Correlations
in real
MULTIPLEXES
HIDDEN GEOMETRIC
Kaj Kolja KLEINEBERG
Universitat de Barcelona
@KoljaKleineberg
kajkoljakleineberg@gmail.com
You have no message
routing
the problem of
maprelies on
underlying
metric space
real systems interact:
multiplexes
between maps of layers?Relation
Hidden metric spaces underlying real complex networks
provide a fundamental explanation of their observed topologies
Nature Physics 5, 74–80 (2008)
Hidden metric spaces underlying real complex networks
provide a fundamental explanation of their observed topologies
How can we infer the coordinates of nodes embedded
in hidden metric spaces?
Hyperbolic geometry emerges from Newtonian model
and similarity × popularity optimization in growing network
S1
p(κ) ∝ κ−γ
r = 1
1+
[
d(θ,θ′)
µκκ′
]1/T
PRL 100, 078701
Hyperbolic geometry emerges from Newtonian model
and similarity × popularity optimization in growing network
S1 H2
p(κ) ∝ κ−γ ri = R − 2 ln κi
κmin
r = 1
1+
[
d(θ,θ′)
µκκ′
]1/T p(xij) = 1
1+e
xij−R
2T
PRL 100, 078701 PRE 82, 036106
Hyperbolic geometry emerges from Newtonian model
and similarity × popularity optimization in growing network
S1 H2 growing
1
2
3
p(κ) ∝ κ−γ ri = R − 2 ln κi
κmin t = 1, 2, 3 . . .
r = 1
1+
[
d(θ,θ′)
µκκ′
]1/T p(xij) = 1
1+e
xij−R
2T
mins [s × ∆θst]
PRL 100, 078701 PRE 82, 036106 Nature 489, 537–540
Constituent network layers of real multiplex systems
are embedded into separate hyperbolic spaces
Invert model → embed real networks: PRE 92, 022807 (2015)
Internet Air and train Drosophila
C. Elegans Human brain arXiv
Rattus Physicians SacchPomb
Constituent network layers of real multiplex systems
are embedded into separate hyperbolic spaces
Invert model → embed real networks: PRE 92, 022807 (2015)
Internet Air and train Drosophila
C. Elegans Human brain arXiv
Rattus Physicians SacchPomb
Constituent network layers of real multiplex systems
are embedded into separate hyperbolic spaces
Internet IPv4 network Internet IPv6 network
Constituent network layers of real multiplex systems
are embedded into separate hyperbolic spaces
Internet IPv4 network Internet IPv6 network
Are coordinates of same nodes in different layers
correlated?
Radial and angular coordinates are correlated
between different layers in many real multiplexes
0
π
2 π
θ1
0
π
2 π
θ2
50
100
150
200
Radial and angular coordinates are correlated
between different layers in many real multiplexes
0
π
2 π
θ1
0
π
2 π
θ2
50
100
150
200
Multiplexes are not random superpositions of layers
but instead are dictated by geometric correlations.
Distance between pairs of nodes in one layer is
an indicator of the connection probability in another layer
Hyperbolic distance in IPv4
Connectionprob.inIPv6
P(2|1)
0 5 10 15 20 25 30 35 40
10-4
10-3
10-2
10-1
100
Distance between pairs of nodes in one layer is
an indicator of the connection probability in another layer
Hyperbolic distance in IPv4
Connectionprob.inIPv6
P(2|1)
0 5 10 15 20 25 30 35 40
10-4
10-3
10-2
10-1
100
Pran(2|1)
Distance between pairs of nodes in one layer is
an indicator of the connection probability in another layer
Hyperbolic distance in IPv4
Connectionprob.inIPv6
P(2|1)
0 5 10 15 20 25 30 35 40
10-4
10-3
10-2
10-1
100
Pran(2|1)
Geometric correlations enable precise trans-layer
link prediction.
navigation?
