| Category | Difficulty |
|---|---|
| HW | 4 |
| Quiz | 4 |
The course is taught by Professor Bhiksha. It can be considered as the ML engineer's toolbox course; you will use the ideas everywhere, regardless of whether you're doing your phd, or working in a company. It was a famous course before neural networks era, and had some of its course projects become best paper awards and PhD theses (in ECE and Civil Engg!) and others have been used to spin off companies.
The course emphasis is on how “Every information is a signal". Speech, text, images. And the homeworks spans all these aspects of Signal mentioned above. As you have probably seen, it is titled "machine learning for ..." this means ML concepts only will be covered. You will learn foundational ML concepts and algorithm behind state of the act applications along with a high level of Algebra and probabilities (I cannot stress this enough). The quizzes and homework stresses this areas a lot. You will see Algebra and probabilities in various forms.
Good news: It isn't as time challenging 11-785 which is more implementation focused. However, this course would be conceptually intensive.
Some of the main topics covered are
- Fundamentals of Linear Algebra
- Eigen decomposition, SVD
- Wavelets, Fourier Transform, Cosine Transforms
- Basics of convex optimization, constrained optimsiation, Lagrange multipliers
- Eigen representations, Karhunen-Loeve, PCA
- Boosting and its application to Face detection
- Cascade Classifier
- ICA for representations and denoising
- Information Theoretical based Methods
- Types of NMF, overcomplete representations and sparsity
- Kernel K-means and Mercer's condition
- Dictionary based representations
- Nearest neighbors, Linear regression, Tikhonov and L1 regularization
- SVM
- Expectation Maximization and GMM
- CCA, LDA
- Hidden Markov Models
- Linear Gaussian Models and Extended Kalman filters
There are 4 HWs, a Final project and Quiz's every week. The Final project is generally for team of 4 and Deep Learning methods are not allowed (though this might change in future).
Attend classes regularly. Assignments will be conceptual based and have potential areas were you might get stuck, so plan for TA hours accordingly.
Regarding the project, there will be an initial project proposal which is due in the first month. There is a midterm review and a final poster presentation. The project encourages students to think out of the box and solve challenging problems.
Tip: Keep all your late days for HW4, as it's a challenging HW relatively much difficult than first 3.
You can check out the previous year's material here: http://mlsp.cs.cmu.edu/courses/fall2019/