how semantic technologies can help to cure hearing loss?

47
Insight Centre for Data Analytics

Upload: andre-freitas

Post on 14-Jul-2015

64 views

Category:

Software


0 download

TRANSCRIPT

Insight Centre for Data Analytics

How Semantic Technologies can

help to cure Hearing Loss

An Introduction to the SIFEM EU Project

André Freitas, Ratnesh Sahay

October 3rd, 2014

SIFEM Team

Kartik Asooja

João Jares

Marggie Jones

Oya Beyan

Yasar Khan

Stefan Decker

Ratnesh Sahay

André Freitas

Insight Centre for Data Analytics Slide 3

Outline

• Motivation: Modelling the Mechanics of Hearing

• Challenges in Contemporary Science

• Semantic Infrastructure

• Demonstration

• Take-away message

Insight Centre for Data Analytics Slide 4

Goals

• Discuss the challenges that contemporary scientific

practice faces

• Discuss how Semantic Technologies can help.

Insight Centre for Data Analytics Slide 5

Outside the computer science

community

Multi-scale Models

Insight Centre for Data Analytics Slide 6

Finite Element Models (Video)

Insight Centre for Data Analytics Slide 7

Geometrical Model

Insight Centre for Data Analytics Slide 8

Physics Model

• FE equilibrium for solid

• FE equilibrium for fluid

Insight Centre for Data Analytics Slide 9

Numerical Models/Solvers

• Incremental-iterative implicit solution scheme

Insight Centre for Data Analytics Slide 10

Experimental Data

• A

Insight Centre for Data Analytics Slide 11

Common Challenges for

‘Big Science’

Insight Centre for Data Analytics Slide 12

Coordination Complexity (L. Floridi)

Insight Centre for Data Analytics Slide 13

Reproducibility

Insight Centre for Data Analytics Slide 14

Efficiency &

Automation

Insight Centre for Data Analytics Slide 15

How to build an infrastructure which

addresses these dimensions?

Insight Centre for Data Analytics Slide 17

Characteristics of the SIFEM Domain

• Most data is at the numeric level

• Highly dependent on visualization (man in the middle)

Insight Centre for Data Analytics Slide 21

Characteristics of the SIFEM Domain

• Relatively small set of concepts

Insight Centre for Data Analytics Slide 22

Characteristics of the SIFEM Domain

• But difficult to represent • Physics, geometrical models, topological relations, algoithmic,

mathematics

Insight Centre for Data Analytics Slide 23

Semantic Infrastructure

• Coordination Complexity • Semantic Web standards and standardized vocabularies for

representing FE resources

• Simulation Platform with built-in standardized data representation

• One-stop shop for FE simulation resources (inner ear)

• Reproducibility • Web platform for sharing FE Simulations

• Simulation output as Linked Data

• Executable papers

• Efficiency & Automation • Facilitating data interpretation

• Attribution & Incentives • ORCID & Altmetrics

Insight Centre for Data Analytics Slide 24

Lid-driven cavity flow

Insight Centre for Data Analytics Slide 25

Physical Model

Solver

FEM Model

If there a vortex close to

the lid?

Automatic Interpretation

Insight Centre for Data Analytics Slide 26

Expected physical behavior (Experiment intent):

Velocity in X starts at zero at the bottom of the box

followed by a slow velocity decrease reaching a

minima which is followed by a very fast velocity

increase close to the lid.

Numeric Level

Symbolic Lifting

IF

Predicates

Physical Object

Insight Centre for Data Analytics Slide 27

Left wall

Bottom

Lid

Right wall

FE Core Conceptual Model

Insight Centre for Data Analytics Slide 28

FE Elements

Insight Centre for Data Analytics Slide 29

Cell

Patch

boundary

Patch

boundary

Mesh

Block

FE Elements

Insight Centre for Data Analytics Slide 30

Physics/Material Properties

Insight Centre for Data Analytics Slide 31

Kinematic

viscosity

Dynamic viscosity

(µ)/Fluid density(ρ)

