Non-linear Machine Learning for Calibration and Classification

Course Description

This course offers a comprehensive overview of modern machine learning with an emphasis on supervised methods. Students begin by establishing a strong foundation in machine learning fundamentals—including essential nomenclature, clear definitions, and key quality metrics—while also exploring the critical concepts of bias vs. variance trade-off and the distinctions between supervised and unsupervised methods.

From there, the curriculum delves into several pivotal machine-learning algorithms:

• Locally Weighted Regression: Techniques for localized modeling using weighted regression and distance measures.

• Support Vector Machines: Core principles, kernel functions, and both classification and regression applications.

• Artificial Neural Networks: A study of various ANN architectures, effective training procedures, strategies to prevent overfitting, and practical deep learning implementations.

• Gradient Boosted Decision Trees: An introduction to decision trees, focusing on gradient boosting concepts and the practical use of tools like XGBoost.

Prerequisites

Linear Algebra for Machine Learning and Chemometrics, MATLAB for Machine Learning and Chemometrics, Chemometrics I–PCA and Chemometrics II–Regression and PLS or equivalent experience.

Course Outline

1. Introduction 
   – Nomenclature and Definitions.
   – Methods: Unsupervised vs. Supervised.
   – Bias vs. variance trade-off.
   – Model Quality metrics.
2. Machine Learning Algorithms (Methods):
   – Locally Weighted Regression.
   – Support Vector Machines.
   – Artificial Neural Networks.
   – Gradient Boosted Decision Trees. (a brief overview)
   – Model Fusion (Model Ensembles)
3. Variable Contribution (Approaching Model Interpretability):
   – Interferent tests.
   – Single variable tests.
   – Shapley values.
4. Choosing the Right Method and Final Remarks:
   – Computational performance, and deployment options for effective model selection.