Upload
zvi-lotker
View
43
Download
0
Embed Size (px)
Citation preview
Recovering the long range links in augmented graphs
Large interaction networks Spontaneous & local construction. Very large number of nodes. Small diameter. High clustering. Power-law degree distr
Milgram Experiment Source person s (e.g., in Wichita) Target person t (e.g., in Cambridge)
Name, professional occupation, city of living, etc.
Letter transmitted via a chain of individuals related on a personal basis
Result: “six degrees of separation”
The small world effect
M. SmithFarmerWyoming
M. SMOOTHPhysicianBosto
Very short paths exists
Milgram 1967
People are able to discover them locally. (navigability)
Navigability Jon Kleinberg (2000)
Why should there exist short chains of acquaintances linking together arbitrary pairs of strangers?
Why should arbitrary pairs of strangers be able to find short chains of acquaintances that link them together?
In other words: how to navigate in a small worlds?
Augmented graphs H=(G,D) Individuals as nodes of a graph G
Edges of G model relations between individuals deducible from their societal positions
A number k of “long links” are added to Gat random, according to some probability distribution D: Du(v)=prob(u→v) Long links model relations between individuals
that cannot be deduced from their societal positions
Greedy Routing in augmented graphs
Source s ∈ V(G) Target t ∈ V(G) Current node x selects among its degG(x)+k
neighbors the closest to t in G, that is according to the distance function distG().
Greedy routing in augmented graphs aims at modeling the routing process performed by social entities in Milgram’s experiment.
Augmented meshesKleinberg (2000)
d-dimensional n-node meshes augmented with d-harmonic links
uv
Du(v)=prob(u→v) ≈ 1/(log(n)*dist(u,v)d)
Kleinberg’s theorems Greedy routing performs in O(log2n / k)
expected #steps in d-dimensional meshes augmented with k links per node, chosen according to the d-harmonic distribution. Note: k = log n ⇒ O(log n) expect. #steps
Greedy routing in d-dimensional meshes augmented with a h-harmonic distribution, h≠d, performs in Ω(nε) expected #steps.
Navigable graphs Let f : N → R be a function
Definition: An n-node graph G is f-navigable if there exist an augmentation D for G such that greedy routing in (G,D) performs in at most f(n) expected #steps.
O(polylog(n))-Navigable graphs
Bounded growth graphs Definition: |B(x,2r)| ≤ k |B(x,r)| Duchon, Hanusse, Lebhar, Schabanel (2005)
Bounded doubling dimension Definition: every B(x,2r) can be covered by at
most 2d balls of radius r, B(xi,r) Slivkins (2005)
Graphs of bounded treewidth Fraigniaud (2005)
Doubling dimensionBounded doubling dimension
Definition: every B(x,2r) can be covered by at most 2d balls of radius r, B(xi,r)
Our resultTheorem (Fraigniaud, Lebhar, Lotker)
Let d such that limn→+∞ loglog n / d(n) = 0
Let F be the family of n-node graphs with doubling dimension at most d(n).
F is not O(polylog(n))-navigable.
Proof of non-navigability The family F contains the graph Hd:
x = x1 x2 ... xdis connected to all nodes
y = y1 y2 ... ydsuch that yi = xi + ai where
ai ∈ -1,0,+1
Hd has doubling dimension d
Intuitive approach Large doubling dimension d implies that
every nodes x ∈ Hd has choices over exponentially many directions
The underlying metric of Hd is L∞:
Case of symmetric distribution
Source s
Target t
Disadvantageddirection
At every step, probability ≤ 1/2d to go in the right direction
Counting argument 2d directions Lines are split in intervals of length L n/L × 2d intervals in total Every node belongs to many intervals,
but can be the certificate of at most one interval
If L<2d there is one interval J0 without certificate
Conclusion Remark: The proof still holds if the long
links are not set independently. Theorem (Peleg)
Every n-node graph is O(n1/2)-navigable, by using a uniform setting of the long links
Open problem: look for the smallest function f such that every n-node graph is f(n)-navigable
Navigability in Kleinbergmodel
A routing algorithm is claimed decentralized if:
1. It knows all links of the mesh,2. It discovers locally the extra random
links. Greedy routing computes paths of
expected length O(log2n) between any pair in this model.
Augmented graph models Augmented graph= (H,φ)
H= base graph, globally known φ= augmented links distribution
φu(v)= probability that v is the long range contact of u.
Recovering the longrange links
1. Validation of the model: Are there long range links? What kind of real links are they?
2. Efficient routing with light encoding: The set of grid coordinates is enough to
route. Distances easy to compute with small
labels.
Goal Design an algorithm
with input G, small world graph that outputs H and a set E of long links such that if G is H’ augmented with E’,
(H,E) is a good approximation of (H’,E’). Good approximation: H close to H’ and
greedy routing efficient knowing only H.
Too big H may destroythe routing
Remaining shortcuts can be problematic!
Greedy diameter Ω(n1/√log n)
Hypothesis on the input G= small world graph We assume that G comes from a
density based augmentation. G= H + E, H of bounded growth and E
produced by φ, density based. We assume that the base metric H has
a high clustering.
Extracting the long range linksIntuition
The base metric has high clustering. Extraction algorithm: tests if links are
highly clustered.
Highly clustered:chances to be close.
Few triangles:chances to be long range.
Extracting the long range links Input: a bounded growth graph G,
clustered augmented by a density based distribution of shortcuts (unknown set E).
Extraction algorithm: for each edge, test # of triangles if it does not correspond to the clustering,
label it as shortcut.
Result Theorem:
if the clustering is high enough (≥logn/loglogn), the algorithm detects all shortcuts of large stretch (≥polylog(n)) except polylog(n) of them w.h.p..
Greedy routing with the output map computes routes of length at most poylylog(n) longer than it should with the original map.
Clustering hypothesis The clustering hypothesis is essential to
recover the long links with a local maximum likelihood algorithm. Theorem: On the ring of n nodes
harmonically augmented, for any 0<ε<1/5, any local maximum
likelihood algorithm misses Ω(n5ε/logn) links of stretch Ω(n1/(5-ε)).