Requires a set of

hyperparameters: correlation length for each dimension.

▸ Hyperparameters fit from data.

X (a) C(X )

Correlation length Short Medium Long

Can an Artificial-Intelligence Win a Nobel Prize

SLIDE 28

Machine learning: Strategy

Can an Artificial-Intelligence Win a Nobel Prize

SLIDE 29

Machine learning: Strategy

(A) (B) (C)

▸ Where would you do the next experiment?

Can an Artificial-Intelligence Win a Nobel Prize

SLIDE 30

Choosing the next point to test

▸ Minimize: ▸ When b=0 learner acts like a “scientist” ▸ When b=1 learner acts like an “engineer” ▸ We swept b between 0 to 1. ▸ Used randomized gradient solver to find

minimum.

Can an Artificial-Intelligence Win a Nobel Prize

SLIDE 31

Overview

Machine Learning  Algorithm Automated  Optimization Evaporation 

f a BEC

X C(X)

X (a) (b) C(X ) X (a) (b) t V C(X )

Can an Artificial-Intelligence Win a Nobel Prize

SLIDE 32

Results I

Common  Training   Data Machine  Learning  10 experiments Nelder-Mead  133 experiments 5 x 105 atoms

Can an Artificial-Intelligence Win a Nobel Prize

SLIDE 33

Using the model

▸ Machine learning algorithms also produces a model ▸ Correlation lengths! ▸ For small data sets we found: ▸ Learner identified important (short correlation length)

parameters correctly.

▸ Learner did not consistently identify unimportant (long

correlation length).

Can an Artificial-Intelligence Win a Nobel Prize

SLIDE 34

Maximum likelihood for hyperparameters

▸ Problem: with small data sets

multiple high likelihood correlation lengths

▸ Solution: Use multiple

Gaussian process and weight them.

▸ Akin to particle filters ▸ More computational time, less

parameters.

Can an Artificial-Intelligence Win a Nobel Prize

SLIDE 35

+ 1 useless parameter=>

X C(X)

X (a) (b) C(X ) X (a) (b) t V C(X )

Test of machine learning model

6 parameters for start and  end of ramps

▸ Can the machine learner identify the useless parameter?

Can an Artificial-Intelligence Win a Nobel Prize

SLIDE 36

Results II

ML (6p)

▲ ■

NM (7p) ML (7p)

▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

evaluation 20 40 60 80 100 120 0.0

0.5
1.0
1.5
2.0
2.5

cost

0.2
0.1

0.0 0.1 0.2

2.4
2.2
2.0
1.8
1.6
1.4
1.2
1.0

normalised parameters cost normalised parameters

Thermal BEC

▸ Machine learner correctly identified

the useless parameter.

▸ Performed much better after

parameter was removed.

Can an Artificial-Intelligence Win a Nobel Prize

SLIDE 37

Conclusions

▸ Automated optimization of quantum experiments works even

for high dimensions: 1016 vs 30 (ML).

▸ Machine learner => Faster than NM ▸ Produces model, predicts importance of parameters. ▸ Can take into account uncertainty in measurements. ▸ Can pick points with most uncertainty or minimum cost (or

something in between)

▸ Uses fast gradient methods on predicted model to find

ptimum.

Can an Artificial-Intelligence Win a Nobel Prize

SLIDE 38

M-LOOP is available now!

▸ Google M-LOOP or go to: 

m-loop.readthedocs.io

Can an Artificial-Intelligence Win a Nobel Prize

SLIDE 39

Continuing Work

Can an Artificial-Intelligence Win a Nobel Prize

Gaussian Process Deep Neural Net

▸ More parameters ▸ Big data

Quantum  memories Gravity  Waves Quantum  transport

SLIDE 40

Cross section of Landscape

Can an Artificial-Intelligence Win a Nobel Prize

SLIDE 41

Discovered ramps

Can an Artificial-Intelligence Win a Nobel Prize

SLIDE 42

Machine learning

▸ Giving computers the ability

to learn without being explicitly programmed.

http://blog.stephenwolfram.com/2015/05/  wolfram-language-artificial-intelligence-the-image-identification-project/ Can an Artificial-Intelligence Win a Nobel Prize

SLIDE 43

Machine learning workflow

▸ Determine your problem: Classification, regression ▸ Pick a statistical model: Neural nets, Gaussian processes ▸ Fit your model to a (large) data set: Maximum likelihood ▸ Use the model to make predictions.

Image/Parameters Learner

Can an Artificial-Intelligence Win a Nobel Prize