Smart Data Analytics Research Group Overview

0

3

/

1

0

/

2

0

1

5

|

R

D

A

F

i

f

t

h

P

l

e

n

a

r

y

M

e

e

t

i

n

g

|

S

a

n

D

i

e

g

o

,

U

S

A

|

P

a

r

a

d

i

s

e

P

o

i

n

t

R

e

s

o

r

t

Markus Götz

Jülich Supercomputing Center (JSC) // University of Iceland

Morris Riedel

Jülich Supercomputing Center (JSC) // University of Iceland

B

i

g

D

a

t

a

I

n

t

e

r

e

s

t

G

r

o

u

p

S

m

a

r

t

D

a

t

a

A

n

a

l

y

t

i

c

s

O

u

t

l

i

n

e

Introduction



Research Group, Research Area

Smart Data Analytics Use Cases and Techniques



Classification, Land Cover Type, piSVM



Clustering, „Drunken Flies“, HPDBSCAN



Deep-Learning, Cortex Layers, pylearn CNN

Conclusion



Results and RDA Context

03/10/2015

2

Markus Götz | Smart Data Analytics | Forschungszentrum Jülich

03/10/2015

3

Research Group



Jülich Supercomputing Center (HPC/HTC)



High Productivity Data Processing Group

Research Area



Smart Data Analytics Methods



Evaluation and Development of Scalable Tools



Processing Platform Requirements



Application 

in

 Scientific Use Case

I

n

t

r

o

d

u

c

t

i

o

n

Markus Götz | Smart Data Analytics | Forschungszentrum Jülich

03/10/2015

4

Land Cover Type Problem



Collaboration with University of Iceland



Determine Land Cover Type in Satellite Images



Different Types - Road, Building, Vegetation, …

Classification



Supervised Learning Technique



Known Set of Groups or Classes



Determine Membership of New Items

C

l

a

s

s

i

f

i

c

a

t

i

o

n

/

/

L

a

n

d

C

o

v

e

r

T

y

p

e

Markus Götz | Smart Data Analytics | Forschungszentrum Jülich

03/10/2015

5

Approach



Support Vector Machines (SVM)



Existing Solution: piSVM (MPI)



In-house Optimization of Parallel Code

C

l

a

s

s

i

f

i

c

a

t

i

o

n

/

/

L

a

n

d

C

o

v

e

r

T

y

p

e

Markus Götz | Smart Data Analytics | Forschungszentrum Jülich

Area

Standard

deviation

Inertia

03/10/2015

6

„Drunken Flies“



Collaboration with University of Cologne



Investigate Influence of Genetics on Alcohol Consumption



Literally Make Flies Drunk

Clustering



Unsupervised Learning Technique



Subdivide Database into Similar Groups



Similarity Metrics

C

l

u

s

t

e

r

i

n

g

/

/

„

D

r

u

n

k

e

n

F

l

i

e

s

“

Markus Götz | Smart Data Analytics | Forschungszentrum Jülich

03/10/2015

7

C

l

u

s

t

e

r

i

n

g

/

/

„

D

r

u

n

k

e

n

F

l

i

e

s

“

Markus Götz | Smart Data Analytics | Forschungszentrum Jülich

Approach



Image Processing Pipeline



HPDBSCAN



In-house Development

(MPI+OpenMP)

03/10/2015

8

Cortex Layer Problem



Institute for Neuro-Medicine (INM) at FZJ



Segment the Seven Layers of the Cortex



Images of Actual Brain Slices



Each Gigabytes (60k square resolution)

Deep Learning



Supervised Learning Technique (Classification)



More Advanced Mathemical Models



Various Flavors of Neural Networks

D

e

e

p

L

e

a

r

n

i

n

g

/

/

C

o

r

t

e

x

L

a

y

e

r

s

Markus Götz | Smart Data Analytics | Forschungszentrum Jülich

03/10/2015

9

Approach



Convolutional Neural Networks



Existing Serial Toolkit



Pylearn 2

/Theano



Scaling Issues

D

e

e

p

L

e

a

r

n

i

n

g

/

/

C

o

r

t

e

x

L

a

y

e

r

s

Markus Götz | Smart Data Analytics | Forschungszentrum Jülich

03/10/2015

10

C

o

n

c

l

u

s

i

o

n

Markus Götz | Smart Data Analytics | Forschungszentrum Jülich

Results



Big Data Challenge is Real!



Gap between Analytics Requirements and Actual Implementations

Interest for RDA



Code is Open-source @ GitHub and Bitbucket



Data is Open and Freely Published @ B2SHARE



Choice of Dataformats



Question of Future Processing Platforms

03/10/2015

11

T

h

a

n

k

s

y

o

u

f

o

r

t

h

e

a

t

t

e

n

t

i

o

n

…

Markus Götz | Smart Data Analytics | Forschungszentrum Jülich

F

i

f

t

h

P

l

e

n

a

r

y

M

e

e

t

i

n

g

08 – 12 March 2015

San Diego, USA | Paradise Point Resort

C

o

n

t

a

c

t

:

m

.

g

o

e

t

z

@

f

z

-

j

u

e

l

i

c

h

.

d

e

S

l

i

d

e

s

:

B

i

g

D

a

t

a

I

G

>

W

i

k

i

>

5

t

h

P

l

e

n

a

r

y

This content provides an overview of the Smart Data Analytics Research Group at the Jülich Supercomputing Center and the University of Iceland. It covers their research areas such as classification, clustering, deep learning, and more. The group focuses on developing scalable tools and platforms for data analytics applications in scientific use cases. Collaboration with various institutions and the application of machine learning algorithms are highlighted.

