SlideShare a Scribd company logo
1 of 13
Download to read offline
Large Scale Problems in Practice:
The effect of dimensionality on the
interaction among variables
Fabio Caraffini∗, Ferrante Neri∗ and Giovanni Iacca∗∗
∗De Montfort University, Leicetser, UK
∗∗RWTH Aachen University, Germany
EvoKNOW, Amsterdam, April 2017
Large scale optimisation problems LSOPs in practice: experimental set-up Results Conclusions
Large Scale Optimisation Problem (LSOP)
LSOPs are optimisation problems plagued by a high number of
design variables (hundreds or even thousands!).
A strict definition cannot be given for LSOP
but such problems are becoming more and more frequent
thus requires further investigation!
Fabio Caraffini EvoKNOW 2017
Large scale optimisation problems LSOPs in practice: experimental set-up Results Conclusions
Hard (large scale) life:
curse of dimensionality
The size of a domain grows
exponentially with the number of
dimensions!
The search space cannot be
covered:
large pop size = high structural
bias
large pop size = no convergence.
Fabio Caraffini EvoKNOW 2017
Large scale optimisation problems LSOPs in practice: experimental set-up Results Conclusions
Hard (large scale) life:
curse of dimensionality
Dimensionality also affects intrinsic features
of the search space itself.
As a square has four sides (1D lines), a cube has six sides (2D
squares) and a tesseract has eight sides (3D cubes):
thus a tesseract of side x has a high “surface volume” of 8x3
.
Similary, hyper-spherical domains behave similarly as:
their surface is Sn(r) = 2π
n
2
Γ( n
2 ) rn−1
,
their volume Vn(r) = π
n
2
Γ( n
2 +1) rn
,
with a ratio Vn(r)
Sn(r) ∝ 1
n .
In high dimensions uniformly sampled points are likely to be lo-
cated on the surface as the volume is a small fraction of the search
space!
Fabio Caraffini EvoKNOW 2017
Large scale optimisation problems LSOPs in practice: experimental set-up Results Conclusions
How to best address LSOPs then?
We may think that, in such a vast space, exploration plays the major role
however, the most successfully algorithms in the literature implement
exploitation mechanisms:
by employing local searchers,
by means of µ-populations,
by decomposing the search space,
by addressing a variable at a time. . .
Fabio Caraffini EvoKNOW 2017
Large scale optimisation problems LSOPs in practice: experimental set-up Results Conclusions
What is the common denominator of these approaches?
How can it be formally explained?
The search logic attempts to quickly achieve improvements with
mechanisms similar to those used for separable problems.
We tried and give a formal explanation of this phenomenon by looking at
the correlation amongst design variables (despite correlation=separability):
On the basis of the usual algorithm and experimental setting, what
happens to the correlation among the variables when the
dimensionality grows?
Fabio Caraffini EvoKNOW 2017
Large scale optimisation problems LSOPs in practice: experimental set-up Results Conclusions
Estimating the Correlation between Pairs of Variables
Step 1: “covariance matrix adaptation”.
We “evolved” a covariance matrix C by borrowing the same working
principle of the CMA-ES algorithm with rank-µ-update and weighted
recombination.
As the covariance matrix adapts itself to a basin of a ac traction, it
can be seen as an approximation of the Hessian matrix, thus giving us
information on the local behaviour of the fitness landscape.
Fabio Caraffini EvoKNOW 2017
Large scale optimisation problems LSOPs in practice: experimental set-up Results Conclusions
Estimating the Correlation between Pairs of Variables
Step 2: “correlation index”.
From C we worked out the Pearson (ρ) and the Spearman (τ) correlations:
|ρi,j | =
Ci,j
√
Ci,i Cj,j
=





1 |ρ1,2| |ρ1,3| ... |ρ1,n|
X 1 |ρ2,3| ... |ρ2,n|
X X 1 ... |ρ3,n|
... ... ... ... ...
X X X X 1





|τi,j | =
m
k=1(rk,i − ¯Ri
) m
k=1(rk,j − ¯Rj
)
m
k=1(rk,i − ¯Ri
)
2 m
k=1(rk,j − ¯Rj
)
2
=





