Abstract:Motivated by the impressive success of deep learning in a wide range of scientific and industrial applications, we explore in this work the application of deep learning into a specific class of optimization problems lacking explicit formulas for both objective function and constraints. Such optimization problems exist in many design problems, e.g., rotor profile design, in which objective and constraint values are available only through experiment or simulation. They are especially challenging when design parameters are high-dimensional due to the curse of dimensionality. In this work, we propose a data-informed deep optimization (DiDo) approach emphasizing on the adaptive fitting of the the feasible region as follows. First, we propose a deep neural network (DNN) based adaptive fitting approach to learn an accurate DNN classifier of the feasible region. Second, we use the DNN classifier to efficiently sample feasible points and train a DNN surrogate of the objective function. Finally, we find optimal points of the DNN surrogate optimization problem by gradient descent. To demonstrate the effectiveness of our DiDo approach, we consider a practical design case in industry, in which our approach yields good solutions using limited size of training data. We further use a 100-dimension toy example to show the effectiveness of our approach for higher dimensional problems. Our results indicate that, by properly dealing with the difficulty in fitting the feasible region, a DNN-based method like our DiDo approach is flexible and promising for solving high-dimensional design problems with implicit objective and constraints.

A Deep Reinforcement Learning Model-based Optimization Method for Graphic Design

Enhancing Topology Optimization with Adaptive Deep Learning

GoodFloorplan: Graph Convolutional Network and Reinforcement Learning-Based Floorplanning

Optimal Design of Composite Grid/skin Structures Based on Deep Learning and Double-Double Layup Strategy

"Deep Reinforcement Learning for Engineering Design Through Topology Optimization of Elementally Discretized Design Domains"

Human-in-the-loop design with machine learning

Deep learning-based topological optimization for representing a user-specified design area

Non-iterative Structural Topology Optimization Using Deep Learning

Generative Design by Reinforcement Learning: Enhancing the Diversity of Topology Optimization Designs

Automated building layout generation using deep learning and graph algorithms

Deep Learning for Explicitly Modeling Optimization Landscapes

Deep learning based inverse method for layout design

Data-informed deep optimization.

Differentiable automatic structural optimization using graph deep learning

Intelligent Generation Method of Innovative Structures Based on Topology Optimization and Deep Learning

Network Topology Optimization via Deep Reinforcement Learning

Algorithmically-consistent deep learning frameworks for structural topology optimization

TO‐NODE: Topology optimization with neural ordinary differential equation

Deep Generative Design: Integration of Topology Optimization and Generative Models

A machine-learning framework for isogeometric topology optimization

Deep learning accelerated topology optimization with inherent control of image quality