How do geometric correlations
affect
Greedy routing in single network using hyperbolic space
allows efficient navigation relying only on local knowledge
Greedy routing in single network using hyperbolic space
allows efficient navigation relying only on local knowledge
Greedy routing in single network using hyperbolic space
allows efficient navigation relying only on local knowledge
Greedy routing in single network using hyperbolic space
allows efficient navigation relying only on local knowledge
Greedy routing in single network using hyperbolic space
allows efficient navigation relying only on local knowledge
Greedy routing in single network using hyperbolic space
allows efficient navigation relying only on local knowledge
Mutual greedy routing allows efficient navigation
using several network layers and metric spaces
Mutual greedy routing allows efficient navigation
using several network layers and metric spaces
Mutual greedy routing allows efficient navigation
using several network layers and metric spaces
Mutual greedy routing allows efficient navigation
using several network layers and metric spaces
Mutual greedy routing allows efficient navigation
using several network layers and metric spaces
Mutual greedy routing allows efficient navigation
using several network layers and metric spaces
Mutual greedy routing allows efficient navigation
using several network layers and metric spaces
Mutual greedy routing allows efficient navigation
using several network layers and metric spaces
How do geometric
correlations affect
performance?
Mutual greedy routing allows efficient navigation
using several network layers and metric spaces
How do geometric
correlations affect
performance?
Do more layers
improve
performance?
Mutual greedy routing allows efficient navigation
using several network layers and metric spaces
How do geometric
correlations affect
performance?
Do more layers
improve
performance?
How close
to the optimum is
the Internet?
Mutual greedy routing allows efficient navigation
using several network layers and metric spaces
How do geometric
correlations affect
performance?
Do more layers
improve
performance?
How close
to the optimum is
the Internet?
Develop a model with tunable geometric
correlations preserving constituent layer topologies.
Geometric multiplex model allows to tune
correlations independently from layer topologies
Constituent layer
topologies by
hyperbolic model
Geometric multiplex model allows to tune
correlations independently from layer topologies
Constituent layer
topologies by
hyperbolic model
Radial correlations
controlled by
ν ∈ [0, 1]
Geometric multiplex model allows to tune
correlations independently from layer topologies
Constituent layer
topologies by
hyperbolic model
Radial correlations
controlled by
ν ∈ [0, 1]
Angular corr.
controlled by
g ∈ [0, 1]
Geometric correlations determine the improvement of
mutual greedy routing by increasing the number of layers
0.0 0.2 0.4 0.6 0.8 1.0
0.2
0.4
0.6
0.8
1.0
Hyperbolic routing
0.980
0.985
0.990
0.995
P
1 2 3 4
0
2
4
6
Layers
Mitigationfactor
opt. correlated
uncorrelated
Angular correlations
Radialcorrelations
Reduction of failure rate
Geometric correlations determine the improvement of
mutual greedy routing by increasing the number of layers
0.0 0.2 0.4 0.6 0.8 1.0
0.2
0.4
0.6
0.8
1.0
Hyperbolic routing
0.980
0.985
0.990
0.995
P
1 2 3 4
0
2
4
6
Layers
Mitigationfactor
opt. correlated
uncorrelated
Angular correlations
Radialcorrelations
Reduction of failure rate
Geometric correlations determine the improvement of
mutual greedy routing by increasing the number of layers
0.0 0.2 0.4 0.6 0.8 1.0
0.2
0.4
0.6
0.8
1.0
Hyperbolic routing
0.980
0.985
0.990
0.995
P
1 2 3 4
0
2
4
6
Layers
Mitigationfactor
opt. correlated
uncorrelated
Angular correlations
Radialcorrelations
Reduction of failure rate
Routing is perfected by using many networks
simultaneously if correlations are present.
Internet?
Do correlations help to navigate
the real
Internet multiplex model allows to study the performance
of mutual greedy routing for arbitrary correlations
Real Model
• Node only in IPv4 • Node in IPv4 and IPv6
Existing correlations in the Internet multiplex increase
performance of mutual greedy routing significantly
0.0 0.2 0.4 0.6 0.8 1.0
0.0
0.2
0.4
0.6
0.8
1.0
g
ν Angular
0.0 0.2 0.4 0.6 0.8 1.0
0.0
0.2
0.4
0.6
0.8
1.0
gν
Hyperbolic
0.63 0.67 0.71 0.75
Success rate
0.85 0.91
Success rate
uncorr. emp. opt. uncorr. emp. opt.