1 |τ1,2| |τ1,3| ... |τ1,n|
X 1 |τ2,3| ... |τ2,n|
X X 1 ... |τ3,n|
... ... ... ... ...
X X X X 1





to obtain the global Pearson (ς) and the Spearman (ϕ) correlation indices:
ς = 2
n2−n
n−1
i=1
n
j=i+1 |ρi,j | ϕ = 2
n2−n
n−1
i=1
n
j=i+1 |τi,j |
Fabio Caraffini EvoKNOW 2017
Large scale optimisation problems LSOPs in practice: experimental set-up Results Conclusions
Test problems
We checked (50 runs) ς and ϕ over
19 test problems from SISC2010 in 10, 30, 50, 100, 500 and
1000 dimensions;
28 test problems from CEC2013 in 10, 30, 50 dimensions;
24 test problems from BBOB2010 in 10, 30, 50 and 100
dimensions;
15 test problems from CEC2013_LSGO in 1000 dimensions.
Fabio Caraffini EvoKNOW 2017
Large scale optimisation problems LSOPs in practice: experimental set-up Results Conclusions
Results
Graphical representation
Figure: Correlation indices for
f13 of SISC2010
Figure: Correlation indices for f8
of CEC2013.
Fabio Caraffini EvoKNOW 2017
Large scale optimisation problems LSOPs in practice: experimental set-up Results Conclusions
Results
Observations
For all the 86 problems it appears clear that the correlation
amongst variables decay when the dimensionality.
As a general trend, optimization problems with at least 100
dimensions show a weak correlation, while problems in 500 and
1000 dimensions show a nearly null correlation.
Regardless of the specific problem, all the LSOPs appear
always characterized by uncorrelated variables.
Separable problems ⇒ low correlation regardless of the n value
(N.B. low correlation separability!)
Fabio Caraffini EvoKNOW 2017
Large scale optimisation problems LSOPs in practice: experimental set-up Results Conclusions
Conclusions and way forward
We noted that in practice common experimental conditions
impose a growing shallowness of the search with the increase
of dimensionality.
Under these conditions LSOPs have a weak correlation
between variables, thus, a practically efficient approach is to
avoid exploratory diagonal moves and exploit the directions
along the axes.
Further studies will propose a lighter replacement for the
CMA-ES part, as well as algorithmic components exploiting
the knowledge from the (weak) correlation between pairs of
variables.
Fabio Caraffini EvoKNOW 2017
Large scale optimisation problems LSOPs in practice: experimental set-up Results Conclusions
Fabio Caraffini EvoKNOW 2017

More Related Content

Similar to Evoknow17 Large Scale Problems in Practice

November, 2006 CCKM'06 1
November, 2006 CCKM'06 1 November, 2006 CCKM'06 1
November, 2006 CCKM'06 1 butest
 
Tutorial rpo
Tutorial rpoTutorial rpo
Tutorial rpomosi2005
 
A General Purpose Exact Solution Method For Mixed Integer Concave Minimizatio...
A General Purpose Exact Solution Method For Mixed Integer Concave Minimizatio...A General Purpose Exact Solution Method For Mixed Integer Concave Minimizatio...
A General Purpose Exact Solution Method For Mixed Integer Concave Minimizatio...Martha Brown
 
Audit report[rollno 49]
Audit report[rollno 49]Audit report[rollno 49]
Audit report[rollno 49]RAHULROHAM2
 
Regression Concept Vectors for Deep Learning
Regression Concept Vectors for Deep LearningRegression Concept Vectors for Deep Learning
Regression Concept Vectors for Deep LearningMara Graziani
 
Joint contrastive learning with infinite possibilities
Joint contrastive learning with infinite possibilitiesJoint contrastive learning with infinite possibilities
Joint contrastive learning with infinite possibilitiestaeseon ryu
 
IMAGE REGISTRATION USING ADVANCED TOPOLOGY PRESERVING RELAXATION LABELING
IMAGE REGISTRATION USING ADVANCED TOPOLOGY PRESERVING RELAXATION LABELING IMAGE REGISTRATION USING ADVANCED TOPOLOGY PRESERVING RELAXATION LABELING
IMAGE REGISTRATION USING ADVANCED TOPOLOGY PRESERVING RELAXATION LABELING cscpconf
 
IMAGE REGISTRATION USING ADVANCED TOPOLOGY PRESERVING RELAXATION LABELING
IMAGE REGISTRATION USING ADVANCED TOPOLOGY PRESERVING RELAXATION LABELING IMAGE REGISTRATION USING ADVANCED TOPOLOGY PRESERVING RELAXATION LABELING
IMAGE REGISTRATION USING ADVANCED TOPOLOGY PRESERVING RELAXATION LABELING csandit
 
result analysis for deep leakage from gradients
result analysis for deep leakage from gradientsresult analysis for deep leakage from gradients
result analysis for deep leakage from gradients國騰 丁
 
Detail description of scientific publications
Detail description of scientific publicationsDetail description of scientific publications
Detail description of scientific publicationsManolis Vavalis
 
Thesis_Presentation
Thesis_PresentationThesis_Presentation
Thesis_PresentationMaruf Alam
 
LNCS 5050 - Bilevel Optimization and Machine Learning
LNCS 5050 - Bilevel Optimization and Machine LearningLNCS 5050 - Bilevel Optimization and Machine Learning
LNCS 5050 - Bilevel Optimization and Machine Learningbutest
 
