Attack, Defense,

and Network
Robustness of
Internet of Things

Security and Privacy in Internet of Things

 • Introduction
• Centrality Attacks

Outline • Network Resilience

• Topological Defense Scheme
Introduction

• The IoT features ubiquitous communication among all kinds of electronic devices.
• It raises security concerns due to seamless penetration and automated integration
among all sorts of applications.
• For example, an adversary may leverage the interconnected devices for malware
propagation.
• Therefore, efficient and effective defense mechanisms are important to ensure the
reliability of the IoT
• In particular, the U.S. Department of Energy (DOE) has identified attack
resistance to be one of the seven major properties required for the operation of
the smart grid, which is an emerging field of the IoT.
Complex connections of the IoT

• By representing the intricate connections of the IoT as a graph

• The network vulnerability of the IoT to various attack schemes can be investigated.
• Three defense schemes are investigated to counter fatal attacks:
• The intrinsic topological defense scheme
• The fusion-based defense scheme
• The sequential defense scheme.
• Network robustness evaluated by using game equilibrium.
• Formulating the interplay between an adversary and a defender as a two-player
zero-sum game.
• They aim to maximize their own payoffs in terms of network connectivity
Network Resilience

• The fraction of the largest connected graph used as a measure of

network resilience to node or edge removals in the IoT.
• Node or edge removals can be viewed as temporal device or
connection failures or targeted attacks in the IoT settings.
• For example, node or edge removals in a graph can be caused by
denial of service (DoS) or jamming attacks, or by natural
occurrences.
Graph Definition

• A basic anatomy of a graph

that is generated by a
bunch of nodes that are
connected by edges.
Graph Mathematical
Representation
• Graphs can build up to become a complex
structure
• For example, Facebook social
network.
• It will be hard to study it just by observing it
visually, so for that, we need to build
mathematical tools that will help us
understand or analyze our graph structure.
• N denotes the set of nodes in our graph
and E is the set of edges we also define the
norm of our graph as the number of nodes
 Adjacency matrix
• We need some sort of tools that encapsulate the information in our graphs and
also easy to do mathematical analysis on it.
• An undirected and unweighted graph G = (N,E) can be used to characterize the
network connectivity structure of the IoT.
• N is the set of nodes (devices) with size n.
• E is the set of edges (connections) with size m.
• The graph can be represented by an n-by-n binary symmetric adjacency matrix
A.
• Ai j = 1 if there is an edge between nodes i and j.
• Ai j = 0, if there no edge between nodes i and j.
 Adjacency matrix

• The adjacency matrix is

symmetric for the undirected
graph (a type of graph
structures)
 Centrality Measure

• A node centrality measure is a quantity that measures the

level of importance of a node in a network.
• The utility of centrality measures is that they can break the
combinatorial bottleneck of searching through all the possible
permutations and combinations of nodes that might reduce
largest component size.
• An attack that removes nodes according to a measure of
centrality will be referred to as a centrality attack.
 Centrality Measure…

• The centrality of a node is a measure of the node’s importance

to the network.
• Centrality measures can be classified into two categories:
• global measure
• Local measure
• Global centrality measures require complete topological
information for their computation
• Local centrality measures require only partial topological
information from neighboring nodes.
 Centrality Measure…

• Acquiring shortest path information between every node pair is

a global method required for the betweenness centrality
measure.
• Acquiring degree information of every node is a local method.
• Centrality measures are scalar values given to each node in the
graph to quantify its importance based on an assumption.
 Centrality Measure…

Centrality Measures:
• Degree (di)
• Betweenness
• Closeness
• Eigenvector centrality (eigen centrality)
• Ego centrality
• Local Fiedler Vector Centrality (LFVC)
Node Degree

• Node degree is one of the

basic centrality measures.
• It's equal to the number of
node neighbors.
• Thus, the more neighbors a
node have the more it's
central and highly connected,
thus have an influence on the
graph.
 Node Degree
• Node degree is local, not global.
• Node degree gives us an idea about
each node connectivity, its a local
measure and doesn't show us the
global picture.
• The two nodes have the same
number of neighbors which means
they have the same node degree but
look at their neighbors.
• Node 1 is connected to nodes with a
high degree and so if we are
choosing based on connectivity we
will choose it for sure thus node
degree as a measure doesn't take
neighbors connectivity in
consideration that why we call it a
local measure.
 Eigenvector centrality
• Eigenvector centrality measures a node’s importance while taking into
consideration its neighbors.
• In other words, it measures the node degree but goes a step further
by counting how many links its connections (neighbors ) have.
• It uses the decomposition of the adjacency matrix.
• So, the Eigen centrality for each node is the entry of the A matrix
eigenvector with the biggest eigenvalue.
 Eigenvector centrality
• Eigenvector centrality depends on the entry of the eigenvector
associated with the largest eigenvalue of the adjacency matrix A.
• It is defined as. eigen(i) = ,
• Where is the largest eigenvalue of A, and is the eigenvector
associated with
• It is a global measure, since the eigenvalue decomposition of A
requires complete topological information of the entire network.
Eigenvector centrality
 Closeness
• Closeness is a global measure of shortest path distance of a
node to all other nodes.
• A node is said to have higher closeness if the sum of its
shortest path distance to all other nodes is smaller.