0.88
Summary: Geometry of multiplex networks can make
interacting decentralized systems perfectly navigable
0.0 0.2 0.4 0.6 0.8 1.0
0.0
0.2
0.4
0.6
0.8
1.0
g
ν
0.80
0.82
0.84
0.86
0.88
0.90
Framework
0
π
2 π
θ1
0
π
2 π
θ2
0.2
0.4
ResultBasis
Navigation
Mutual greedy routing
(local knowledge)
Geometric
correlations
Layers embedded in
hyperbolic space
Multidimensional
community structure
Precise trans-layer
link prediction
Correlations increase
routing performance
Many correlated layers
perfect routing
Distances between
nodes are correlated
Hyperbolic distance in IPv4
Connectionprob.inIPv6
P(2|1)
0 5 10 15 20 25 30 35 40
10-4
10-3
10-2
10-1
100
Pran(2|1)
Nature Physics, doi: 10.1038/NPHYS3782
Node coordinates
are correlated
Multiplexes not random
combinations of layers
Multiplex
organization
Summary: Geometry of multiplex networks can make
interacting decentralized systems perfectly navigable
0.0 0.2 0.4 0.6 0.8 1.0
0.0
0.2
0.4
0.6
0.8
1.0
g
ν
0.80
0.82
0.84
0.86
0.88
0.90
Framework
0
π
2 π
θ1
0
π
2 π
θ2
0.2
0.4
ResultBasis
Navigation
Mutual greedy routing
(local knowledge)
Geometric
correlations
Layers embedded in
hyperbolic space
Multidimensional
community structure
Precise trans-layer
link prediction
Correlations increase
routing performance
Many correlated layers
perfect routing
Distances between
nodes are correlated
Hyperbolic distance in IPv4
Connectionprob.inIPv6
P(2|1)
0 5 10 15 20 25 30 35 40
10-4
10-3
10-2
10-1
100
Pran(2|1)
Nature Physics, doi: 10.1038/NPHYS3782
Node coordinates
are correlated
Multiplexes not random
combinations of layers
Multiplex
organization
Summary: Geometry of multiplex networks can make
interacting decentralized systems perfectly navigable
0.0 0.2 0.4 0.6 0.8 1.0
0.0
0.2
0.4
0.6
0.8
1.0
g
ν
0.80
0.82
0.84
0.86
0.88
0.90
Framework
0
π
2 π
θ1
0
π
2 π
θ2
0.2
0.4
ResultBasis
Navigation
Mutual greedy routing
(local knowledge)
Geometric
correlations
Layers embedded in
hyperbolic space
Multidimensional
community structure
Precise trans-layer
link prediction
Correlations increase
routing performance
Many correlated layers
perfect routing
Distances between
nodes are correlated
Hyperbolic distance in IPv4
Connectionprob.inIPv6
P(2|1)
0 5 10 15 20 25 30 35 40
10-4
10-3
10-2
10-1
100
Pran(2|1)
Nature Physics, doi: 10.1038/NPHYS3782
Node coordinates
are correlated
Multiplexes not random
combinations of layers
Multiplex
organization
Geometric correlations in real multiplexes:
Reference and contact information
Thanks to: M. Boguñá, M. A. Serrano, F. Papadopoulos
Reference:
»Hidden geometric correlations in real multiplex networks«
Nature Physics, doi: 10.1038/NPHYS3782
K.-K. Kleineberg, M. Boguñá, M. A. Serrano, F. Papadopoulos
Kaj Kolja Kleineberg:
• kajkoljakleineberg@gmail.com
• @KoljaKleineberg
• koljakleineberg.wordpress.com
Geometric correlations in real multiplexes:
Reference and contact information
Thanks to: M. Boguñá, M. A. Serrano, F. Papadopoulos
Reference:
»Hidden geometric correlations in real multiplex networks«
Nature Physics, doi: 10.1038/NPHYS3782
K.-K. Kleineberg, M. Boguñá, M. A. Serrano, F. Papadopoulos
Kaj Kolja Kleineberg:
• kajkoljakleineberg@gmail.com
• @KoljaKleineberg ← Slides
• koljakleineberg.wordpress.com
IMAGE CREDITS
Compass: Martin Fisch
Message in bottle: Susanne
Nilsson
Old globe: jayneandd
Compass: Creative Stall
Compass (navigate): Creative
Stall
Internet router: Thomas Uebe
Train: Naomi Atkinson
fly (drosophila): Daan Kauwenberg
worm (celegans): anbileru adaleru
bain network: parkjisun
coauthor: Matt Wasser
community: Edward Boatman
Link: Rafaël Massé
Icons: thenounproject
Pictures: flickr

Hidden geometric correlations in real multiplex networks

  • 1.
