MACHINE LEARNING MODELS FOR DIRECTED CURATION OF DESIGN SOLUTION SPACE

Search for this publication on Google Scholar
Sjoberg, C. (2017). MACHINE LEARNING MODELS FOR DIRECTED CURATION OF DESIGN SOLUTION SPACE. Unc Charlotte Electronic Theses And Dissertations.
Download PDF

Analytics

107 views ◎
63 downloads ⇓


Abstract

The expanding role of computational models in the process of design is producing exponentially growing parameter spaces. As designers, we must create and implement new methods for searching these parameter spaces considering not only quantitative optimization metrics but also qualitative features. This paper proposes a methodology leveraging pattern modeling properties of artificial neural networks to capture designer’s inexplicit selection criteria and create user-selection based fitness functions for a genetic solver. Through emulation of learned selection patterns, fitness functions based on trained networks provide a method for qualitative evaluation of designs in the context of a given population. The application of genetic solvers for the generation of new populations based on the trained networks selections creates emergent high-density clusters in the parameter space, allowing for the identification of solutions which satisfy the designer’s inexplicit criteria.

Details
Author

Sjoberg, Christian

Title

MACHINE LEARNING MODELS FOR DIRECTED CURATION OF DESIGN SOLUTION SPACE

Physical Description

1 online resource (54 pages) : PDF

Date

2017

Degree Granting Institution

University of North Carolina at Charlotte

Abstract

The expanding role of computational models in the process of design is producing exponentially growing parameter spaces. As designers, we must create and implement new methods for searching these parameter spaces considering not only quantitative optimization metrics but also qualitative features. This paper proposes a methodology leveraging pattern modeling properties of artificial neural networks to capture designer’s inexplicit selection criteria and create user-selection based fitness functions for a genetic solver. Through emulation of learned selection patterns, fitness functions based on trained networks provide a method for qualitative evaluation of designs in the context of a given population. The application of genetic solvers for the generation of new populations based on the trained networks selections creates emergent high-density clusters in the parameter space, allowing for the identification of solutions which satisfy the designer’s inexplicit criteria.

Genre

masters theses

Subjects--Topics

Artificial intelligence
Architecture
Design

Degree

M.S.

Keywords

Evolutionary
Learning
Network
Neural
Parameter

Subject Area

Architecture

Advisor(s)

Sauda, Eric

Committee Members

Dickey, Rachel
Gero, John
Ellinger, Jefferson
Hadzikadic, Mirsad

Degree Note

Thesis (M.S.)--University of North Carolina at Charlotte, 2017.

Rights Statement

This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s). For additional information, see http://rightsstatements.org/page/InC/1.0/.

Rights Holder Information

Copyright is held by the author unless otherwise indicated.

Identifier

Sjoberg_uncc_0694N_11447

Permalink

http://hdl.handle.net/20.500.13093/etd:114