EECS 448: Applied Machine Learning for Modeling Human Behavior
Everywhere we look, machine learning is uncovering new ways of sensing and modeling human behavior. But, how does this work? Does this even work? The course will cover current practices in measuring and sensing human behavior via machine learning.
Enforced Prerequisites: Students should have taken EECS 281 and (MATH 214 or MATH 217 or MATH 296 or MATH 417) or graduate standing.
Machine learning, with a focus on human behavior, across multiple modalities including speech and text. Teams complete projects based primarily on their individual interests centered on modeling an aspect of human behavior. Prior experience with speech/language or other data modeling is not needed.
- Understand the value of human behavior computing in industry and research.
- Develop an understanding for the common signals used to measure behavior (speech, text, multimodal).
- Learn machine learning methods in affective computing.
- Demonstrate an understanding of the concepts by building systems that sense and interpret human behavior.
- Demonstrate an understanding of the limitations of the technologies critically interpreting the newest advances in human-centered technologies.
- Understand how python tools can be used in these domains.
The evaluation of this course will include four homeworks, a midterm, and a semester-long project.
Lecture-by-Lecture (Winter 2022, subject to change!)
- Linear Regression and Research Spotlight
- Logistic regression and research Spotlight
- Probability prep
- Gaussian Mixture Models and research spotlight
- Hidden Markov Models 1
- Hidden Markov Models 2
- Hidden Markov Models 3
- Hidden Markov Models 4
- Hidden Markov Models Applications
- Data collection in the real world
- Midterm ← date still to be finalized (approximate only)
- Hyperparameter optimization, regularization, feature selection
- Neural Networks and human behavior
- Project Milestone Presentation Day
- Neural Networks and human attention
- Convolutional Neural Networks (CNN) and facial emotion recognition
- Recurrent Neural Networks (RNN) and age detection
- Multimodal Fusion
- Autoencoders and Feature Learning
- Representation Learning and Regularization
- Transformers in Affective Computing
- Anonymization in Human Behavioral Data
- Project presentations
- Project presentations