.........What is IEC?

The Interactive Evolutionary Computation encompasses computational models based on the joint search of the human being and the EC. Thus, in these techniques the adequacy values (fitness values) must be estimated by a human being (Interactive Evolutionary Computation, IEC)

As can be seen in figure 1, the general scheme of an IEC algorithm is very simple: the user interacts with the IEC system trough the interfaces, evaluating the outputs evolved by the EC application, responsible of applying the optimization mechanisms in order to generate the best outputs based on the users' past selections.

Figure 1. General scheme of an IEC system

Therefore, the IEC is a technique that allows to collect the users' preferences, the intuition, the emotions or determined psychological aspects. Then, it uses them to evolve the system towards the next generations.

There are two main definitions of IEC, \cite{Takagi01}, a narrow one and a wide one, that contain the main features of IEC:

1. An EC technique whose fitness function is replaced by a human user.

2. The IEC is the technology that EC optimizes the target systems having an interactive human-machine interface.

Both definitions are based on the term 'EC'. The term evolutionary computation (EC) covers several techniques using mechanisms similar to natural selection perform evolution within a computer. Examples of evolutionary computation systems include genetic algorithms (GA), genetic programming (GP), evolutionary computation (EC) or evolutionary strategies (ES).

IEC main features

The Interactive Evolutionary Computation is a technique based on the joint search of the human being and the EC. This technique tries to reach a common target by establishing a relation between the human being search space (psychological) and the parameters and outputs of the algorithm. The relation is set up during the following evaluations made by the human being over the different outputs from the algorithm.

Two types of approach are commonly used for developing systems with subjective evaluation of humans:

  • Analytical approach: the knowledge of the human or expert is collected and later integrated in the system in the evaluation process. They are frequently implemented in research works related to Artificial Intelligence (AI), but obviously, that knowledge is difficult to detect and to store. This objection makes this approach complex and problem dependant.
  • Synthetic approach: this approach includes the user evaluation like a 'black box' evaluation process \cite{Takagi00}. Namely, the user does not evaluate directly the internal features of the individual generated by the EC algorithm, instead evaluates the visible characteristics.

However, in spite of implementing any of both approaches, there is a unavoidable problem inherent to human being: the doubts or preferences change. After a determined proposed solution of the system the user can change his or her preferences, and the IEC algorithm must be able to collect those changes. Thus, the search must be robust to face up to the changes of the users' preferences \cite{Takagi96}.

Another problem linked to IEC techniques is the user fatigue during the evaluation process. This problem forces the IEC algorithms to work in each iteration with very small populations, due to the fact that the user can not evaluate simultaneously numerous individuals (images, sounds, etc.), and besides, for the same reason the maximum number of generations is limited to 10 to 20, \cite{Takagi98}, \cite{saez03}.

To solve these problems it is necessary to improve the design of simply, intuitive and efficient human-computer interfaces, and to improve the existing EC algorithms in order to converge in less iterations and work with smaller populations.

IEC Related Work

Computer Graphics (CG)
Music & Sound

Sound & Signal Treatment
Image processing
Document design/edition
Industrial Design
E-Commerce & Internet

BDD Searching tools
Data Mining and Knowledge Management
Face Recognition

2005, University CARLOS III of Madrid