Page 126 - Artificial Intelligence in the Age of Neural Networks and Brain Computing

1. From Aristotle’s Logic to Artificial Neural Networks and Hybrid Systems 115

• Recurrent ANN and reinforcement learning [20].
• Other [21].

Some of the ANN were considered “black boxes” as it was difficult to interpret
their internal structures and to articulate the essential knowledge learned. That led to
the development of hybrid ANN ruleebased systems.

1.4 INTEGRATING ANN WITH RULE-BASED SYSTEMS: HYBRID
CONNECTIONIST SYSTEMS
In order to incorporate human knowledge into an intelligent system, an ANN module
can be combined with a rule-based module in the same system. The rules can be
fuzzy rules as a partial case [22]. An exemplar system is shown in Fig. 6.3, where,
at a lower level, an ANN module predicts the next day value of a stock index and,
at a higher level, a fuzzy reasoning module combines the predicted values with
some macroeconomic variables, using the following types of fuzzy rules [22]:
IF <the stock value predicted by the ANN module is high>
AND <the economic situation is good>
THEN <buy stock>
Hybrid systems can also use crisp propositional rules, along with fuzzy rules [23].
The type of hybrid systems illustrated in Fig. 6.3 is suitable to use when decision rules
are available to integrate with a machine learning module that learns from incoming
data.
Another group of ANN methods can be used not only to learn from data, but to
extract rules from a trained ANN and/or insert rules into an ANN as initialization
procedure. These are the neurofuzzy systems as discussed in the next section on
the case of the evolving connectionist systems (ECOS).
The integration of ANN and fuzzy systems into one system attracted many
researchers. The integration of fuzzy rules into a single neuron model and then

Fuzzified data
Rules
Neural Trading rules
extraction
network (fuzzy)
module
Current price Predicted price
Neural
network
Yesterday's price (crisp value)
(crisp values)

Political situation Fuzzy Decision (buy/sell/hold)
rule based
Economic situation decision (fuzzy & crisp values)
(fuzzy values)
FIGURE 6.3
A hybrid ANN-fuzzy ruleebased expert system for financial decision support [22].