    Correlations in real MULTIPLEXES HIDDEN GEOMETRIC KajKolja KLEINEBERG Universitat de Barcelona @KoljaKleineberg kajkoljakleineberg@gmail.com
  • 2.
    You have nomessage
  • 3.
  • 4.
  • 5.
  • 6.
    between maps oflayers?Relation
  • 7.
    Hidden metric spacesunderlying real complex networks provide a fundamental explanation of their observed topologies Nature Physics 5, 74–80 (2008)
  • 8.
    Hidden metric spacesunderlying real complex networks provide a fundamental explanation of their observed topologies How can we infer the coordinates of nodes embedded in hidden metric spaces?
  • 9.
    Hyperbolic geometry emergesfrom Newtonian model and similarity × popularity optimization in growing network S1 p(κ) ∝ κ−γ r = 1 1+ [ d(θ,θ′) µκκ′ ]1/T PRL 100, 078701
  • 10.
    Hyperbolic geometry emergesfrom Newtonian model and similarity × popularity optimization in growing network S1 H2 p(κ) ∝ κ−γ ri = R − 2 ln κi κmin r = 1 1+ [ d(θ,θ′) µκκ′ ]1/T p(xij) = 1 1+e xij−R 2T PRL 100, 078701 PRE 82, 036106
  • 11.
    Hyperbolic geometry emergesfrom Newtonian model and similarity × popularity optimization in growing network S1 H2 growing 1 2 3 p(κ) ∝ κ−γ ri = R − 2 ln κi κmin t = 1, 2, 3 . . . r = 1 1+ [ d(θ,θ′) µκκ′ ]1/T p(xij) = 1 1+e xij−R 2T mins [s × ∆θst] PRL 100, 078701 PRE 82, 036106 Nature 489, 537–540
  • 12.
    Constituent network layersof real multiplex systems are embedded into separate hyperbolic spaces Invert model → embed real networks: PRE 92, 022807 (2015) Internet Air and train Drosophila C. Elegans Human brain arXiv Rattus Physicians SacchPomb
  • 13.
    Constituent network layersof real multiplex systems are embedded into separate hyperbolic spaces Invert model → embed real networks: PRE 92, 022807 (2015) Internet Air and train Drosophila C. Elegans Human brain arXiv Rattus Physicians SacchPomb
  • 14.
    Constituent network layersof real multiplex systems are embedded into separate hyperbolic spaces Internet IPv4 network Internet IPv6 network
  • 15.
    Constituent network layersof real multiplex systems are embedded into separate hyperbolic spaces Internet IPv4 network Internet IPv6 network Are coordinates of same nodes in different layers correlated?
  • 16.
    Radial and angularcoordinates are correlated between different layers in many real multiplexes 0 π 2 π θ1 0 π 2 π θ2 50 100 150 200
  • 17.
    Radial and angularcoordinates are correlated between different layers in many real multiplexes 0 π 2 π θ1 0 π 2 π θ2 50 100 150 200 Multiplexes are not random superpositions of layers but instead are dictated by geometric correlations.
  • 18.
    Distance between pairsof nodes in one layer is an indicator of the connection probability in another layer Hyperbolic distance in IPv4 Connectionprob.inIPv6 P(2|1) 0 5 10 15 20 25 30 35 40 10-4 10-3 10-2 10-1 100
  • 19.
    Distance between pairsof nodes in one layer is an indicator of the connection probability in another layer Hyperbolic distance in IPv4 Connectionprob.inIPv6 P(2|1) 0 5 10 15 20 25 30 35 40 10-4 10-3 10-2 10-1 100 Pran(2|1)
  • 20.
    Distance between pairsof nodes in one layer is an indicator of the connection probability in another layer Hyperbolic distance in IPv4 Connectionprob.inIPv6 P(2|1) 0 5 10 15 20 25 30 35 40 10-4 10-3 10-2 10-1 100 Pran(2|1) Geometric correlations enable precise trans-layer link prediction.
