This course offers a foundational introduction to the principles, methodologies, and applications of modern machine learning (ML); from foundational principles to cutting-edge methods across diverse data modalities. It is organized into three broad parts. Part I focuses on the foundations of supervised learning, emphasizing classical models such as linear and logistic regression, decision trees, and ensemble methods. The goal is to equip students with the understanding of the ML workflow, model evaluation, and generalization principles. Core concepts such as bias-variance tradeoff, interpretability, uncertainty, robustness, and fairness are introduced early, preparing students to think critically about model performance and reliability. This part also introduces the contemporary learning paradigms including self-supervised, semi-supervised, active, meta, continual, reinforcement, and federated learning to showcase the diverse settings under which ML is developed.

Building on this foundation, Part II introduces modern models including neural networks and advanced architectures such as convolutional, recurrent, and attention-based models. Finally, Part III focuses on ML techniques for specialized data modalities---tabular, graph, temporal, text, and image data---and their practical applications such as forecasting, recommendation, fraud detection.

By the end of the course, students will have a comprehensive understanding of both the theoretical underpinnings and the practical considerations involved in designing and applying machine learning techniques.

• Foundations of Machine Learning: definitions, types, applications, and the ML workflow
• Data preprocessing, feature engineering
• Classical models: linear and logistic regression, decision trees, and ensemble methods (random forests, gradient boosting)
• Model selection, generalization, and hyperparameter optimization
• Core ML concepts: bias–variance trade-off, uncertainty estimation, calibration, interpretability, robustness, and fairness
• Learning paradigms: supervised, semi-supervised, self-supervised, unsupervised, active, meta, continual, federated, and reinforcement learning
• Neural networks and modern architectures: feedforward, convolutional, recurrent, and transformer-based models
• Specialized data modalities: tabular, graph/relational, temporal/sequential, text and image
• Emerging trends: foundation models, multimodal learning

• Explain the fundamental concepts and paradigms of ML, including un-/semi-/supervised learning.
• Apply classical ML algorithms (linear models, decision trees, and ensemble methods) to real-world tasks.
• Compare models using appropriate metrics, cross-validation, while managing overfitting and generalization.
• Understand key ML concepts; including uncertainty estimation, interpretability, robustness, and fairness.
• Describe and compare neural network architectures; CNNs, RNNs, and Transformers, for diverse tasks.
• Differentiate between learning paradigms; self-supervised/reinforcement/meta/continual/federated learning.
• Design ML solutions for specialized data types, including tabular, graph, temporal, text, and image data.
• Critically assess recent developments in foundation and multimodal models and their applications.

This course does not assume prior experience with machine learning theory or practice. However, students are expected to have the following background:

• A basic understanding of probability and statistics
• A working knowledge of linear algebra
• Basic programming skills, including familiarity with Python
• Experience using NumPy, pandas, and matplotlib for data manipulation and visualization

There is no required textbook for the course. I will post course slides for each lecture as well as several recommended chapters for reading (optional) from various books. See Resources for the list of recommended books that could help supplement your understanding of the course material.