Machine Learning II - CS5806
Fall 2023
Instructor
Force Add requests
Please submit this form.
Class Info
- Mondays and Wednesdays
- 2:30 - 3:45 PM
- LITRV 1860
- 08/21/23 - 12/14/23
Description
This course provides an in-depth understanding of classical machine learning theory using Bayesian models, statistical machine learning and pattern recognition techniques, advanced machine learning methods, and their applications.
Learning Objectives
- Utilize mathematical tools in theoretical machine learning
- Identify classical ML methods and be able to implement them from scratch
- Formulate and implement advanced ML methods
- Analyze the ML models in terms of generalization, performance, and scalability
- Design generative and discriminative machine learning models for real problems
- Utilize ML toolboxes and platforms for model development
Course Objectives
Machine learning methods have enabled significant progress in real-world applications from text classification and machine translation to image and scene understanding, robotics, etc. Advanced methods such as Bayesian inference and probabilistic models have found increasing interest both in industry and academia, as they allow encoding domain knowledge, quantifying uncertainty, and sampling from the underlying data distribution. It is therefore essential to learn the fundamental knowledge and practical skill sets in both discriminative and generative machine learning approaches.
What to Expect
This course provides an in-depth understanding of concepts, state-of-the-art techniques and algorithms, and real-world applications of advanced machine learning. Students will gain extensive exposure to both mathematical foundations and practical development of advanced machine learning methods. The course will prepare students to tackle challenging tasks and real-world problems with different types of data (text, images, graphs, etc.) and learning tasks (supervised, unsupervised, etc.). Through this course, students will gain mathematical intuition on how each machine learning method works, form mathematical connections between different machine learning algorithms, learn how to transform practical problems into formalized learning objectives, and how to encode priors and other forms of structure into probabilistic models.
Topics
1. Overview and Learning Theory
- Maximum Likelihood (MLE) / Maximum A-Posteriori (MAP)
- Bias-Variance tradeoff
- PAC Learning, VC Dimension, etc.
- Ethics in AI
2. Classical ML Methods
- Bayesian Linear Models
- Mixture Models
- Ensemble learning: Bagging, Boosting
3. Advanced ML Methods
- Relational learning
- Relational ensemble learning
- Collective ensemble inference
- Neural networks: Multilayer neural networks, Deep neural networks
- Timeseries in Deep Learning
- Few-shot learning
- Different Types of Learning: Supervised, Unsupervised, etc.
4. Design and implementation of ML methods by using existing commonly- used ML toolboxes and platforms
Prerequisites
CS 5805 Machine Learning I. CS 5806 is taught at the 5000 level because it builds upon general programming and analytical skills found in undergraduate courses related to linear algebra, probability, statistics, algorithm design and analysis, and computer programming. It further expands upon the material presented in its prerequisite of CS 5805 - Machine Learning I.
