Advanced Modularity Optimization Assignment Help

For any help regarding Computer Network Assignment Help
Visit :- https://www.computernetworkassignmenthelp.com/ ,
Email :- support@computernetworkassignmenthelp.com or
Call us at :- +1(315) 557-6473
computernetworkassignmenthelp.com

Introduction
Welcome to the presentation on Advanced Modularity
Optimization.
In this presentation, we will explore complex mathematical
expressions and their solutions.
The questions and solutions provided here are curated by
experts from ComputerNetworkAssignmentHelp.com to aid
students in their understanding of network modularity
optimization

Problems and Solutions:
(a) Show that if we divide the network into two parts by cutting any single edge, such that one part
has r vertices and the other has n r, the modularity, takes the value:
(b) Hence show that when n is even the optimal such division, in terms of modularity, is the
division that splits the network exactly down the middle.
Solution:
(a) From the equation in Newman, we can write our modularity, Q, as:
where m is the total number of edges, Bij = Aij k 2 ik m j , where ki is the de P gree of node i and
let (ci, cj ) = 1 2 (s si j + 1). Using the fact that j Bij = 0 (Eq 11.41 in Newman), we can write our
modularity as:

Since we have two classes (partitions), we can assign all si = 1 if node i is in and c j lass are 1, in
and the si = same 1 i class f nod and e i s is s in = class 1 i 2 f . th No e ti y a ce r t e i h n at di si ↵
s e j ren = 1 t cl if as n s o es d . es i i j Using this, we break up our summation as follows:

Notice that the sum over each Aij is just the sum of the degrees of the corresponding nodes, and
our degree products kikj take the following form:
Using the above gives us:

Plugging back into our equation for Q, setting m = (n 1) and simplifying, we get our desired answer
of:
(b) We can find the value of r which maximizes our modularity, Q, by setting its derivative to 0. We
have:
Setting dQ dr = 0, and solving for r, we get r = n 2 .
Problem 3.2
Using your favorite numerical software for finding eigenvectors of matrices, construct the
Laplacian and modularity matrix for this small network:

(a) Find the eigenvector of the Laplacian corresponding to the second smallest eigenvalue and
hence perform a spectral bisection of the network into two equally sized parts.
(b) Find the eigenvector of the modularity matrix corresponding to the largest eigenvalue and
hence divide the network into two communities.
You should find that the division of the network generated by the two methods is in this case, the
same.
Solution:
(a) For parts (a) and
(b) we use the following adjacency matrix:

We should get the following laplacian matrix:
The second smallest eigenvalue is: 0.438 The eigenvector
corresponding to the second smallest eigenvalue is:

This gives us the following partition vector:
(b) We should get the following modularity matrix:

The largest eigenvalue is: 1.732
The eigenvector corresponding to the largest eigenvalue is:
This gives us the following partition vector:
For both methods, you should see that our graph is partitioned straight down the middle.
Problem 3.3
Consider an Erd¨os-Renyi random graph G(n, p)

(a) Let A1 denotes the event that node 1 has at least l 2 Z+ neighbors. Do we observe a phase
transition for this event? If so, find the threshold function and explain your reasoning.
(b) Let B denote the event that a cycle with k edges (for a fixed k) emerges in the graph. Do we
observe a phase transition of this event? If so, find the threshold function and explain your
reasoning

Note that this implies P(Al|p(n) ! 0, since otherwise, the expected degree would be strictly positive.
Next assume that p(n) t(n) ! 1. It follows that p(n) > r n for any r 2 R+ and suciently large n. The
probability that Al does not occur can be bounded as follows:

Here the third line follows because if the graph was generated using t(n) instead of p(n), each link
would be present with a smaller probability and 7
hence the probability that node 1 has less than l neighbors (the event Al c ) would be larger. Since
the above is true for any r 2 R+, considering arbitrarily large r, it follows that:
(b) We observe phase transition for this part as well. Similar to part (a), consider the candidate
threshold function t(n) = r n for any r 2 R+. We will prove that for a fixed k, the event B satisfies:

In order to prove (i), assume that p(n) t(n) ! 0. Denote the number of distinct cycles on k nodes
by Ck. Note that over n nodes, n k (k1)! 2 di↵erent cycles (of k nodes) can be observed and each
cycle is realized with p(n)k probability. Therefore, the expectation of Ck can be found as:
hence E[Ck] ! 0 as p(n) t(n) ! 0. Note that this implies P(B|p(n)) ! 0, since otherwise, the
expectation of Ck would be strictly positive. 8