Fluid Solid

Velocity

Velocity

Pressure

Navier-Stokes

Equation

Physics/Material Properties

Insight Centre for Data Analytics Slide 32

Insight Centre for Data Analytics Slide 33

are Connected

are Connected

are Connected

are Connected

Box

Wall

Is Part of

Fluid

is Inside

Topology

Topology

Insight Centre for Data Analytics Slide 34

Simulation Results

Insight Centre for Data Analytics Slide 35

FE Core Conceptual Model

Insight Centre for Data Analytics Slide 36

The SIFEM Conceptual Model

Insight Centre for Data Analytics Slide 37

Data View

Insight Centre for Data Analytics Slide 38

Data Selection

y

0.05

Feature Extraction (Symbolic Lifting)

Insight Centre for Data Analytics Slide 39

Minima=(0.055,-0.20)

fast increase

slow decrease

followed by

(avg first derivative > 35)

velocity starts

at 0 at the

bottom

maximum

velocity is 0.93

at the lid

Data Interpretation Statements

Insight Centre for Data Analytics Slide 40

:DataView1 :hasDimensionY :VelocityX .

:DataView1 :hasDimensionX :DistanceFromTheCavityBase .

:DataView1 :x0 “0.0"^^xsd:double .

:DataView1 :y0 “0.0"^^xsd:double .

:DataView1 :hasMinimumX “-0.055"^^xsd:double .

:DataView1 :hasMinimumY “-0.20"^^xsd:double .

:DataView1 :hasFeature :PositiveSecondDerivative .

:DataView1 :hasBehaviour :BehaviourRegion1 .

:DataView1 :hasBehaviour :BehaviourRegion2 .

:BehaviourRegion1 :avgFirstDerivative “-3.63"^^xsd:double .

:BehaviourRegion1 :hasFeature EndRegion .

:BehaviourRegion1 :hasFeature :Decreases .

:BehaviourRegion1 :hasFeature :DecreasesSlowly .

:BehaviourRegion2 :avgFirstDerivative “33.35"^^xsd:double .

:BehaviourRegion2 :hasFeature EndRegion .

:BehaviourRegion2 :hasFeature :Increases .

:BehaviourRegion2 :hasFeature :IncreasesFast .

:BehaviourRegion1 :isFollowedBy :BehaviourRegion1 .

: LidSimulation :hasInterpretation :ValidVelocityBehaviour .

Data Analysis Rule

Data Analysis Rules

Insight Centre for Data Analytics Slide 41

CONSTRUCT

{ :LidSimulation sif: hasInterpretation :ValidVelocityBehaviour }

WHERE {

?dataview rdf:type dao:DataView .

?dataview dao:hasFeature ?x .

...

}

IF( minima(velocity) is negative AND

decreases very slowly(velocity) AND

increases very fast (velocity) )

VALID VELOCITY BEHAVIOUR

SPARQL Rule

Data Analysis Workflow

Insight Centre for Data Analytics Slide 42

FE Data Analysis

Insight Centre for Data Analytics Slide 43

FE Data Analysis

Insight Centre for Data Analytics Slide 44

Insight Centre for Data Analytics Slide 45

Insight Centre for Data Analytics Slide 46

:DataView1 :hasDimensionY :BasilarMembraneMagnitude .

:DataView1 :hasDimensionX :DistanceFromTheCochleaBasis .

:DataView1 :hasFeature :isSingleWave .

:DataView1 :hasMaximumAmplitude “0.0031 "^^xsd:double.

:DataView1 :hasMaximumY “0.0020 e^-6 "^^xsd:double .

:DataView1 :hasMaximumX “14"^^xsd:double .

:DataView1 :hasMinimumY “-0.0011 e^-6 "^^xsd:double .

:DataView1 :hasMinimumX “17"^^xsd:double .

Architecture

Insight Centre for Data Analytics Slide 47

Infrastructure/Data Attribution

Insight Centre for Data Analytics Slide 48

Future Directions

• Finalization of the semantic infrastructure

• Explore heuristics for the automatic exploration of the

parameter space

• Replicate an existing scientific discovery

• Engage users

Insight Centre for Data Analytics Slide 49

Demonstration (Video)

Insight Centre for Data Analytics Slide 50

Take-away message

• Contemporary science demands new infrastructures to

scale scientific discovery in a complex knowledge

environment.

• In SIFEM we aim at experimenting with new

infrastructures based on Semantic Web standards to

support better: • Resource Coordination

• Reproducibility

• Efficiency & Automation

• Infrastructure/Data Attribution

• This institute can be a protagonist in this process.

Insight Centre for Data Analytics Slide 51

must!