Saliency Based Hookworm and Infection Detection for Wireless Capsule Endoscop...
Saliency Based Hookworm and Infection Detection for Wireless Capsule Endoscop...Saliency Based Hookworm and Infection Detection for Wireless Capsule Endoscop...
Saliency Based Hookworm and Infection Detection for Wireless Capsule Endoscop...IRJET Journal
 
教師なし画像特徴表現学習の動向 {Un, Self} supervised representation learning (CVPR 2018 完全読破...
教師なし画像特徴表現学習の動向 {Un, Self} supervised representation learning (CVPR 2018 完全読破...教師なし画像特徴表現学習の動向 {Un, Self} supervised representation learning (CVPR 2018 完全読破...
教師なし画像特徴表現学習の動向 {Un, Self} supervised representation learning (CVPR 2018 完全読破...cvpaper. challenge
 
Recommender system
Recommender systemRecommender system
Recommender systemBhumi Patel
 
Cannonical Correlation
Cannonical CorrelationCannonical Correlation
Cannonical Correlationdomsr
 
Cannonical correlation
Cannonical correlationCannonical correlation
Cannonical correlationdomsr
 
IEEE Pattern analysis and machine intelligence 2016 Title and Abstract
IEEE Pattern analysis and machine intelligence 2016 Title and AbstractIEEE Pattern analysis and machine intelligence 2016 Title and Abstract
IEEE Pattern analysis and machine intelligence 2016 Title and Abstracttsysglobalsolutions
 
Deep learning ensembles loss landscape
Deep learning ensembles loss landscapeDeep learning ensembles loss landscape
Deep learning ensembles loss landscapeDevansh16
 

Similar to Evoknow17 Large Scale Problems in Practice (20)

November, 2006 CCKM'06 1
November, 2006 CCKM'06 1 November, 2006 CCKM'06 1
November, 2006 CCKM'06 1
 
Tutorial rpo
Tutorial rpoTutorial rpo
Tutorial rpo
 
A General Purpose Exact Solution Method For Mixed Integer Concave Minimizatio...
A General Purpose Exact Solution Method For Mixed Integer Concave Minimizatio...A General Purpose Exact Solution Method For Mixed Integer Concave Minimizatio...
A General Purpose Exact Solution Method For Mixed Integer Concave Minimizatio...
 
Audit report[rollno 49]
Audit report[rollno 49]Audit report[rollno 49]
Audit report[rollno 49]
 
Regression Concept Vectors for Deep Learning
Regression Concept Vectors for Deep LearningRegression Concept Vectors for Deep Learning
Regression Concept Vectors for Deep Learning
 
Joint contrastive learning with infinite possibilities
Joint contrastive learning with infinite possibilitiesJoint contrastive learning with infinite possibilities
Joint contrastive learning with infinite possibilities
 
IMAGE REGISTRATION USING ADVANCED TOPOLOGY PRESERVING RELAXATION LABELING
IMAGE REGISTRATION USING ADVANCED TOPOLOGY PRESERVING RELAXATION LABELING IMAGE REGISTRATION USING ADVANCED TOPOLOGY PRESERVING RELAXATION LABELING
IMAGE REGISTRATION USING ADVANCED TOPOLOGY PRESERVING RELAXATION LABELING
 
IMAGE REGISTRATION USING ADVANCED TOPOLOGY PRESERVING RELAXATION LABELING
IMAGE REGISTRATION USING ADVANCED TOPOLOGY PRESERVING RELAXATION LABELING IMAGE REGISTRATION USING ADVANCED TOPOLOGY PRESERVING RELAXATION LABELING
IMAGE REGISTRATION USING ADVANCED TOPOLOGY PRESERVING RELAXATION LABELING
 
result analysis for deep leakage from gradients
result analysis for deep leakage from gradientsresult analysis for deep leakage from gradients
result analysis for deep leakage from gradients
 
Detail description of scientific publications
Detail description of scientific publicationsDetail description of scientific publications
Detail description of scientific publications
 
Thesis_Presentation
Thesis_PresentationThesis_Presentation
Thesis_Presentation
 
LNCS 5050 - Bilevel Optimization and Machine Learning
LNCS 5050 - Bilevel Optimization and Machine LearningLNCS 5050 - Bilevel Optimization and Machine Learning
LNCS 5050 - Bilevel Optimization and Machine Learning
 
Saliency Based Hookworm and Infection Detection for Wireless Capsule Endoscop...
Saliency Based Hookworm and Infection Detection for Wireless Capsule Endoscop...Saliency Based Hookworm and Infection Detection for Wireless Capsule Endoscop...
Saliency Based Hookworm and Infection Detection for Wireless Capsule Endoscop...
 
