Degree Requirements – Machine Learning Sub-Plan

Bayesian modeling and inference is a powerful modern approach to representing the statistics of the world, reasoning about the world in the face of uncertainty, and learning about it from data. It cleanly separates the notions of representation, reasoning, and learning. It provides a principled framework for combining multiple source of information such as prior knowledge about the world with evidence about a particular case in observed data. This course will provide a solid introduction to the methodology and associated techniques, and show how they are applied in diverse domains ranging from computer vision to molecular biology to astronomy.  Graduate-level requirements include different exams requiring greater depth of understanding of topics, and will be assigned questions based on graduate-student specific assignments topics.

This course will introduce students to the concepts and techniques of data mining for knowledge discovery. It includes methods developed in the fields of statistics, large-scale data analytics, machine learning, pattern recognition, database technology and artificial intelligence for automatic or semi-automatic analysis of large quantities of data to extract previously unknown interesting patterns. Topics include understanding varieties of data, data preprocessing, classification, association and correlation rule analysis, cluster analysis, outlier detection, and data mining trends and research frontiers. We will use software packages for data mining, explaining the underlying algorithms and their use and limitations. The course include laboratory exercises, with data mining case studies using data from many different resources such as social networks, linguistics, geo-spatial applications, marketing and/or psychology.

This course introduces the key concepts underlying statistical natural language processing. Students will learn a variety of techniques for the computational modeling of natural language, including: n-gram models, smoothing, Hidden Markov models, Bayesian Inference, Expectation Maximization, Viterbi, Inside-Outside Algorithm for Probabilistic Context-Free Grammars, and higher-order language models.  Graduate-level requirements include assignments of greater scope than undergraduate assignments. In addition to being more in-depth, graduate assignments are typically longer and additional readings are required.

This course provides a broad technical introduction to the tools, techniques and concepts of artificial intelligence. The course will focus on methods for automating decision making under a variety of conditions, including full and partial information, and dealing with uncertainty. Students will gain practical experience writing programs that use these techniques to solve a variety of problems.

Topics include problem solving (search spaces, uninformed and informed search, games, and constraint satisfaction), principles of knowledge representation and reasoning (propositional and first-­‐order logic, logical inference, planning), and representing and reasoning with uncertainty (decision theory, reinforcement learning, Bayesian networks, probabilistic inference, basic discrete-­‐time probabilistic models).

Most of web data today consists of unstructured text. This course will cover the fundamental knowledge necessary to organize such texts, search them a meaningful way, and extract relevant information from them. This course will teach natural language processing through the design and development of end-to-end natural language understanding applications, including sentiment analysis (e.g., is this review positive or negative?), information extraction (e.g., extracting named entities and their relations from text), and question answering (retrieving exact answers to natural language questions such as "What is the capital of France" from large document collections). We will use several natural language processing toolkits, such as NLTK and Stanford's CoreNLP. The main programming language used in the course will be Python, but code written in Java or Scala will be accepted as well.  Graduate-level requirements include implementing more complex, state-of-the-art algorithms for the three proposed projects. This will require additional reading of conference papers and journal articles.

Most of the web data today consists of unstructured text. Of course, the fact that this data exists is irrelevant, unless it is made available such that users can quickly find information that is relevant for their needs. This course will cover the fundamental knowledge necessary to build such systems, such as web crawling, index construction and compression, boolean, vector-based, and probabilistic retrieval models, text classification and clustering, link analysis algorithms such as PageRank, and computational advertising. The students will also complete one programming project, in which they will construct one complex application that combines multiple algorithms into a system that solves real-world problems.  Graduate level requirements include implementing more complex, state-of-the-art algorithms for the programming project, which might require additional reading of research articles. Written assignments will have additional questions for graduate students.

Neural networks are a branch of machine learning that combines a large number of simple computational units to allow computers to learn from and generalize over complex patterns in data. Students in this course will learn how to train and optimize feed forward, convolutional, and recurrent neural networks for tasks such as text classification, image recognition, and game playing.