Course information for COMS 4772 Advanced Machine Learning (Fall 2013)
Basic information
- When: Wednesday 4:10–6:00 PM
- Where: 545 Mudd
- Instructor: Daniel Hsu
Course description
- This course is about probabilistic and linear algebraic tools for unsupervised machine learning.
- We will study various methods used by machine learning practitioners (e.g., hashing, projection pursuit, PCA, spectral clustering) which usually fall out of the purview of supervised learning.
- The goal is to develop tools for both (i) analyzing unsupervised learning methods and (ii) developing new methods.
Prerequisites
- Algorithms, calculus, linear algebra, and probability.
- This is a theory course, so mathematical maturity is essential.
- It is okay to take COMS 4771 concurrently with COMS 4772 this term.
- Please contact me if you have concerns about the prerequisites.
Tentative topics
The actual set of topics covered in the course will have a non-trivial intersection with the following list.
- Basic probability: Gaussian distributions, Chernoff bounds, random vectors
- Random linear embeddings: Johnson-Lindenstrauss, hashing trick, subspace embeddings, compressed sensing, projection pursuit, locality-sensitive hashing
- Spectral analysis: singular value decomposition, random matrices, covariance estimation, low-rank approximation, canonical correlation analysis
- Quantization and mixture models: optimization formulations, planted partition models, Gaussian mixture model
Course work
Homework assignments (50%) and final project (50%).