T Cell–Inspired Algorithms - Immunological Computation: Theory and Applications - page 103

Dasgupta and Gonzalez (2002) compared the negative characterization
approach to a positive characterization method (“ kd -tree”). Although the positive
characterization approach gave more precise results, it was more costly in time and
space when compared to the negative characterization technique. In another work,
Gonzalez (2003) made a comparison of “hybrid neuro-immune system” (HNIS)
and “self-organizing maps” (SOM) for anomaly detection application.
Ji and Dasgupta (2006) used “support vector machine” (SVM) as the positive
detection technique and compared with the V-detector algorithm. SVM appears
to have provided more generalization, but the performance primarily depends on a
suitable choice of the kernel function.
Both PS and NS can be a reasonable choice based on various reasons. An
application with a large amount of self-data seems appropriate for NS.

4.9 Negative Database

h e negative database (NDB; Esponda et al., 2004; Esponda, 2005) is one of latest
developments in NSAs. Accordingly, the “self ” is a database that stores a collection
of data records referred to as positive database (PDB), and the complementary data-
base, NDB, constitutes all possible records that are not in DB. h e interesting
property of this representation concerns the di culty of inferring DB from a given
NDB, thus, enhancing the privacy of the sensitive information without any
encryption.

4.9.1 Negative Database Representation

Given an arbitrary set of strings l -length defi ned over {0, 1, *}, where * is the “don't
care” symbol, determining which strings are not represented in NDB is an NP-
hard problem. As all the possible strings that are not in DB constitute the NDB,
NDB

= −

DB, where U denotes the universe defi ned over the same alphabet
(Esponda et al., 2004). A string s is in DB if and only if it fails to match all entries
in the NDB (Table 4.3).

U

Table 4.3 Example of a DB, Its Corresponding U-DB,

and a Possible NDB Representing U-DB

DB (Self) (U-DB) NDB (Nonself)

010 000 *00

101 001 0*1

011 *11

100
110
111