Pattern Recognition and Analysis
Course facts
- Payment: Free, no certificate
- Credits: Non Credit
- Learning style: Self-paced, start anytime
- Resources: Problem Sets; Programming Assignments; Projects
MIT OpenCourseWare details
- As taught in: Fall 2006
- Department: Media Arts and Sciences, Civil and Environmental Engineering
| Course Details | ||||||
|---|---|---|---|---|---|---|
|
||||||
| Outcomes | ||||||
|
||||||
| Price & Terms | ||||||
|
||||||
| Technical | ||||||
|
||||||
| Specification | ||||||
|
||||||
Foundations of pattern recognition: decision theory, statistical classification, clustering, hidden Markov models and Bayesian networks, with data sets.
Finding patterns and features of interest in numerical data is the subject of this class, with applications to user modeling, affect recognition, speech, computer vision and physiological analysis. It teaches decision theory, statistical classification, maximum likelihood and Bayesian estimation, nonparametric methods and clustering. Later topics reach hidden Markov models, Bayesian networks and reinforcement learning, with six data-driven problem sets.
What You Will Achieve
- Apply decision theory and ROC curves to classification
- Estimate parameters with maximum likelihood and Bayesian methods
- Use clustering and expectation-maximization on unlabeled data
- Build and evaluate a pattern recognition project on real data
Program Overview
Probability and Decision Theory
The first topics review probability and set up decisions.
- Probability and Bayes rule
- Linear transformations
- Decision theory and ROC curves
- Fisher discriminant
Learning Methods
Training methods and classifiers take up the middle.
- Maximum likelihood and Bayesian estimation
- Support vector machines
- K-nearest-neighbor classification
- Density estimation
Mixture Models and Bayesian Networks
The last topics treat more complex models.
- Clustering and mixture modeling
- Hidden Markov models
- Kalman filtering
- Bayesian networks and reinforcement learning
Course Completion
Job Outlook
These methods are used in machine learning, speech and vision systems, and in analysis of physiological and behavioral data.
Good for: Students interested in signal understanding problems who want a broad grounding in classic pattern recognition.
Keep in mind: Problem sets and data sets are published, but no lecture notes or solutions are listed. Treat it as foundations, since its topic list stops at classic methods.
Bottom line: A grounding in classic pattern recognition, with data sets to work on.
Frequently Asked Questions
Are data sets provided?
Yes. Each of the six problem sets comes with its data set files, including separate training and test data for one problem in Problem Set 4.
What did students do for projects?
Sample projects include intrinsic rewards in reinforcement learning, non-parametric classification of facial features, and speech act classification of video game data.
Which sequence models are covered?
Hidden Markov models with the Viterbi and Baum-Welch algorithms, and linear dynamical systems with Kalman filtering.
User Reviews
Write your review
You must be logged in to post a review.
Reviews can only be posted by registered learners. Log in or create a free account, then you will return here to post.
There are no reviews yet.