M.S. in Data Science Courses

Description: This course is tailored to cover core concepts in probability theory and foundational statistics. It introduces students to the principles and mathematical tools necessary for understanding variability, uncertainty, and decision‐making processes under conditions of uncertainty.

Technical description: This course is tailored to cover core concepts in probability theory and foundational statistics. It introduces students to the principles and mathematical tools necessary for understanding variability, uncertainty, and decision‐making processes under conditions of uncertainty. Emphasizing theoretical understanding, the course prepares learners to grasp more complex topics in future studies.

Description: This course focuses on integrating methods in software and data engineering to provide a comprehensive understanding of data analysis. Students will learn to employ both Python and R to conduct thorough exploratory data analyses, enhancing their practical skills in data handling and visualization.

Technical description: This course focuses on integrating methods in software and data engineering to provide a comprehensive understanding of data analysis. The course also emphasizes good software practices, ensuring participants can manage data effectively and maintain robust analysis processes. It is designed to equip learners with the essential tools and techniques for insightful data exploration and effective decision‐making in real‐world applications.

Description: This advanced course covers regression analysis topics in depth. Students will engage with a range of important regression methods, starting with basic linear regression and progressing through multiple regression techniques, including instrumental variable estimation.

Technical description: This advanced course covers regression analysis topics in depth. Regression analysis refers to a class of methods for predicting continuous variables based on observed data. The prediction itself may not be the ultimate goal. Instead, it may serve as a tool for understanding deeper insights regarding the nature of the phenomena responsible for generating the data.

Description: This course focuses on the application of maximum likelihood techniques in statistical modeling. Maximum likelihood estimation refers to a class of methods for predicting both discrete and continuous variables. The ultimate goal of the investigation may be either the prediction itself, or gaining deeper insights into mechanisms behind the data.

Technical description: This course focuses on the application of maximum likelihood techniques in statistical modeling. The course introduces maximum likelihood estimation and its general properties. Then it explores discrete‐outcome models such as logit, probit, and Poisson regression. It equips students with the skills needed to apply these methods effectively in advanced statistical analysis.

Description: This course provides a detailed introduction to machine learning and its intersection with statistical methods, focusing on foundational concepts, advanced techniques, and practical applications across various domains. Students will explore a range of machine learning paradigms, which include both unlabelled and labelled data. They will also gain a detailed understanding of model evaluation techniques and performance metrics. The course addresses complexities in model building and shows how to make the model capture meaningful patterns in the data without taking too seriously aspects of the data that are present just by random chance. The methods discussed include generalized linear models and decision tree methods, including gradient boosted trees. The course also introduces students to the fundamentals of deep learning, including various neural network architectures and their optimization.

Technical description: This course provides a detailed introduction to machine learning and its intersection with statistical methods, focusing on foundational concepts, advanced techniques, and practical applications across various domains. Students will explore supervised, unsupervised, and self-supervised learning, and gain a detailed understanding of model evaluation metrics. The course addresses complexities in model building (regularization, underfitting, overfitting), generalized linear models, decision tree methods (including gradient boosted trees), and fundamentals of deep learning (various architectures and their optimization).

Description: This course explores important applied areas of machine learning. For computer vision, the course provides detailed understanding of two main types of models: convolutional neural networks and transformers. Convolutional neural networks are traditional neural networks used for image processing. Transformer neural networks are more recent and utilize the so‐called attention mechanism, which retrieves information from other parts of the data as needed. Best performing models often combine both of these neural network types. Through various visualization techniques, students will see firsthand the operation of both convolutions and attention mechanisms. Further, the course covers techniques that leverage unlabelled data, such as contrastive learning and masked autoencoders. The course goes beyond computer vision, exploring, for example, graph neural networks that are useful in many domains.

Technical description: This course examines convolutional neural networks (CNNs) and transformer architectures (which use attention mechanisms), including visualization techniques showing how these layers operate. It also covers methods for unlabelled data like contrastive learning and masked autoencoders, plus applications of graph neural networks for broader tasks.