教師なし画像特徴表現学習の動向 {Un, Self} supervised representation learning (CVPR 2018 完全読破...
教師なし画像特徴表現学習の動向 {Un, Self} supervised representation learning (CVPR 2018 完全読破...教師なし画像特徴表現学習の動向 {Un, Self} supervised representation learning (CVPR 2018 完全読破...
教師なし画像特徴表現学習の動向 {Un, Self} supervised representation learning (CVPR 2018 完全読破...
 
Recommender system
Recommender systemRecommender system
Recommender system
 
Cannonical Correlation
Cannonical CorrelationCannonical Correlation
Cannonical Correlation
 
Cannonical correlation
Cannonical correlationCannonical correlation
Cannonical correlation
 
IEEE Pattern analysis and machine intelligence 2016 Title and Abstract
IEEE Pattern analysis and machine intelligence 2016 Title and AbstractIEEE Pattern analysis and machine intelligence 2016 Title and Abstract
IEEE Pattern analysis and machine intelligence 2016 Title and Abstract
 
1607.01152.pdf
1607.01152.pdf1607.01152.pdf
1607.01152.pdf
 
Deep learning ensembles loss landscape
Deep learning ensembles loss landscapeDeep learning ensembles loss landscape
Deep learning ensembles loss landscape
 

More from Fabio Caraffini

The Importance of Being Structured
The Importance of Being StructuredThe Importance of Being Structured
The Importance of Being StructuredFabio Caraffini
 
A seriously simple memetic approach with a high performance
A seriously simple memetic approach with a high performanceA seriously simple memetic approach with a high performance
A seriously simple memetic approach with a high performanceFabio Caraffini
 
Three rotational invariant variants of the 3SOME algorithms
Three rotational invariant variants of the 3SOME algorithmsThree rotational invariant variants of the 3SOME algorithms
Three rotational invariant variants of the 3SOME algorithmsFabio Caraffini
 
Meta-Lamarckian 3some algorithm for real-valued optimization
Meta-Lamarckian 3some algorithm for real-valued optimizationMeta-Lamarckian 3some algorithm for real-valued optimization
Meta-Lamarckian 3some algorithm for real-valued optimizationFabio Caraffini
 
Micro Differential Evolution with Extra Moves alonf the Axes
Micro Differential Evolution with Extra Moves alonf the AxesMicro Differential Evolution with Extra Moves alonf the Axes
Micro Differential Evolution with Extra Moves alonf the AxesFabio Caraffini
 
Evo star2012 Robot Base Disturbance Optimization with Compact Differential Ev...
Evo star2012 Robot Base Disturbance Optimization with Compact Differential Ev...Evo star2012 Robot Base Disturbance Optimization with Compact Differential Ev...
Evo star2012 Robot Base Disturbance Optimization with Compact Differential Ev...Fabio Caraffini
 

More from Fabio Caraffini (9)

The Importance of Being Structured
The Importance of Being StructuredThe Importance of Being Structured
The Importance of Being Structured
 
A seriously simple memetic approach with a high performance
A seriously simple memetic approach with a high performanceA seriously simple memetic approach with a high performance
A seriously simple memetic approach with a high performance
 
Three rotational invariant variants of the 3SOME algorithms
Three rotational invariant variants of the 3SOME algorithmsThree rotational invariant variants of the 3SOME algorithms
Three rotational invariant variants of the 3SOME algorithms
 
Ri-some algorithm
Ri-some algorithmRi-some algorithm
Ri-some algorithm
 
Meta-Lamarckian 3some algorithm for real-valued optimization
Meta-Lamarckian 3some algorithm for real-valued optimizationMeta-Lamarckian 3some algorithm for real-valued optimization
Meta-Lamarckian 3some algorithm for real-valued optimization
 
Micro Differential Evolution with Extra Moves alonf the Axes
Micro Differential Evolution with Extra Moves alonf the AxesMicro Differential Evolution with Extra Moves alonf the Axes
Micro Differential Evolution with Extra Moves alonf the Axes
 
Evo star2012 Robot Base Disturbance Optimization with Compact Differential Ev...
Evo star2012 Robot Base Disturbance Optimization with Compact Differential Ev...Evo star2012 Robot Base Disturbance Optimization with Compact Differential Ev...
Evo star2012 Robot Base Disturbance Optimization with Compact Differential Ev...
 
Pechakucha
PechakuchaPechakucha
Pechakucha
 
DEFENSE
DEFENSEDEFENSE
DEFENSE
 

Recently uploaded

SBFT Tool Competition 2024 -- Python Test Case Generation Track
SBFT Tool Competition 2024 -- Python Test Case Generation TrackSBFT Tool Competition 2024 -- Python Test Case Generation Track
SBFT Tool Competition 2024 -- Python Test Case Generation TrackSebastiano Panichella
 