  • 21.
    navigation? How do geometriccorrelations affect
  • 22.
    Greedy routing insingle network using hyperbolic space allows efficient navigation relying only on local knowledge
  • 23.
    Greedy routing insingle network using hyperbolic space allows efficient navigation relying only on local knowledge
  • 24.
    Greedy routing insingle network using hyperbolic space allows efficient navigation relying only on local knowledge
  • 25.
    Greedy routing insingle network using hyperbolic space allows efficient navigation relying only on local knowledge
  • 26.
    Greedy routing insingle network using hyperbolic space allows efficient navigation relying only on local knowledge
  • 27.
    Greedy routing insingle network using hyperbolic space allows efficient navigation relying only on local knowledge
  • 28.
    Mutual greedy routingallows efficient navigation using several network layers and metric spaces
  • 29.
    Mutual greedy routingallows efficient navigation using several network layers and metric spaces
  • 30.
    Mutual greedy routingallows efficient navigation using several network layers and metric spaces
  • 31.
    Mutual greedy routingallows efficient navigation using several network layers and metric spaces
  • 32.
    Mutual greedy routingallows efficient navigation using several network layers and metric spaces
  • 33.
    Mutual greedy routingallows efficient navigation using several network layers and metric spaces
  • 34.
    Mutual greedy routingallows efficient navigation using several network layers and metric spaces
  • 35.
    Mutual greedy routingallows efficient navigation using several network layers and metric spaces How do geometric correlations affect performance?
  • 36.
    Mutual greedy routingallows efficient navigation using several network layers and metric spaces How do geometric correlations affect performance? Do more layers improve performance?
  • 37.
    Mutual greedy routingallows efficient navigation using several network layers and metric spaces How do geometric correlations affect performance? Do more layers improve performance? How close to the optimum is the Internet?
  • 38.
    Mutual greedy routingallows efficient navigation using several network layers and metric spaces How do geometric correlations affect performance? Do more layers improve performance? How close to the optimum is the Internet? Develop a model with tunable geometric correlations preserving constituent layer topologies.
  • 39.
    Geometric multiplex modelallows to tune correlations independently from layer topologies Constituent layer topologies by hyperbolic model
  • 40.
    Geometric multiplex modelallows to tune correlations independently from layer topologies Constituent layer topologies by hyperbolic model Radial correlations controlled by ν ∈ [0, 1]
  • 41.
    Geometric multiplex modelallows to tune correlations independently from layer topologies Constituent layer topologies by hyperbolic model Radial correlations controlled by ν ∈ [0, 1] Angular corr. controlled by g ∈ [0, 1]
  • 42.
    Geometric correlations determinethe improvement of mutual greedy routing by increasing the number of layers 0.0 0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0 Hyperbolic routing 0.980 0.985 0.990 0.995 P 1 2 3 4 0 2 4 6 Layers Mitigationfactor opt. correlated uncorrelated Angular correlations Radialcorrelations Reduction of failure rate
  • 43.
    Geometric correlations determinethe improvement of mutual greedy routing by increasing the number of layers 0.0 0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0 Hyperbolic routing 0.980 0.985 0.990 0.995 P 1 2 3 4 0 2 4 6 Layers Mitigationfactor opt. correlated uncorrelated Angular correlations Radialcorrelations Reduction of failure rate
  • 44.
    Geometric correlations determinethe improvement of mutual greedy routing by increasing the number of layers 0.0 0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0 Hyperbolic routing 0.980 0.985 0.990 0.995 P 1 2 3 4 0 2 4 6 Layers Mitigationfactor opt. correlated uncorrelated Angular correlations Radialcorrelations Reduction of failure rate Routing is perfected by using many networks simultaneously if correlations are present.
  • 45.
    Internet? Do correlations helpto navigate the real
  • 46.
    Internet multiplex modelallows to study the performance of mutual greedy routing for arbitrary correlations Real Model • Node only in IPv4 • Node in IPv4 and IPv6
  • 47.
    Existing correlations inthe Internet multiplex increase performance of mutual greedy routing significantly 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 g ν Angular 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 gν Hyperbolic 0.63 0.67 0.71 0.75 Success rate 0.85 0.91 Success rate uncorr. emp. opt. uncorr. emp. opt. 0.88
  • 48.
