Machine Learning for Engineers: Using data to solve problems for physical systems

✍ Scribed by Ryan G. McClarren

• Publisher: Springer
• Year: 2021
• Tongue: English
• Leaves: 252
• Category: Library

✦ Synopsis

All engineers and applied scientists will need to harness the power of machine learning to solve the highly complex and data intensive problems now emerging. This text teaches state-of-the-art machine learning technologies to students and practicing engineers from the traditionally “analog” disciplines―mechanical, aerospace, chemical, nuclear, and civil. Dr. McClarren examines these technologies from an engineering perspective and illustrates their specific value to engineers by presenting concrete examples based on physical systems. The book proceeds from basic learning models to deep neural networks, gradually increasing readers’ ability to apply modern machine learning techniques to their current work and to prepare them for future, as yet unknown, problems. Rather than taking a black box approach, the author teaches a broad range of techniques while conveying the kinds of problems best addressed by each. Examples and case studies in controls, dynamics, heat transfer, and other engineering applications are implemented in Python and the libraries scikit-learn and tensorflow, demonstrating how readers can apply the most up-to-date methods to their own problems. The book equally benefits undergraduate engineering students who wish to acquire the skills required by future employers, and practicing engineers who wish to expand and update their problem-solving toolkit.

✦ Table of Contents

Preface
Contents
Acronyms
Part I Fundamentals
1 The Landscape of Machine Learning
1.1 Supervised Learning
1.1.1 Regression
1.1.1.1 Overfitting
1.1.2 Classification
1.1.3 Time Series
1.1.4 Reinforcement Learning
1.2 Unsupervised Learning
1.2.1 Finding Structure
1.2.2 Association Rules
1.3 Optimization and Machine Learning vs. Simulation
1.4 Bayesian Probability
1.5 Cross-Validation
Notes and Further Reading
Problems
References
2 Linear Models for Regression and Classification
2.1 Motivating Example: An Object in Free Fall
2.2 General Linear Model
2.2.1 Nonlinear Transformations
2.3 Logistic Regression Models
2.3.1 Comparing to the Null Model
2.3.2 Interpreting the Logistic Model
2.3.3 Multivariate Logistic Models
2.3.4 Multinomial Models
2.4 Regularized Regression
2.4.1 Ridge Regression
2.4.2 Lasso and Elastic Net Regression
2.5 Case Study: Determining Governing Equations from Data
2.5.1 Determining the Form of the Coefficients
2.5.2 Discussion of the Results
Notes
Problems
References
3 Decision Trees and Random Forests for Regression and Classification
3.1 Decision Trees for Regression
3.1.1 Building Regression Trees
3.2 Classification Trees
3.3 Random Forests
3.3.1 Comparison of Random Forests and Tree Models
3.4 Case Study: Predicting the Result of a Simulation Using Random Forests
3.4.1 Using the Random Forest Model to Calibrate the Simulation
Notes
Problems
References
4 Finding Structure Within a Data Set: Data Reduction and Clustering
4.1 Singular Value Decomposition
4.2 Case Study: SVD to Understand Time Series
4.3 K-means
4.3.1 K-means Example
4.4 t-SNE
4.4.1 Computing t-SNE
4.4.2 Example of t-SNE
4.5 Case Study: The Reflectance and Transmittance Spectra of Different Foliages
4.5.1 Reducing the Spectra to Colors
Notes
Problems
References
Part II Neural Networks
5 Feed-Forward Neural Networks
5.1 Simple Neural Network
5.1.1 Example Neural Network
5.1.2 Why Is This Called a Neural Network?
5.1.3 The Activation Function, σ(u)
5.2 Training Single Layer Neural Networks
5.2.1 Multiple Training Points
5.2.2 Data Normalization and Training Neural Networks
5.2.2.1 Stochastic Gradient Descent
5.2.2.2 Issues with Gradient Descent
5.3 Deep Neural Networks
5.3.1 Example Deep Neural Network
5.3.2 Training Deep Neural Networks
5.3.3 Neural Networks for Classification Problems
5.4 Regularization and Dropout
5.4.1 Dropout
5.5 Case Study: The Strength of Concrete as a Function of Age and Ingredients
5.5.1 Neural Networks with Raw Independent Variables
5.5.2 Neural Networks with Additional Features
Notes and Further Reading
Problems
References
6 Convolutional Neural Networks for Scientific Images and Other Large Data Sets
6.1 Convolutions
6.1.1 Discrete Convolutions
6.1.2 Two-Dimensional Convolutions
6.1.3 Multi-Channel 2-D Convolutions
6.2 Pooling
6.2.1 2-D Pooling
6.3 Convolutional Neural Networks
6.4 Case Study: Classification with Fashion MNIST
6.5 Case Study: Finding Volcanoes on Venus with Pre-fit Models
Notes and Further Reading
Problems
References
Part III Advanced Topics
7 Recurrent Neural Networks for Time Series Data
7.1 Basic Recurrent Neural Networks
7.1.1 Training RNNs and Vanishing Gradients
7.1.2 Example RNN: Finding the Frequency and Shift of an Oscillating Signal
7.2 Long Short-Term Memory (LSTM) Networks
7.2.1 How LSTMs Get Around the Vanishing Gradients Problem
7.2.2 Variants to the LSTM Network
7.2.3 Example LSTM: Finding the Frequency and Shift of an Oscillating Signal
7.3 Case Study: Determining the Behavior of a Cart-Mounted Pendulum
7.3.1 Model Design and Results
Notes and Further Reading
Problems
References
8 Unsupervised Learning with Neural Networks: Autoencoders
8.1 Fully Connected Autoencoder
8.2 Compressing Leaf Spectra with an Autoencoder
8.3 Convolutional Autoencoders
8.3.1 Fashion MNIST Demonstration of Convolutional Autoencoders
8.3.2 Noise Reduction with Convolutional Autoencoders
8.4 Case Study: Reducing the Data from a Physics Simulation with Autoencoders
8.4.1 Convolutional Autoencoder Set Up and Training
Removing Noise with the Autoencoder
8.4.2 LSTM with Inputs from the Autoencoder
Notes and Further Reading
Problems
References
9 Reinforcement Learning with Policy Gradients
9.1 Reinforcement Learning to Swing the Cart-Mounted Pendulum
9.1.1 Reward Function
9.1.2 Controlling the Cart Pendulum System
9.2 Training a Neural Network with Policy Gradients
9.3 Policy Gradient Training and Results for the Cart-Mounted Pendulum
9.4 Case Study: Control of Industrial Glass Cooling with Policy Gradients
Notes and Further Reading
Problems
References
A Data and Implementation of the Examples and Case Studies
A.1 Scikit-Learn
A.1.1 Training and Using ML Models
A.1.2 Evaluating Model Performance and Cross-Validation
A.1.3 Preprocessing and Loading Data
A.2 Tensorflow
Index