Russian Call Girls in Kolkata Vaishnavi 🤌 8250192130 🚀 Vip Call Girls Kolkata
Russian Call Girls in Kolkata Vaishnavi 🤌  8250192130 🚀 Vip Call Girls KolkataRussian Call Girls in Kolkata Vaishnavi 🤌  8250192130 🚀 Vip Call Girls Kolkata
Russian Call Girls in Kolkata Vaishnavi 🤌 8250192130 🚀 Vip Call Girls Kolkataanamikaraghav4
 
CTAC 2024 Valencia - Henrik Hanke - Reduce to the max - slideshare.pdf
CTAC 2024 Valencia - Henrik Hanke - Reduce to the max - slideshare.pdfCTAC 2024 Valencia - Henrik Hanke - Reduce to the max - slideshare.pdf
CTAC 2024 Valencia - Henrik Hanke - Reduce to the max - slideshare.pdfhenrik385807
 
Genesis part 2 Isaiah Scudder 04-24-2024.pptx
Genesis part 2 Isaiah Scudder 04-24-2024.pptxGenesis part 2 Isaiah Scudder 04-24-2024.pptx
Genesis part 2 Isaiah Scudder 04-24-2024.pptxFamilyWorshipCenterD
 
Call Girls in Sarojini Nagar Market Delhi 💯 Call Us 🔝8264348440🔝
Call Girls in Sarojini Nagar Market Delhi 💯 Call Us 🔝8264348440🔝Call Girls in Sarojini Nagar Market Delhi 💯 Call Us 🔝8264348440🔝
Call Girls in Sarojini Nagar Market Delhi 💯 Call Us 🔝8264348440🔝soniya singh
 
George Lever - eCommerce Day Chile 2024
George Lever -  eCommerce Day Chile 2024George Lever -  eCommerce Day Chile 2024
George Lever - eCommerce Day Chile 2024eCommerce Institute
 
The 3rd Intl. Workshop on NL-based Software Engineering
The 3rd Intl. Workshop on NL-based Software EngineeringThe 3rd Intl. Workshop on NL-based Software Engineering
The 3rd Intl. Workshop on NL-based Software EngineeringSebastiano Panichella
 
WhatsApp 📞 9892124323 ✅Call Girls In Juhu ( Mumbai )
WhatsApp 📞 9892124323 ✅Call Girls In Juhu ( Mumbai )WhatsApp 📞 9892124323 ✅Call Girls In Juhu ( Mumbai )
WhatsApp 📞 9892124323 ✅Call Girls In Juhu ( Mumbai )Pooja Nehwal
 
Navi Mumbai Call Girls Service Pooja 9892124323 Real Russian Girls Looking Mo...
Navi Mumbai Call Girls Service Pooja 9892124323 Real Russian Girls Looking Mo...Navi Mumbai Call Girls Service Pooja 9892124323 Real Russian Girls Looking Mo...
Navi Mumbai Call Girls Service Pooja 9892124323 Real Russian Girls Looking Mo...Pooja Nehwal
 
Call Girls in Rohini Delhi 💯Call Us 🔝8264348440🔝
Call Girls in Rohini Delhi 💯Call Us 🔝8264348440🔝Call Girls in Rohini Delhi 💯Call Us 🔝8264348440🔝
Call Girls in Rohini Delhi 💯Call Us 🔝8264348440🔝soniya singh
 
call girls in delhi malviya nagar @9811711561@
call girls in delhi malviya nagar @9811711561@call girls in delhi malviya nagar @9811711561@
call girls in delhi malviya nagar @9811711561@vikas rana
 
Philippine History cavite Mutiny Report.ppt
Philippine History cavite Mutiny Report.pptPhilippine History cavite Mutiny Report.ppt
Philippine History cavite Mutiny Report.pptssuser319dad
 
OSCamp Kubernetes 2024 | SRE Challenges in Monolith to Microservices Shift at...
OSCamp Kubernetes 2024 | SRE Challenges in Monolith to Microservices Shift at...OSCamp Kubernetes 2024 | SRE Challenges in Monolith to Microservices Shift at...
OSCamp Kubernetes 2024 | SRE Challenges in Monolith to Microservices Shift at...NETWAYS
 
NATIONAL ANTHEMS OF AFRICA (National Anthems of Africa)
NATIONAL ANTHEMS OF AFRICA (National Anthems of Africa)NATIONAL ANTHEMS OF AFRICA (National Anthems of Africa)
NATIONAL ANTHEMS OF AFRICA (National Anthems of Africa)Basil Achie
 
Simulation-based Testing of Unmanned Aerial Vehicles with Aerialist
Simulation-based Testing of Unmanned Aerial Vehicles with AerialistSimulation-based Testing of Unmanned Aerial Vehicles with Aerialist
Simulation-based Testing of Unmanned Aerial Vehicles with AerialistSebastiano Panichella
 
Microsoft Copilot AI for Everyone - created by AI
Microsoft Copilot AI for Everyone - created by AIMicrosoft Copilot AI for Everyone - created by AI
Microsoft Copilot AI for Everyone - created by AITatiana Gurgel
 
