Machine Learning for Ocean Engineers

An introductory, project-based machine learning course for students with an ocean engineering background. The emphasis is on building working intuition for the standard toolkit and applying it to ocean engineering data through a term project.

Course content

Feature engineering

• Singular value decomposition (SVD), matrix approximations, pseudo-inverse, and least squares
• Principal component analysis (PCA); truncation and alignment
• Fourier series and transforms, DFT/FFT, Gabor transform, and spectrograms
• Wavelets, multi-resolution analysis, and image processing

Regression and classification

• Curve fitting, nonlinear regression, and gradient descent
• Under- and over-determined systems; optimization and the Pareto front
• Cross validation and information criteria for model selection
• Unsupervised learning: k-means, dendrograms, mixture models, and the EM algorithm
• Supervised learning: linear discriminants, perceptron, SVMs, classification trees, and random forests

Neural networks

• Neural networks, activation functions, and backpropagation
• Stochastic gradient descent and deep convolutional networks
• Neural networks for dynamical systems; diversity of network architectures

Learning objectives

Upon successful completion of the course, students will be able to:

• Understand the use of SVD and PCA in data decompositions and apply them to fluid flows
• Develop features for tailoring a problem for application of ML algorithms
• Apply regression and classification approaches on ocean related datasets
• Build and tune data driven models using neural networks for ocean engineering applications
• Develop their own computer program to create and analyze an engineering dataset using the methods introduced in the course