Author: Eesh Kamrah
Date: Monday, April 10th, 2023 at 1:30 pm
Location: EGR-4164B
Committee Members:
- Dr. Mark Fuge / Chair
- Dr. Shapour Azarm
- Dr. Nikhil Chopra
Title of Thesis: ‘A STUDY TO EVALUATE WHEN TO NOT USE DIVERSE EXAMPLES
Abstract:
Design researchers have struggled to provide quantifiable forecasts on whether diversity helps or hinders the design search process. [1] This thesis studies this problem by answering the following question “how and when do diverse initial stimuli lead to better quality designs?” It does so by presenting a design research study on a modular ND test problem and Delta Design Game [2].
During our investigation some methods were developed as they couldn’t be found in popular literature. For example, a sampling method that can sample both relatively less and highly diverse initial data; we addressed this by developing a fast DPP rank based diversity method. Next, we could not find a modular test function that has parameters that would allow us to control the function’s ruggedness/complexity, we addressed this by creating a modular test function based on an existing concept that can be used to generate test functions of similar and varying complexity.
The thesis is thus laid out in a manner so that initially we address the methods that are necessary to understand the study. This begins with Chapter 2, where we look at how the diversity data sampling method can be used to generate diverse and less-diverse examples. In Chapter 3, we look at the different test problems that have been developed that are used in the study. Once familiar with this, the latter part of the thesis shows the results that provide evidence that initializing an optimizer with diverse examples is not always beneficial. We also identify the conditions in which the less diverse initial examples perform better. In the last chapter we identify the limitations of the provided results and how further experiments can be designed to investigate the identified limitations.
[1] Fu, Katherine, Joel Chan, Jonathan Cagan, Kenneth Kotovsky, Christian Schunn, and Kristin Wood. “The Meaning of ‘Near’ and ‘Far’: The Impact of Structuring Design Databases and the Effect of Distance of Analogy on Design Output.” Journal of Mechanical Design 135, no. 2 (January 7, 2013). https://doi.org/10.1115/1.4023158.
[2] Nepf, Heidi, Herbert Einstein, and Louis Buicciarelli. “Delta Game | Introduction to Civil and Environmental Engineering Design I | Civil and Environmental Engineering | MIT OpenCourseWare.” MIT – Course Information. INTRODUCTION TO CIVIL AND ENVIRONMENTAL ENGINEERING DESIGN I, Fall 2016. https://ocw.mit.edu/courses/1-101-introduction-to-civil-and-environmental-engineering-design-i-fall-2006/pages/delta-game/.