SBFT Tool Competition 2024 - CPS-UAV Test Case Generation Track
SBFT Tool Competition 2024 - CPS-UAV Test Case Generation TrackSBFT Tool Competition 2024 - CPS-UAV Test Case Generation Track
SBFT Tool Competition 2024 - CPS-UAV Test Case Generation TrackSebastiano Panichella
 
Night 7k Call Girls Noida Sector 128 Call Me: 8448380779
Night 7k Call Girls Noida Sector 128 Call Me: 8448380779Night 7k Call Girls Noida Sector 128 Call Me: 8448380779
Night 7k Call Girls Noida Sector 128 Call Me: 8448380779Delhi Call girls
 
Governance and Nation-Building in Nigeria: Some Reflections on Options for Po...
Governance and Nation-Building in Nigeria: Some Reflections on Options for Po...Governance and Nation-Building in Nigeria: Some Reflections on Options for Po...
Governance and Nation-Building in Nigeria: Some Reflections on Options for Po...Kayode Fayemi
 
Open Source Camp Kubernetes 2024 | Running WebAssembly on Kubernetes by Alex ...
Open Source Camp Kubernetes 2024 | Running WebAssembly on Kubernetes by Alex ...Open Source Camp Kubernetes 2024 | Running WebAssembly on Kubernetes by Alex ...
Open Source Camp Kubernetes 2024 | Running WebAssembly on Kubernetes by Alex ...NETWAYS
 

Recently uploaded (20)

SBFT Tool Competition 2024 -- Python Test Case Generation Track
SBFT Tool Competition 2024 -- Python Test Case Generation TrackSBFT Tool Competition 2024 -- Python Test Case Generation Track
SBFT Tool Competition 2024 -- Python Test Case Generation Track
 
Russian Call Girls in Kolkata Vaishnavi 🤌 8250192130 🚀 Vip Call Girls Kolkata
Russian Call Girls in Kolkata Vaishnavi 🤌  8250192130 🚀 Vip Call Girls KolkataRussian Call Girls in Kolkata Vaishnavi 🤌  8250192130 🚀 Vip Call Girls Kolkata
Russian Call Girls in Kolkata Vaishnavi 🤌 8250192130 🚀 Vip Call Girls Kolkata
 
CTAC 2024 Valencia - Henrik Hanke - Reduce to the max - slideshare.pdf
CTAC 2024 Valencia - Henrik Hanke - Reduce to the max - slideshare.pdfCTAC 2024 Valencia - Henrik Hanke - Reduce to the max - slideshare.pdf
CTAC 2024 Valencia - Henrik Hanke - Reduce to the max - slideshare.pdf
 
Genesis part 2 Isaiah Scudder 04-24-2024.pptx
Genesis part 2 Isaiah Scudder 04-24-2024.pptxGenesis part 2 Isaiah Scudder 04-24-2024.pptx
Genesis part 2 Isaiah Scudder 04-24-2024.pptx
 
Call Girls in Sarojini Nagar Market Delhi 💯 Call Us 🔝8264348440🔝
Call Girls in Sarojini Nagar Market Delhi 💯 Call Us 🔝8264348440🔝Call Girls in Sarojini Nagar Market Delhi 💯 Call Us 🔝8264348440🔝
Call Girls in Sarojini Nagar Market Delhi 💯 Call Us 🔝8264348440🔝
 
George Lever - eCommerce Day Chile 2024
George Lever -  eCommerce Day Chile 2024George Lever -  eCommerce Day Chile 2024
George Lever - eCommerce Day Chile 2024
 
The 3rd Intl. Workshop on NL-based Software Engineering
The 3rd Intl. Workshop on NL-based Software EngineeringThe 3rd Intl. Workshop on NL-based Software Engineering
The 3rd Intl. Workshop on NL-based Software Engineering
 
WhatsApp 📞 9892124323 ✅Call Girls In Juhu ( Mumbai )
WhatsApp 📞 9892124323 ✅Call Girls In Juhu ( Mumbai )WhatsApp 📞 9892124323 ✅Call Girls In Juhu ( Mumbai )
WhatsApp 📞 9892124323 ✅Call Girls In Juhu ( Mumbai )
 
Navi Mumbai Call Girls Service Pooja 9892124323 Real Russian Girls Looking Mo...
Navi Mumbai Call Girls Service Pooja 9892124323 Real Russian Girls Looking Mo...Navi Mumbai Call Girls Service Pooja 9892124323 Real Russian Girls Looking Mo...
Navi Mumbai Call Girls Service Pooja 9892124323 Real Russian Girls Looking Mo...
 
