01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
Elsevier AMS Job code: BFN Ch05-N52710 23-1-2007 8:05 a.m. Page:149 Trimsize:165×240 MM
Quantitative Structure-Activity Relationships (QSAR) for Pesticide Regulatory Purposes 149
Edited by Emilio Benfenati
© 2007 Elsevier B.V. All rights reserved.
Chapter 5
Hybrid systems
Nicolas Amaury
1
, Emilio Benfenati
2
, Severin Bumbaru
3
, Antonio Chana
2
, Marian
Craciun
3
, Jacques R. Chrétien
1
, Giuseppina Gini
4
, Gongde Guo
5
, Frank Lemke
6
,
Viorel Minzu
3
, Johann-Adolf Müller
6
, Daniel Neagu
5
, Marco Pintore
1
, Silviu Augustin
Stroia
3
, Paul Trundle
5
1
BCX, Biochemics Consulting SAS, Orléans, France
2
Laboratory of Environmental Chemistry and Toxicology, Istituto di Ricerche
Farmacologiche “Mario Negri”. Milano, Italy
3
Department of Computer Science and Engineering, University “Dunarea de Jos”, Galati,
Romania
4
Department of Electronic and Information, Politecnico di Milano, Milano, Italy
5
Department of Computing, School of Informatics, University of Bradford, Bradford, UK
6
KnowledgeMiner, Berlin, Germany
1. INTRODUCTION: GOALS OF THE HYBRID SYSTEMS
The term ‘hybrid system’ dates back to the development of expert systems
(Jackson, 1999) from the initial rule-based aspect to the modern modeling and
interpretation systems. Most of the accent in the beginning has been on the idea of
making use of more representations of the problem, more paradigms of knowledge
representation, and more algorithms to find a solution.
A seminal work by Gallant (1993) introduced a way to look together to
neural networks and rule-based systems. In his approach, a net, built from data and
in absence of symbolic knowledge, is used to extract rules. This idea developed
in the artificial intelligence (AI) community the well-known area of integrating
connectionist and symbolic systems.
The starting machine learning community developed in the same years
another way to make use of data in the absence of knowledge that led to
the development of inductive trees, well exemplified by C4.5 (Quinlan, 1993).
Integrating different representations and solutions is a direction taken in AI in the
years around 1995. The term ‘expert system’ in those years was almost replaced
by the term ‘intelligent system’ or ‘intelligent agent’.
Using different representations to reach a common agreement or a problem
solution led to the idea of using computational different methods on different
problem representations, to make use of their relative strengths. Examples are the