• Data Analytics
• Research Group
• Machine Learning
• Jülich Supercomputing

---

---

Please find below an Image/Link to download the presentation.

The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author.If you encounter any issues during the download, it is possible that the publisher has removed the file from their server.

You are allowed to download the files provided on this website for personal or commercial use, subject to the condition that they are used lawfully. All files are the property of their respective owners.

The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author.

Presentation Transcript

---

1. Big Data Interest Group Smart Data Analytics Markus G tz J lich Supercomputing Center (JSC) // University of Iceland Member of the Helmholtz Association Morris Riedel J lich Supercomputing Center (JSC) // University of Iceland 03/10/2015 | RDA Fifth Plenary Meeting | San Diego, USA | Paradise Point Resort

2. Outline Introduction Research Group, Research Area Smart Data Analytics Use Cases and Techniques Classification, Land Cover Type, piSVM Clustering, Drunken Flies , HPDBSCAN Deep-Learning, Cortex Layers, pylearn CNN Conclusion Results and RDA Context Member of the Helmholtz Association 03/10/2015 Markus G tz | Smart Data Analytics | Forschungszentrum J lich 2

3. Introduction Research Group J lich Supercomputing Center (HPC/HTC) High Productivity Data Processing Group Parallel Data Analytics Data Mining Methods Machine Learning Algorithms Smart Data Analytics Research Area Smart Data Analytics Methods Evaluation and Development of Scalable Tools Processing Platform Requirements Application in Scientific Use Case Scientific Community Application Data Analzsis Tools Generic Data Methods Member of the Helmholtz Association 03/10/2015 Markus G tz | Smart Data Analytics | Forschungszentrum J lich 3

4. Classification // Land Cover Type Land Cover Type Problem Collaboration with University of Iceland Determine Land Cover Type in Satellite Images Different Types - Road, Building, Vegetation, Classification Supervised Learning Technique Known Set of Groups or Classes Determine Membership of New Items Member of the Helmholtz Association 03/10/2015 Markus G tz | Smart Data Analytics | Forschungszentrum J lich 4

5. Classification // Land Cover Type Approach Support Vector Machines (SVM) Existing Solution: piSVM (MPI) In-house Optimization of Parallel Code Member of the Helmholtz Association Inertia Standard deviation Area 03/10/2015 Markus G tz | Smart Data Analytics | Forschungszentrum J lich 5

6. Clustering // Drunken Flies Drunken Flies Collaboration with University of Cologne Investigate Influence of Genetics on Alcohol Consumption Literally Make Flies Drunk Clustering Unsupervised Learning Technique Subdivide Database into Similar Groups Similarity Metrics Member of the Helmholtz Association 03/10/2015 Markus G tz | Smart Data Analytics | Forschungszentrum J lich 6

7. Clustering // Drunken Flies Approach Image Processing Pipeline HPDBSCAN In-house Development (MPI+OpenMP) Member of the Helmholtz Association 03/10/2015 Markus G tz | Smart Data Analytics | Forschungszentrum J lich 7

8. Deep Learning // Cortex Layers Cortex Layer Problem Institute for Neuro-Medicine (INM) at FZJ Segment the Seven Layers of the Cortex Images of Actual Brain Slices Each Gigabytes (60k square resolution) Deep Learning Supervised Learning Technique (Classification) More Advanced Mathemical Models Various Flavors of Neural Networks Member of the Helmholtz Association 03/10/2015 Markus G tz | Smart Data Analytics | Forschungszentrum J lich 8

9. Deep Learning // Cortex Layers Approach Convolutional Neural Networks Existing Serial Toolkit Pylearn 2/Theano Scaling Issues Member of the Helmholtz Association 03/10/2015 Markus G tz | Smart Data Analytics | Forschungszentrum J lich 9

10. Conclusion Results Big Data Challenge is Real! Gap between Analytics Requirements and Actual Implementations Interest for RDA Code is Open-source @ GitHub and Bitbucket Data is Open and Freely Published @ B2SHARE Choice of Dataformats Question of Future Processing Platforms Member of the Helmholtz Association 03/10/2015 Markus G tz | Smart Data Analytics | Forschungszentrum J lich 10

11. Thanks you for the attention Fifth Plenary Meeting 08 12 March 2015 San Diego, USA | Paradise Point Resort Member of the Helmholtz Association Contact: m.goetz@fz-juelich.de Slides: Big Data IG > Wiki > 5th Plenary 03/10/2015 Markus G tz | Smart Data Analytics | Forschungszentrum J lich 11