Call Girls in Rohini Delhi 💯Call Us 🔝8264348440🔝
Call Girls in Rohini Delhi 💯Call Us 🔝8264348440🔝Call Girls in Rohini Delhi 💯Call Us 🔝8264348440🔝
Call Girls in Rohini Delhi 💯Call Us 🔝8264348440🔝
 
call girls in delhi malviya nagar @9811711561@
call girls in delhi malviya nagar @9811711561@call girls in delhi malviya nagar @9811711561@
call girls in delhi malviya nagar @9811711561@
 
Philippine History cavite Mutiny Report.ppt
Philippine History cavite Mutiny Report.pptPhilippine History cavite Mutiny Report.ppt
Philippine History cavite Mutiny Report.ppt
 
OSCamp Kubernetes 2024 | SRE Challenges in Monolith to Microservices Shift at...
OSCamp Kubernetes 2024 | SRE Challenges in Monolith to Microservices Shift at...OSCamp Kubernetes 2024 | SRE Challenges in Monolith to Microservices Shift at...
OSCamp Kubernetes 2024 | SRE Challenges in Monolith to Microservices Shift at...
 
NATIONAL ANTHEMS OF AFRICA (National Anthems of Africa)
NATIONAL ANTHEMS OF AFRICA (National Anthems of Africa)NATIONAL ANTHEMS OF AFRICA (National Anthems of Africa)
NATIONAL ANTHEMS OF AFRICA (National Anthems of Africa)
 
Simulation-based Testing of Unmanned Aerial Vehicles with Aerialist
Simulation-based Testing of Unmanned Aerial Vehicles with AerialistSimulation-based Testing of Unmanned Aerial Vehicles with Aerialist
Simulation-based Testing of Unmanned Aerial Vehicles with Aerialist
 
Microsoft Copilot AI for Everyone - created by AI
Microsoft Copilot AI for Everyone - created by AIMicrosoft Copilot AI for Everyone - created by AI
Microsoft Copilot AI for Everyone - created by AI
 
SBFT Tool Competition 2024 - CPS-UAV Test Case Generation Track
SBFT Tool Competition 2024 - CPS-UAV Test Case Generation TrackSBFT Tool Competition 2024 - CPS-UAV Test Case Generation Track
SBFT Tool Competition 2024 - CPS-UAV Test Case Generation Track
 
Night 7k Call Girls Noida Sector 128 Call Me: 8448380779
Night 7k Call Girls Noida Sector 128 Call Me: 8448380779Night 7k Call Girls Noida Sector 128 Call Me: 8448380779
Night 7k Call Girls Noida Sector 128 Call Me: 8448380779
 
Governance and Nation-Building in Nigeria: Some Reflections on Options for Po...
Governance and Nation-Building in Nigeria: Some Reflections on Options for Po...Governance and Nation-Building in Nigeria: Some Reflections on Options for Po...
Governance and Nation-Building in Nigeria: Some Reflections on Options for Po...
 
Open Source Camp Kubernetes 2024 | Running WebAssembly on Kubernetes by Alex ...
Open Source Camp Kubernetes 2024 | Running WebAssembly on Kubernetes by Alex ...Open Source Camp Kubernetes 2024 | Running WebAssembly on Kubernetes by Alex ...
Open Source Camp Kubernetes 2024 | Running WebAssembly on Kubernetes by Alex ...
 

