Progress Report on the Learnable Evolution Model
|
Permanent citation URL:
|
Files in This Item:
| Title: | Progress Report on the Learnable Evolution Model |
| Author(s): | Michalski, Ryszard S.; Wojtusiak, Janusz; Kaufman, Kenneth |
| Keywords: | Evolutionary Computation; Function Optimization; Learnable Evolution Model; Guided Evolutionary Computation |
| Issue Date: | 18-Nov-2007 |
| Series/Report no.: | Reports of the Machine Learning and Inference Laboratory MLI 07-2 |
| Abstract: | This report reviews recent research on Learnable Evolution Model (LEM), and presents selected results from its application to the optimization of complex functions and engineering designs. Among the most significant new contributions is a multi operator methodology for generating individuals (candidate solutions) and the employment of a more advanced learning program, AQ21, as the learning module. The new features have been implemented in the LEM3 program. To evaluate LEM3’s performance, it was experimentally compared to other evolutionary computation programs, such as, EA--a conventional, Darwinian-type evolutionary computation program, CA--a cultural evolution algorithm, and EDA--an estimation of distribution algorithm on selected function optimization problems. To determine the scalability of LEM3 and compared programs, the number of variables in the optimized functions was varied from 2 up to 1000. In every experiment, LEM3 outperformed the other programs in terms of the evolution length, sometimes more than an order of magnitude. Another recent research result is the development of early versions of two LEM-based systems, ISHED and ISCOD, for the optimization of heat exchangers evaporators and condensers, respectively. This work was done in collaboration with scientists from the National Institute of Science and Technology. In experimental testing, the systems produced designs that matched or were superior to human designs, particularly, in the cases of nonuniform air flows. This collaboration continues, and may ultimately produce systems that NIST will use to develop better designs of heat exchangers and have them implemented by the industry. |
| URI: | http://hdl.handle.net/1920/2876 |
| Appears in Collections: | Machine Learning and Inference Laboratory |
Items in MARS are protected by copyright, with all rights reserved, to the extent allowed by law.