Title of paper: 

(1) On Vehicle Surrogate Learning with Genetic Programming Ensembles
(2) A method to learn high-performing and novel product layouts and its application to vehicle design


Authors:

Victor Parque, Waseda University, 3-4-1 Okubo, Shinjuku-ku, Tokyo, Japan, (+81)-3-5286-3249, parque@aoni.waseda.jp
Tomoyuki Miyashita,  Waseda University, 3-4-1 Okubo, Shinjuku-ku, Tokyo, Japan, (+81)-3-5286-3249, tomo.miyashita@waseda.jp

Corresponding Author: Victor Parque (parque@aoni.waseda.jp)

Abstract:

In the above papers we aim at tackling the problem of searching for novel and high-performing product designs. Generally speaking, the conventional schemes usually optimize a (multi) objective function on a dynamic model/simulation, then perform a number of representative real-world experiments to validate and test the accuracy of the some product performance metric. However, in a number of scenarios involving complex product configuration, e.g. optimum vehicle design and large-scale spacecraft layout design, the conventional schemes using simulations and experiments are restrictive, inaccurate and expensive.

In the above papers, we propose a new approach to search for novel and high-performing product designs by optimizing not only a proposed novelty metric, but also a performance function which is learned from historical data. Rigorous computational experiments using more than twenty thousand vehicle models over the last thirty years and a relevant set of well-known gradient-free optimization algorithms show the feasibility and usefulness to obtain novel and high performing vehicle layouts under tight and relaxed search space scenarios.

The promising results of the proposed method opens new possibilities to build unique and high-performing systems in a wider set of design engineering problems.

Criteria List: B, D, G

Statement on Criteria:

Criterion B: 

The research on novel and high-performing product designs is of practical and economic relevance.  We present a method that tackles the problem of designing unique and high-performing products, and shows its effectiveness and usefulness in vehicle design. Though benchmarks are unexistent for algorithmic comparison, it is possible to observe the effectiveness and practical implications based on the rendered vehicle layouts.

In our experiments, we have learned performance surrogate functions based on more than twenty thousand vehicle models over the last thirty years, and used Evolutionary Computing (Differential Evolution, Particle Swarm Optimization and their latest variants) to obtain unique and high-performing vehicle layouts.


Criterion D: In addition to the significant improvement in Big Data-driven product design using Evolutionary Computing, we introduce (1) novelty metrics based on hierarchical clustering algorithm, and (2) surrogates for vehicle performance functions over clusters, allowing finer granularity and enhanced performance in testing phase.


Criterion G: It is clear that Product Design is an important and difficult problem that has been addressed using a wide variety of techniques. Also, learning unique and high-performing products are widely addressed, and rendered useful methods are of practical and economic relevance.


Citation:

V. Parque, T. Miyashita, "On Vehicle Surrogate Learning with Genetic Programming Ensembles", In ACM Genetic and Evolutionary Computation Conference (GECCO), Workshop on Real-world Applications of Continuous and Mixed-integer Optimization

V. Parque, T. Miyashita, "A method to learn high-performing and novel product layouts and its application to vehicle design", Neurocomputing 248: 41-56 (2017).
https://doi.org/10.1016/j.neucom.2016.12.082


Prize Money: Any prize money is to be divided evenly among the co-authors.


Statement to the Judges: 

We have proposed a new approach to search for unique and high-performing vehicle design layouts given observations of historical data. Our application involving real world vehicle data of more than twenty thousand vehicles from the last 30 years, along with rigorous and extensive computational experiments based on Genetic Programming and a eight relevant Evolutionay Optimization algorithms, has shown the feasibility and usefulness of our proposed approach. More specifically, our results and methods allows (1) to learn vehicle performance functions being highly accurate, succinct and economical, (2) to obtain novel vehicle layouts with fairly competitive fuel efficiency ratios under historical boundaries, and (3) to obtain unique vehicle layouts with enhanced fuel-efficiency ratios compared to the actual and historical existing vehicles.

We believe that our approach brings promising results to further advance on learning and pattern recognition algorithms for real world product design, and brings new insights to build unique and high-performing products and high-performing systems, being applicable in a wider set of design engineering problems.


General Type: Genetic Programming, Differential Evolution, Particle Swarm Optimization, Evolutionary Strategy


Date of Publication: 

Accepted: April 13
V. Parque, T. Miyashita, "On Vehicle Surrogate Learning with Genetic Programming Ensembles", In ACM Genetic and Evolutionary Computation Conference (GECCO), Workshop on Real-world Applications of Continuous and Mixed-integer Optimization

26 July 2017
V. Parque, T. Miyashita, "A method to learn high-performing and novel product layouts and its application to vehicle design", Neurocomputing 248: 41-56 (2017).
https://doi.org/10.1016/j.neucom.2016.12.082