Evoknow17 Large Scale Problems in Practice

  • 1. Large Scale Problems in Practice: The effect of dimensionality on the interaction among variables Fabio Caraffini∗, Ferrante Neri∗ and Giovanni Iacca∗∗ ∗De Montfort University, Leicetser, UK ∗∗RWTH Aachen University, Germany EvoKNOW, Amsterdam, April 2017
  • 2. Large scale optimisation problems LSOPs in practice: experimental set-up Results Conclusions Large Scale Optimisation Problem (LSOP) LSOPs are optimisation problems plagued by a high number of design variables (hundreds or even thousands!). A strict definition cannot be given for LSOP but such problems are becoming more and more frequent thus requires further investigation! Fabio Caraffini EvoKNOW 2017
  • 3. Large scale optimisation problems LSOPs in practice: experimental set-up Results Conclusions Hard (large scale) life: curse of dimensionality The size of a domain grows exponentially with the number of dimensions! The search space cannot be covered: large pop size = high structural bias large pop size = no convergence. Fabio Caraffini EvoKNOW 2017
  • 4. Large scale optimisation problems LSOPs in practice: experimental set-up Results Conclusions Hard (large scale) life: curse of dimensionality Dimensionality also affects intrinsic features of the search space itself. As a square has four sides (1D lines), a cube has six sides (2D squares) and a tesseract has eight sides (3D cubes): thus a tesseract of side x has a high “surface volume” of 8x3 . Similary, hyper-spherical domains behave similarly as: their surface is Sn(r) = 2π n 2 Γ( n 2 ) rn−1 , their volume Vn(r) = π n 2 Γ( n 2 +1) rn , with a ratio Vn(r) Sn(r) ∝ 1 n . In high dimensions uniformly sampled points are likely to be lo- cated on the surface as the volume is a small fraction of the search space! Fabio Caraffini EvoKNOW 2017
  • 5. Large scale optimisation problems LSOPs in practice: experimental set-up Results Conclusions How to best address LSOPs then? We may think that, in such a vast space, exploration plays the major role however, the most successfully algorithms in the literature implement exploitation mechanisms: by employing local searchers, by means of µ-populations, by decomposing the search space, by addressing a variable at a time. . . Fabio Caraffini EvoKNOW 2017
  • 6. Large scale optimisation problems LSOPs in practice: experimental set-up Results Conclusions What is the common denominator of these approaches? How can it be formally explained? The search logic attempts to quickly achieve improvements with mechanisms similar to those used for separable problems. We tried and give a formal explanation of this phenomenon by looking at the correlation amongst design variables (despite correlation=separability): On the basis of the usual algorithm and experimental setting, what happens to the correlation among the variables when the dimensionality grows? Fabio Caraffini EvoKNOW 2017
  • 7. Large scale optimisation problems LSOPs in practice: experimental set-up Results Conclusions Estimating the Correlation between Pairs of Variables Step 1: “covariance matrix adaptation”. We “evolved” a covariance matrix C by borrowing the same working principle of the CMA-ES algorithm with rank-µ-update and weighted recombination. As the covariance matrix adapts itself to a basin of a ac traction, it can be seen as an approximation of the Hessian matrix, thus giving us information on the local behaviour of the fitness landscape. Fabio Caraffini EvoKNOW 2017
  • 8. Large scale optimisation problems LSOPs in practice: experimental set-up Results Conclusions Estimating the Correlation between Pairs of Variables Step 2: “correlation index”. From C we worked out the Pearson (ρ) and the Spearman (τ) correlations: |ρi,j | = Ci,j √ Ci,i Cj,j =      1 |ρ1,2| |ρ1,3| ... |ρ1,n| X 1 |ρ2,3| ... |ρ2,n| X X 1 ... |ρ3,n| ... ... ... ... ... X X X X 1      |τi,j | = m k=1(rk,i − ¯Ri ) m k=1(rk,j − ¯Rj ) m k=1(rk,i − ¯Ri ) 2 m k=1(rk,j − ¯Rj ) 2 =      1 |τ1,2| |τ1,3| ... |τ1,n| X 1 |τ2,3| ... |τ2,n| X X 1 ... |τ3,n| ... ... ... ... ... X X X X 1      to obtain the global Pearson (ς) and the Spearman (ϕ) correlation indices: ς = 2 n2−n n−1 i=1 n j=i+1 |ρi,j | ϕ = 2 n2−n n−1 i=1 n j=i+1 |τi,j | Fabio Caraffini EvoKNOW 2017
  • 9. Large scale optimisation problems LSOPs in practice: experimental set-up Results Conclusions Test problems We checked (50 runs) ς and ϕ over 19 test problems from SISC2010 in 10, 30, 50, 100, 500 and 1000 dimensions; 28 test problems from CEC2013 in 10, 30, 50 dimensions; 24 test problems from BBOB2010 in 10, 30, 50 and 100 dimensions; 15 test problems from CEC2013_LSGO in 1000 dimensions. Fabio Caraffini EvoKNOW 2017
  • 10. Large scale optimisation problems LSOPs in practice: experimental set-up Results Conclusions Results Graphical representation Figure: Correlation indices for f13 of SISC2010 Figure: Correlation indices for f8 of CEC2013. Fabio Caraffini EvoKNOW 2017
  • 11. Large scale optimisation problems LSOPs in practice: experimental set-up Results Conclusions Results Observations For all the 86 problems it appears clear that the correlation amongst variables decay when the dimensionality. As a general trend, optimization problems with at least 100 dimensions show a weak correlation, while problems in 500 and 1000 dimensions show a nearly null correlation. Regardless of the specific problem, all the LSOPs appear always characterized by uncorrelated variables. Separable problems ⇒ low correlation regardless of the n value (N.B. low correlation separability!) Fabio Caraffini EvoKNOW 2017
  • 12. Large scale optimisation problems LSOPs in practice: experimental set-up Results Conclusions Conclusions and way forward We noted that in practice common experimental conditions impose a growing shallowness of the search with the increase of dimensionality. Under these conditions LSOPs have a weak correlation between variables, thus, a practically efficient approach is to avoid exploratory diagonal moves and exploit the directions along the axes. Further studies will propose a lighter replacement for the CMA-ES part, as well as algorithmic components exploiting the knowledge from the (weak) correlation between pairs of variables. Fabio Caraffini EvoKNOW 2017
  • 13. Large scale optimisation problems LSOPs in practice: experimental set-up Results Conclusions Fabio Caraffini EvoKNOW 2017