    Summary: Geometry ofmultiplex networks can make interacting decentralized systems perfectly navigable 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 g ν 0.80 0.82 0.84 0.86 0.88 0.90 Framework 0 π 2 π θ1 0 π 2 π θ2 0.2 0.4 ResultBasis Navigation Mutual greedy routing (local knowledge) Geometric correlations Layers embedded in hyperbolic space Multidimensional community structure Precise trans-layer link prediction Correlations increase routing performance Many correlated layers perfect routing Distances between nodes are correlated Hyperbolic distance in IPv4 Connectionprob.inIPv6 P(2|1) 0 5 10 15 20 25 30 35 40 10-4 10-3 10-2 10-1 100 Pran(2|1) Nature Physics, doi: 10.1038/NPHYS3782 Node coordinates are correlated Multiplexes not random combinations of layers Multiplex organization
  • 49.
    Summary: Geometry ofmultiplex networks can make interacting decentralized systems perfectly navigable 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 g ν 0.80 0.82 0.84 0.86 0.88 0.90 Framework 0 π 2 π θ1 0 π 2 π θ2 0.2 0.4 ResultBasis Navigation Mutual greedy routing (local knowledge) Geometric correlations Layers embedded in hyperbolic space Multidimensional community structure Precise trans-layer link prediction Correlations increase routing performance Many correlated layers perfect routing Distances between nodes are correlated Hyperbolic distance in IPv4 Connectionprob.inIPv6 P(2|1) 0 5 10 15 20 25 30 35 40 10-4 10-3 10-2 10-1 100 Pran(2|1) Nature Physics, doi: 10.1038/NPHYS3782 Node coordinates are correlated Multiplexes not random combinations of layers Multiplex organization
  • 50.
    Summary: Geometry ofmultiplex networks can make interacting decentralized systems perfectly navigable 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 g ν 0.80 0.82 0.84 0.86 0.88 0.90 Framework 0 π 2 π θ1 0 π 2 π θ2 0.2 0.4 ResultBasis Navigation Mutual greedy routing (local knowledge) Geometric correlations Layers embedded in hyperbolic space Multidimensional community structure Precise trans-layer link prediction Correlations increase routing performance Many correlated layers perfect routing Distances between nodes are correlated Hyperbolic distance in IPv4 Connectionprob.inIPv6 P(2|1) 0 5 10 15 20 25 30 35 40 10-4 10-3 10-2 10-1 100 Pran(2|1) Nature Physics, doi: 10.1038/NPHYS3782 Node coordinates are correlated Multiplexes not random combinations of layers Multiplex organization
  • 51.
    Geometric correlations inreal multiplexes: Reference and contact information Thanks to: M. Boguñá, M. A. Serrano, F. Papadopoulos Reference: »Hidden geometric correlations in real multiplex networks« Nature Physics, doi: 10.1038/NPHYS3782 K.-K. Kleineberg, M. Boguñá, M. A. Serrano, F. Papadopoulos Kaj Kolja Kleineberg: • kajkoljakleineberg@gmail.com • @KoljaKleineberg • koljakleineberg.wordpress.com
  • 52.
    Geometric correlations inreal multiplexes: Reference and contact information Thanks to: M. Boguñá, M. A. Serrano, F. Papadopoulos Reference: »Hidden geometric correlations in real multiplex networks« Nature Physics, doi: 10.1038/NPHYS3782 K.-K. Kleineberg, M. Boguñá, M. A. Serrano, F. Papadopoulos Kaj Kolja Kleineberg: • kajkoljakleineberg@gmail.com • @KoljaKleineberg ← Slides • koljakleineberg.wordpress.com
  • 53.
    IMAGE CREDITS Compass: MartinFisch Message in bottle: Susanne Nilsson Old globe: jayneandd Compass: Creative Stall Compass (navigate): Creative Stall Internet router: Thomas Uebe Train: Naomi Atkinson fly (drosophila): Daan Kauwenberg worm (celegans): anbileru adaleru bain network: parkjisun coauthor: Matt Wasser community: Edward Boatman Link: Rafaël Massé Icons: thenounproject Pictures: flickr