Next as ume that p(n) t(n) k ! 1. In a graph with n nodes, there can be at most N = n k ( t k e t h
1)! ese ⇠= cy c cl 0n es an dis d tinc for t i cyc 1 les ...N with de k fine no a des random (for a v
cons ariable tant I c0 su ). ch We enumera 2 i that Ii = 1 if the ith cycle is realized and 0
otherwise. Note that the probability that no cycle is realized satifies:
Calculating the expectations of the indicator variables we conclude that:
Next we upper bound var. In order to do so, we use the following identity

Using the properties of Bernoulli random variables, it follows that
On the other hand, cov(Ii, Ij ) = E[IiIj ] E[Ii][Ij ] = 0 if the cycles i, j do not have an edge in common,
since in this case the cycles are independent. Assume the cycles i and j have l common edges
(hence l + 1) common nodes. In this case:
Also note that there are at most n l+1 n cl such i, j pairs. This can 2( m k o l n 1) be obtained first
by identifying the com odes, and then choosing the n remaining nodes of both graphs and then
considering ordergins of nodes in cycle (which is captured by the constant cl). Combining the
above, and calculating the sum of the covariances by condition on the number of common edges
between i, j it follows that:

Since p(n) t(n) ! 0 this equation implies that

as claimed.
Problem 3.4
We can make a simple random graph model of a network with clustering or transitivity as follows.
We take n ve n rt ices and go through each distinct trio of three vertices, of which there are , and
with indepen- 3 dent probability p we connect the members of the trio together using three edges
to form a triangle, where p = c ( n 1 2 ) with c constant.
(a) Show that the mean degree of a vertex in this network is 2c.
(b) (b) Show that the degree distribution is
(c) Show that the clustering coecient is C = 1 2c+1 .
(d) Show that when there is a giant component in the network, its expected size S, as a fraction
of the network size, satisfies S = 1 ecS(2S) .
(e) What is the value of the clustering coecient when the giant component fills half of the
network?

Solution:
(a) For each vertex, there are n1 2 pairs of others with which it cold form a triangle, and each
triangle is present with probability c ( n 1 2 ) , for an average number of triangles c per vertex.
Each triangle contributes two edges to the degree, so the average degree is 2c.
(b) The probability pt of having t triangles follows the binomial distribution:
where the final equality is exact in the limit of large n. The degree is twice the number of triangles
and hence t = k/2 and:
so long as k is even. Odd values of k cannot occur, so pk = 0 for k odd.

(c) As shown above, there are on average c triangles around each vertex and hence nc is the
total number of triangles in the network times three (since each triangle appears around three
di↵erent vertices and gets counted three times).
T eh number of connected triples around a vertex of degree k = 2t is 2t 2 = t(2t 1) and there are
npt vertices with t triangles, with pt as above. So the total number of connected triples is:
(The sum is a standard one that can be found in tables, but it’s also reasonably straightforward
to do by hand if you know the right tricks) Now the clustering coefficient is

(d) Let u be the probability that a vertex is not in the giant component. If n a v 1 e rtex is not in
the giant component, then it must be that for each of the 2 distinct pairs of other vertices in
the network. Either (a) that pair does not form a triangle with our vertex (probability 1 p) or (b)
the pair does form a triangle (probability p) but neither member of the pair is itself in the giant
component (probability u2). Thus the analog of Eq. (12.12) for this model is
Putting p = c/n1 2 and taking the limit of large n this becomes u = ec(1u2) . Putting S = 1 u we
then find that S = 1 ecS(2S) .
(e) Rearranging for c in terms of S we have
and for S = 1/2 this gives c = 4 3 ln(2). Substituting into the expression for the clustering
coefficient above then gives

Problem 3.5
Consider the variation on the small-world model proposed by Newman and Watts: consider a
ring lattice with n nodes in which each node is connected to its neighbors k hops or less away.
For each edge, with one probability p, add a new edge to the ring lattice between two nodes
chosen uniformly at random.
(a) Find the degree distribution of this model.
(b) Show that when p = 0, the overall clustering coefficient of this graph is given by
(c) Show that when p > 0, the overall clustering coecient is given by

