In previous work, a method for isohedral tiling design was proposed, which analytically produces a tillable shape similar to a given shape. Whereas this method can obtain an optimal figure, its output changes depending on the small change of the input shape. Thus, it requires trial and error for slightly changes of the input shape.

The proposed method selects input point set and tiling pattern using genetic algorithm, allowing us to design better tillable shape with less trial and error effort than the previous method.

Asuka Hisatomi, Hitomi Koba, Makoto Kamizono, Kazunori Mizuno, Satoshi Ono: “Escher-like Tiling Design Using Hierarchical Optimization”, Genetic and Evolutionary Computation Conference (GECCO) , companion, pp.89-90 (2017).

Makoto Kamizono, Kigo Shimomura, Masayuki Tajiri, Satoshi Ono: “Two-Dimensional Barcode Decoration Using Module-wise Non-systematic Coding and Cooperative Evolution by User and System”, Genetic and Evolutionary Computation Conference (GECCO) , pp.245-252 (2016).

Free access until 26th March, 2014:
http://elsarticle.com/1ej7JRE

—

This study proposes a cooperative evolutionary optimization method between a user and system (CEUS) for problems involving quantitative and qualitative optimization criteria. In a general Interactive Evolutionary Computation (IEC) model, both the system and user have their own role in the evolution, such as individual reproduction or evaluation. In contrast, the proposed CEUS allows the user to dynamically change the allocation of search roles between the system and user, resulting in simultaneous optimization of qualitative and quantitative objective functions without increasing user fatigue. This is achieved by a combination of user evaluation prediction and the integration of interactive and non-interactive EC. For instance, the system performs a global search at the beginning, the user then intensifies the search area, and finally the system conducts a local search in the intensified search area. This study applies CEUS to an image processing filter design problem that involves both quantitative (filter output accuracy) and qualitative (filter behavior) criteria. Experiments have shown that the proposed CEUS can design image filters in accordance with user preferences, and CEUS interacting with a non-naive user enhanced the initial global search so that it converged and found a reasonable solution more than four times faster than a non-interactive search.

In general Interactive Evolutionary Computation, a role assignment between a user and system and timing of user operation were determined and not changed during the search. In addition, although some methods for user fatigue alleviation by prediction have been proposed, the methods require a certain amount of learning data.

The proposed method allows the user to dynamically change search role allocation between the system and the user by user evaluation prediction and integration of interactive and non-interactive evolutionary computation, resulting in simultaneous optimization of qualitative and quantitative objective functions without increasing user fatigue.