Introduction to machine learning

This course provides an overview of machine learning concepts and algorithms. It focuses mostly on techniques related to classification and regression, such as nearest neighbors methods, generalized linear models, tree-based methods, feed-forward neural networks. Simple clustering techniques (k-means clustering, hierarchical clustering) are also introduced. Lecture covers main principles behind different algorithms, model evaluation strategies and basics of statistical learning theory. Connections with topics known from cognitive modeling (e.g., categorization models, signal detection theory) or statistics (e.g., sampling, probability density estimation, logistic regression) are made. During laboratory classes students learn practical applications of the introduced methods using libraries from Python ecosystem (scikit-learn, XGBoost, PyTorch).

Machine learning is a somewhat eclectic field drawing from artificial intelligence, statistics, optimization and data mining. It is pragmatically oriented as it focuses on solving practical problems using various heuristics rather than looking for ‘correct’ solutions. The center of interest is building predictive models based on available data. Machine learning provides general schemes for framing problems (e.g., classification, regression, clustering) and a set of diverse techniques for handling them.

The course is designed to provide an overview of machine learning techniques and foster a specific way of thinking about problems. In accordance with a motto TIMTOWTDI (There is more than one way to do it), multiple ways of approaching similar tasks will be discussed. Material covered will seek a balance between simple algorithms illustrating particular concepts, models interesting from cognitive science perspective, and modern techniques with practical applications.

Lectures will cover necessary theoretical background, including key concepts and some mathematical details. Classes in computer laboratory will focus on applications of the introduced algorithms. Generally, ready-made software packages will be used, students will not be required to implement algorithms on their own.