Information Technology Reference
In-Depth Information
2.2 Dendritic Cell Algorithm
This paper focuses specifically on the Dendritic Cell Algorithm [5,6,7] of Greensmith
et al
, which abstracted a number of properties of DCs that are possibly advantageous
to design AIS for anomaly detection.
In the human immune system, during the antigen ingestion process, immature DCs
experience different types of signals that indicate the context (either safe or
dangerous) of an environment where the digested antigens exist. The different types
of signals lead DCs to differentiate into two types: mature and semi-mature. Chemical
messages known as cytokines produced by mature and semi-mature DCs are different
and influence the differentiation of naïve T-cells into several distinctive paths such as
helper T-cells or killer T-cells. In order to employ these properties of DCs,
Greensmith
et al.
categorised DC input signals into four groups -
PAMPs
(signals
known to be pathogenic),
Safe Signals
(signals known to be normal),
Danger Signals
(signals that may indicate changes in behaviour) and
Inflammatory Cytokines
(signals
that amplify the effects of other signals). When each artificial DC experiences the
combination of these four different signal groups released by the artificial tissue, it
interprets the context of ingested antigens by using a signal processing function,
which weights each type of input signal differently. The output of a signal processing
function determines the differentiation status of DCs (either semi-mature or mature).
3 Artificial Immune Systems Applied to Sensor Networks
The parallels between intrusion detection and immunity have long been the source of
inspiration for AIS researchers, but conventional computer networks do not closely
resemble the dynamic, distributed and fluid nature of organisms and their immune
systems well. There is, however, a type of network that does share many of these
features: sensor networks. In the following sections, we introduce this type of network
and outline one popular routing protocol, known as
Directed Diffusion
[9].
3.1 Sensor Network Overview
Sensor networks are an emerging technology and research area in the rapidly growing
field of ubiquitous computing [4], aimed at providing distributed and massively
parallel monitoring in heterogeneous physical environments. Sensors are typically
low-cost, limited capacity, mass production units, consisting of no more than (i) a
sensing unit, (ii) a processing unit, (iii) memory, (iv) a transceiver and (v) a power
unit [2]. Their aim is two fold: (i) to faithfully execute their intended task, and (ii) to
efficiently manage their limited resources, such as energy, so as to maximise their
lifetime. The following features of sensor networks distinguish them from traditional
computing environments [2, 4]:
P1:
Constrained resources - limited in physical capacity, bandwidth, cost, etc.
P2:
High-density - number and density of sensor nodes can be several orders of
magnitude higher than the mobile nodes in an ad hoc mobile network.
