One-shot generation of near-optimal topology through theory-driven machine learning

Ruijin Cang, Hope Yao, Yi Ren

Research output: Contribution to journalArticle

1 Citation (Scopus)

Abstract

We introduce a theory-driven mechanism for learning a neural network model that performs generative topology design in one shot given a problem setting, circumventing the conventional iterative process that computational design tasks usually entail. The proposed mechanism can lead to machines that quickly respond to new design requirements based on its knowledge accumulated through past experiences of design generation. Achieving such a mechanism through supervised learning would require an impractically large amount of problem–solution pairs for training, due to the known limitation of deep neural networks in knowledge generalization. To this end, we introduce an interaction between a student (the neural network) and a teacher (the optimality conditions underlying topology optimization): The student learns from existing data and is tested on unseen problems. Deviation of the student's solutions from the optimality conditions is quantified, and used for choosing new data points to learn from. We call this learning mechanism “theory-driven”, as it explicitly uses domain-specific theories to guide the learning, thus distinguishing itself from purely data-driven supervised learning. We show through a compliance minimization problem that the proposed learning mechanism leads to topology generation with near-optimal structural compliance, much improved from standard supervised learning under the same computational budget.

Original languageEnglish (US)
Pages (from-to)12-21
Number of pages10
JournalCAD Computer Aided Design
Volume109
DOIs
StatePublished - Apr 1 2019

Fingerprint

Learning systems
Supervised learning
Topology
Students
Neural networks
Shape optimization
Compliance

Keywords

  • Active learning
  • Meta-learning
  • Topology optimization

ASJC Scopus subject areas

  • Computer Science Applications
  • Computer Graphics and Computer-Aided Design
  • Industrial and Manufacturing Engineering

Cite this

One-shot generation of near-optimal topology through theory-driven machine learning. / Cang, Ruijin; Yao, Hope; Ren, Yi.

In: CAD Computer Aided Design, Vol. 109, 01.04.2019, p. 12-21.

Research output: Contribution to journalArticle

@article{3f51477bba854c77b6e0edf89167b5ec,
title = "One-shot generation of near-optimal topology through theory-driven machine learning",
abstract = "We introduce a theory-driven mechanism for learning a neural network model that performs generative topology design in one shot given a problem setting, circumventing the conventional iterative process that computational design tasks usually entail. The proposed mechanism can lead to machines that quickly respond to new design requirements based on its knowledge accumulated through past experiences of design generation. Achieving such a mechanism through supervised learning would require an impractically large amount of problem–solution pairs for training, due to the known limitation of deep neural networks in knowledge generalization. To this end, we introduce an interaction between a student (the neural network) and a teacher (the optimality conditions underlying topology optimization): The student learns from existing data and is tested on unseen problems. Deviation of the student's solutions from the optimality conditions is quantified, and used for choosing new data points to learn from. We call this learning mechanism “theory-driven”, as it explicitly uses domain-specific theories to guide the learning, thus distinguishing itself from purely data-driven supervised learning. We show through a compliance minimization problem that the proposed learning mechanism leads to topology generation with near-optimal structural compliance, much improved from standard supervised learning under the same computational budget.",
keywords = "Active learning, Meta-learning, Topology optimization",
author = "Ruijin Cang and Hope Yao and Yi Ren",
year = "2019",
month = "4",
day = "1",
doi = "10.1016/j.cad.2018.12.008",
language = "English (US)",
volume = "109",
pages = "12--21",
journal = "CAD Computer Aided Design",
issn = "0010-4485",
publisher = "Elsevier Limited",

}

TY - JOUR

T1 - One-shot generation of near-optimal topology through theory-driven machine learning

AU - Cang, Ruijin

AU - Yao, Hope

AU - Ren, Yi

PY - 2019/4/1

Y1 - 2019/4/1

N2 - We introduce a theory-driven mechanism for learning a neural network model that performs generative topology design in one shot given a problem setting, circumventing the conventional iterative process that computational design tasks usually entail. The proposed mechanism can lead to machines that quickly respond to new design requirements based on its knowledge accumulated through past experiences of design generation. Achieving such a mechanism through supervised learning would require an impractically large amount of problem–solution pairs for training, due to the known limitation of deep neural networks in knowledge generalization. To this end, we introduce an interaction between a student (the neural network) and a teacher (the optimality conditions underlying topology optimization): The student learns from existing data and is tested on unseen problems. Deviation of the student's solutions from the optimality conditions is quantified, and used for choosing new data points to learn from. We call this learning mechanism “theory-driven”, as it explicitly uses domain-specific theories to guide the learning, thus distinguishing itself from purely data-driven supervised learning. We show through a compliance minimization problem that the proposed learning mechanism leads to topology generation with near-optimal structural compliance, much improved from standard supervised learning under the same computational budget.

AB - We introduce a theory-driven mechanism for learning a neural network model that performs generative topology design in one shot given a problem setting, circumventing the conventional iterative process that computational design tasks usually entail. The proposed mechanism can lead to machines that quickly respond to new design requirements based on its knowledge accumulated through past experiences of design generation. Achieving such a mechanism through supervised learning would require an impractically large amount of problem–solution pairs for training, due to the known limitation of deep neural networks in knowledge generalization. To this end, we introduce an interaction between a student (the neural network) and a teacher (the optimality conditions underlying topology optimization): The student learns from existing data and is tested on unseen problems. Deviation of the student's solutions from the optimality conditions is quantified, and used for choosing new data points to learn from. We call this learning mechanism “theory-driven”, as it explicitly uses domain-specific theories to guide the learning, thus distinguishing itself from purely data-driven supervised learning. We show through a compliance minimization problem that the proposed learning mechanism leads to topology generation with near-optimal structural compliance, much improved from standard supervised learning under the same computational budget.

KW - Active learning

KW - Meta-learning

KW - Topology optimization

UR - http://www.scopus.com/inward/record.url?scp=85059572783&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85059572783&partnerID=8YFLogxK

U2 - 10.1016/j.cad.2018.12.008

DO - 10.1016/j.cad.2018.12.008

M3 - Article

VL - 109

SP - 12

EP - 21

JO - CAD Computer Aided Design

JF - CAD Computer Aided Design

SN - 0010-4485

ER -