Introduction to Machine Learning

Aim

After completion of the course the student should at an advanced level be able to:

• use relevant concepts and methods from machine learning in order to formulate, structure and solve practical problems that involve large or complex data,
• apply an inference for the parameter values for commonly used machine learning models,
• use machine learning models for prediction and decision making,
• estimate the quality of the machine learning models,
• critically evaluate given task in order to select a suitable model in situations with a limited or no information about the underlying dependencies in the data,
• implement machine learning models in a programming language and use existing machine learning software in order to analyze large and/or complex datasets, make predictions and estimate the uncertainty of these predictions.
• reflect on ethical and societal aspects relevant to machine learning,
• reflect on the possibilities and limitations of machine learning, its role in society and people's responsibility for how it is used,
• plan and carry out course work within given time frames.

Content

The course focuses on the main concepts and tools in probabilistic machine learning that are necessary for professional work and research in data analysis.

The course covers the following topics:

• introduction and overview of machine learning (including regression, classification, supervised and unsupervised learning) and its application areas,

• Nearest Neighbors, logistic regression and decision trees,

• binary classification: precision, recall, F1 score and ROC curves.

• model selection and uncertainty estimation: holdout method, cross-validation, degrees of freedom, bias-variance tradeoff and confidence intervals,

• linear regression and regularization methods (Ridge, LASSO),

• latent variables and principal component analysis (PCA)

• kernel smoothers, kernel trick and support vector machines,

• neural networks,

• bagging, boosting and random forests,

• mixture models,

• optimization for machine learning: gradient descent and stochastic gradient descent,

• implicit regularization: early stopping.