• Let ρ(i, j) denote the shortest path distance between node i and node
j in a connected graph
Closeness(i) =
Closeness
Closeness
 Betweenness
• A close measure to closeness centrality in terms of emphasizing
on shortest paths is the betweenness centrality.
• Betweenness is the fraction of shortest paths passing through a
node relative to the total number of shortest paths in the
network.
• Specifically, it is a global measure defined as
betweenness (i) = ,
where is the total number of shortest paths from k to j, and (i) is the
number of such shortest paths passing through i.
Betweenness

• In the Figure, either

node B or node C can
be removed and there
will still be paths
leading to node E.
Node D, however, is
quite central, since it is
required for any path
leading to node E.
 Local Fiedler Vector Centrality (LFVC)
• A measure that characterizes vulnerability to node removals.
• A node with higher LFVC is more important for network
connectivity structure.
• LFVC is a global centrality measure.
 Network Resilience
• When evaluating network resilience to different centrality attacks.
• Often compare the number of node removals needed by a centrality attack
to reduce the largest component size to a certain amount.
• The number of nodes required to reduce the largest component size to
10% of its original size .
 Network
Resilience
• This network contains 149 nodes
(routers) and 193 edges (physical
connections).
• The topological information needed to
compute the centrality measures are
updated when a node is removed from
the graph
• Resilience of network connectivity to
different centrality attacks on the
Europe Internet backbone network
topology (GTS-CE dataset).
• The largest component size can be
reduced to 20% of its original size by
removing 10 nodes based on LFVC or
betweenness attacks.
 Topological defense scheme
• A topological defense scheme allows change of network
topology to enhance network resilience.
• By swapping a small number of edges in the network topology,
one is able to greatly improve network resilience without
including additional edges.
• The Europe Internet backbone network can be secured by
swapping 20 edges.
• The rewired network is more robust to centrality attacks.
Topological
defense scheme
• Network connectivity of the
edge rewiring method when
restricted to 10 greedy
node removals on the
Europe Internet backbone
network topology (GTS- CE
dataset).
• The edge rewiring method
can greatly improve
network resilience without
introducing additional
edges into the network.
 Game-Theoretic Analysis of Network Robustness
and Fusion-Based Defense Scheme
• In many cases, edge rewire is not permitted in the IoT due to
circumstances such as protocol confinement, geolocation
constraint, and so on.
• A fusion-based defense mechanism is used to infer the
presence of an attack based on the feedbacks from each node.
• The feedback information can be as simple as a binary status
report reflecting that each node is, or is not, under attack, based
on the node-level detection capabilities.
• Then, a network-level attack inference scheme is carried out at
the fusion center.
 Game-Theoretic Analysis of Network Robustness and Fusion-
Based Defense Scheme

• A two-player game between the defender (the

fusion center) and the attacker is naturally formed,
given the critical value of network resilience(e.g.,
the largest component can be no less than 50% of
its original size) and the node-level detection
configurations.
• Intuitively, from the adversary’s perspective, too few
node removals cause hardly any harm to the
network connectivity, while too many node removals
are prone to be detected by the fusion center, which
means that the attack is eventually in vain.
• From the defender’s perspective, inferring attacks
using all feedbacks might treat the topological
attack as a false alarm, since only a small subset of
nodes are targeted. An illustration of the attack and fusion-based defense model
for the IoT
 Game-Theoretic Analysis of Network Robustness and Fusion-Based
Defense Scheme

• On the other hand, inferring attacks using only

a few feedbacks might suffer from information
insufficiency and therefore fail to detect the
presence of attacks.
• Consequently, there exists a balance point at
which both attacker and defender are satisfied
with their own strategies, which is exactly the
notion of Nash equilibrium in game theory .
• At game equilibrium, no player’s payoff can
be increased by unilaterally changing
strategy. As a result, the game payoff at game
equilibrium can be used to study the
robustness of a network. An illustration of the attack and fusion-based defense model
for the IoT
 Network Resilience of IoT
• One can improve network resilience of the IoT by implementing
network-level defense mechanisms.
• However, one main disadvantage of fusion-based defense is
the acquisition of feedbacks from all nodes, which may not be
applicable to the IoT due to its enormous number of devices.
• Nonetheless, fusion-based defense can be used in a
hierarchical manner for multilayer defense.
 Sequential Defense Scheme
• A sequential defense scheme that sequentially collects feedbacks from high
degree nodes for attack inference.
• The advantage of sequential defense is that there is no need to acquire
feedbacks from all nodes.
• Terminates the collection process once sufficient feedback have been
collected for attack inference.
• The enormous network size (e.g., Internet routers or sensors in the IoT)
renders simultaneous data transmissions infeasible, especially for wireless
networks with scarce radio resources.
• Moreover, due to the large network size and limited computational power,
analyzing the collected information from all nodes incurs tremendous
computation overheads, and it may fail to provide timely defense.
Summary
• This chapter introduces several centrality attacks that aim to
maximally disrupt the connectivity of an IoT network.
• Introduce three defense schemes to counter these fatal attacks
• The topological defense scheme
• Allows edge swapping to enhance intrinsic network resilience.
• The fusion-based defense mechanism and the game-theoretic
perspective of network robustness.
• The sequential defense scheme
• Enables efficient attack inference with only a few feedbacks from the network.