Description: This course offers an introduction to natural language processing (NLP) and large language models (LLMs). It begins with word vectors and embeddings (a word vector being a set of characteristics related to a word’s meaning) and then moves to the core aspects of LLMs built on the Transformer architecture, including models that predict one word (token) at a time and those that generate their entire output at once.

Students will explore how modern AI systems “think” by reasoning in steps, much like breaking down a complex math problem, and by testing multiple approaches to arrive at the best answer. The course also covers how LLMs integrate with external tools—such as search engines and calculators—to refine their responses, and how they are trained to follow instructions to ensure accuracy and appropriateness.

Practical aspects of model training are covered through traditional fine-tuning and efficient low-rank approximations, alongside techniques like retrieval-augmented generation and large-scale similarity search.

Technical description: This course offers an in-depth exploration of modern natural language processing (NLP) techniques, beginning with foundational concepts such as distributional semantics, word embeddings, and the Transformer architecture, including attention mechanisms and positional encoding. It covers the spectrum of language models, contrasting autoregressive generative models (like GPT) with non-autoregressive masked language models.

Building on these basics, the course proceeds with more developments, examining large language models (LLMs) and contemporary training paradigms, including supervised instruction tuning, few-shot prompting, retrieval-augmented generation, and advanced reasoning strategies like chain-of-thought and tree-of-thought prompting. Students explore recent alignment methods, comparing traditional reinforcement learning from human feedback (RLHF) with innovative techniques such as direct preference optimization (DPO) or group relative policy optimization (GRPO), which enhance model alignment and efficiency.

Additionally, the curriculum addresses the role of synthetic data generation for model improvement, methods for extending context lengths, memory-augmented architectures, and parameter-efficient fine-tuning strategies. Throughout, the emphasis is placed on understanding key concepts, critically evaluating new research, and practical engineering methods necessary for deploying NLP technologies.

Description: This course provides an in‐depth exploration of generative models for image and video data. These are artificial intelligence systems capable of creating images or videos with specified content. The models are composed of statistical variables. Their training builds on traditional statistical methods, such as Bayesian inference, and their functioning is often related to physical phenomena. The course covers generative models such as variational autoencoders, diffusion models, and generative adversarial networks. Generative models of this kind are useful not just for generation itself, but also for capturing deeper meaning behind the data. They are capable of extracting the most important aspects of the data, which is useful for many purposes, such as visualization, data understanding, and data compression.

Technical description: This course provides an in‐depth exploration of generative models for image and video data, with an emphasis on diffusion-based approaches. Foundational techniques such as variational autoencoders (VAEs) are introduced first—detailing the intuition behind their components, the evidence lower bound (ELBo) used in training, and their pivotal role in advancing generative modeling. The course also covers generative adversarial networks (GANs) to offer a well-rounded perspective on classical methods before progressing into more advanced territory.

The curriculum places diffusion models at the forefront. It begins by explaining the physics-inspired intuition behind denoising diffusion probabilistic models and then advances to modern techniques like classifier‐free diffusion guidance, progressive distillation, and denoising diffusion implicit models. Special attention is given to the latest innovations in multimodal generation, such as extending text-to-image diffusion models into text-to-video frameworks by incorporating temporal layers and specialized training methods that ensure temporal consistency. Additionally, theoretical frameworks like flow matching and generator matching are discussed. This course equips students with a comprehensive and forward-looking understanding of generative modeling in today’s rapidly evolving field.

Description: This course covers the essentials of economics and game theory. It introduces the game‐theory perspective on the behavior of humans and other agents. After exploring interactions between few agents, the course provides a comprehensive introduction to markets with different market structures. Then the course covers applications to designing socially optimal policies, various microeconomic and macroeconomic issues, and financial markets.

Technical description: This course covers the essentials of economics and game theory. It introduces the game‐theory perspective on the behavior of humans and other agents. After exploring interactions between few agents, the course provides a comprehensive introduction to markets with different market structures. Then the course covers applications to designing socially optimal policies, various microeconomic and macroeconomic issues, and financial markets.

The Applied Data Science Practicum serves as the capstone experience for the MS in Data Science. Students entering the practicum have completed all other coursework. This course provides an opportunity for students to apply their cumulative knowledge and skills in a real‐world setting.