Solution:
(a) Degree of vertex = 2k + (number of shortcut edges attached to it). The number of shortcut edges
is nk. For each such edge: add shortcut with probability p. There are nkp shorts on average. There
are 2nkp ends of shortcuts on average. There are 2kp ends of shortcuts on average per vertex.
Similar to Erdos-Renyi, we will have a Poisson distribution in the limit of large n:
Therefore, the degree distribution is
(b) First, we give the labels 1,...,n to the nodes in a counter clockwise fashion starting from an
arbitrary node. when p = 0, two nodes with labels u and v have an edge if they are at most k hops
away, i.e., if |u v| k. We will compute the clustering coecient by using the following definition:

The following expression gives the total number of triangles that agenet 1 forms with agents 2 to
k + 1 (note that this is not the total number of triangles that agent 1 can form, since we are not
cunting (yet) the triangles with agents n k + 1 to n):
From symmetry we conclude that the total number of triangles is simply n k(k1) 2 (note that by
considering the triangles that 1 forms just with agents 2 to k + 1 and continuing in the same
fashion with agents 2 to n we avoided double-counting any triangles). Similarly, the number of
triples that agent 1 forms with agents 2 to 2k + 1 is given by:

where the second term in the summation comes from the fact that each triangle is counted just
once in the first term. Thus, the overall clustering coecient is simply given by the ratio:
(c) The overall clustering coecient is defined as the average of the clustering coecients of individual
nodes in the graph. The clustering coecient for agient i is defined as the ratio of all links between
the neighbors of i over the number of all potential links between the neighbors. If agent i as ni
neighbors, then the number of potential links between i’s neighbors is ni(ni1) 2 . As n ! 1 the above
definition is equivalent with the
following:
Next note that two neighbors of agent i that were connected at p = 0 will remain connected and
linked with i with probability (1 p)3 when p > 0. Thus, the expected number of links between the
neighbors of a node is equal to 2k(k1) 2 (1 resp p) onds 3 + to O a (1/n new ), whe triangle re the b
s eing econd formed term by is t ne wo gligible edges th as n ! 1 and cor at were rewired and one
that was not rewired, etc

On the other hand, the expected number of potential links between the neighbors of a node
remains the same as in the case of p = 0 and
Problem 3.6 Consider a society of n individuals. A randomly chosen node is infected with a
contagious infection. Assume that the network of interactions in the society is represented by a
configuration model with degree distribution pk. Assume that any individual is immune
independently with probability ⇡. We would like to investigate whether the infection can spread
to a nontrivial fraction of the society.
(a) Find a threshold for the immunity probability (in terms of the moments of the degree
distribution) below which the infection spreads to a large portion of the population.
(b) What is this threshold for a k-regular random graph, i.e., a configuration model network in
which all nodes have the same degree?
(c) What is this threshold for a power-law graph with exponents less than 3, i.e., pk ⇠ k↵ with ↵
< 3? The Internet graph (representing connections between routers) has a power-law
distribution with exponent ⇠ 2.1 2.7. What does this result imply for the Internet graph?
(d) (d) Find the size of the infected population (you can assume that the infection spreads to a
large portion of the population).

Solution:
(a) Degree distribution of a neighboring node:
Therefore, the expected number of infected children, , is:

Now emply branching process approximation:
(i) if < 1, then the disease dies out after a finite number of stages
(ii) if > 1, then with positive probability, the disease persists by infecting a large portion of the
population
Therefore

(b) For k ¯-regular random graph
(c) in Pow te e rne r l t a ( w ↵ : pk 2. ⇠ 1 k↵ 2 , .7) ↵ is < 3. robust. < k2 If >= you 1, s remo o t v h e
res 98% hol o d i f s the 1. no The des, refo it’s re, the ⇠ still connected. (d) We want to compute
the size of the giant component. Consider a node and the event that this node is in the giant
component, or equivalently, that the branching process does not die out. Let q˜ = probability that
the branching process does not die out. Starting from neighboring node:

where the first term is the probability that the neighbor is immune, and the second term
(summation) is the probability that the neighbor is not immune (however, none of its other
neighbors sustains the process).
Where the right hand side is the probability that none of the neighbors manages to sustain the
branching process.

Advanced Modularity Optimization Assignment Help

More Related Content

Similar to Advanced Modularity Optimization Assignment Help (20)

More from Computer Network Assignment Help (20)

Recently uploaded (20)

Advanced Modularity Optimization Assignment Help