Download 7. Decision Trees and Decision Rules

國立雲林科技大學
National Yunlin University of Science and Technology
2005.ACM GECCO.8.Discriminating and
visualizing anomalies using negative
selection and self-organizing maps
Advisor : Dr. Hsu
Presenter : Chih-Ling Wang
Author
: Fabio A. Gonzalez, Juan
Carlos Galeano
ACM GECCO 2005
Intelligent Database Systems Lab
Outline
2

Motivation

Objection

Proper noun

NS-SOM model structure.

Experimentation

Introduction

Background work

Conclusion

My opinion
N.Y.U.S.T.
I. M.
Intelligent Database Systems Lab
Motivation

3
N.Y.U.S.T.
I. M.
The anomaly detection problem could be seen as a classification
problem.
─
First, in many real world problems, only normal samples are available at the
training phase.
─
Second, the set of possible anomalies could be potentially infinite.
Intelligent Database Systems Lab
Objective

4
N.Y.U.S.T.
I. M.
This paper presents a model that can detect anomalies, even when
trained only with normal samples, and can learn from encounters
with new anomalies.
Intelligent Database Systems Lab
Proper noun

5
N.Y.U.S.T.
I. M.
Negative Selection Algorithm
─
The Negative Selection (NS) algorithm is based on the principles of self/noself discrimination in the immune system.
─
It uses as input a set of strings that represents the normal data(self set) in
order to generate detectors in the non-self space. The negative detectors are
chosen by matching them to the self strings: if a detector matches a self string,
it is discarded, otherwise, it is kept.
Intelligent Database Systems Lab
NS-SOM model structure
6
N.Y.U.S.T.
I. M.
Intelligent Database Systems Lab
Experimentation
Unknown
 Iris abnormal
known
abnormal
Primary response
normal

N.Y.U.S.T.
I. M.
Secondary response
Confusion matrices for the secondary response
Wisconsin Breast Cancer
Primary response and secondary response
Confusion matrices of the primary and secondary response
7
Intelligent Database Systems Lab
Background work

8
N.Y.U.S.T.
I. M.
Self/non-self discrimination and immune learning
─
Artificial immunes systems (AIS) which model the self/non-self discrimination
function of the natural immune system (NIS) are mainly based on the NS
algorithm, nevertheless, new models have been proposed, which are based on
danger theory.
─
Both categories of AIS models, self/non-self discrimination based and immune
learning based, have been extensively and independently investigated since the
beginning of AIS research; however, there is not much work on combining
these two approaches in one model.
Intelligent Database Systems Lab
Background work (cont.)

9
N.Y.U.S.T.
I. M.
Anomaly visualization
─
Usually, the anomaly detection problem arises in context where the monitored
system is very complex in structure and function.
─
Information visualization techniques could help to deal with this complexity,
since human perception could detect unexpected features in visual displays and
recall related images to detect anomalies.
─
Most of the work done in anomaly visualization has been restricted to the area
of computer security.
Intelligent Database Systems Lab
Conclusion
10
N.Y.U.S.T.
I. M.

The model combines a negative selection algorithm and a selforganizing map (SOM) in an immune inspired architecture.

One remarkable characteristic of the model is its ability to generate
a 2-dimensional visual representation of the feature space.

This representation facilitates the understanding of the structure of
the self/non-self space by producing a visual discrimination of the
normal, known abnormal and unknown abnormal regions.

This feature could be useful for building interactive visualization
tools.
Intelligent Database Systems Lab
My opinion
11
N.Y.U.S.T.
I. M.

Advantage:Combine AIS mode and immune learning based.

Disadvantage:Lack the mathematic formula.

Apply:Anomaly detection.
Intelligent Database Systems Lab