CSE517a Machine Learning

Spring 2020 entry exam will be assigned for one week and will available before January 7th 2020 from this website (find links below).

This course assumes a profound understanding of the fundamentals of machine learning (including the theoretical foundations and principles of ML as well as hands-on implementation experience). CSE517a covers advanced topics at the frontier of the field in-depth. Topics to be covered include kernel methods (support vector machines, Gaussian processes), neural networks (deep learning), and unsupervised learning. Depending on developments in the field, the course will also cover some advanced topics, which may include learning from structured data, active learning, and practical machine learning (feature selection, dimensionality reduction, model evaluation and comparison).

Prerequisites: CSE 247, CSE 417T, ESE 326, Math 233, Math 309, and profound experience in Matlab/Octave or Python (numpy/scipy). Please, scroll down to the bottom of this page for more information.

This class counts towards the Certificate in Data Mining and Machine Learning as required course.

Interested in ML outside the classroom? Check out the Physics Machine Learning Club or my student projects page!

IMPORTANT: CSE517A has CSE417T as an enforced prerequisite

  • CSE514A, CSE511A, or CSE515T cannot replace CSE417T.
  • Given that we assume that students in CSE417T spend 10hrs a week to study the conceptual materials and work on the class assignments, it is not possible to acquire the same level of knowledge through self-study. Note that online classes like the Coursera course Introduction to ML do not provide the same level of depth as CSE417T.
  • If you have taken a machine learning course from another university covering the content of CSE417T, you should be familiar with the topics listed below. If that is the case you can be conditionally enrolled in the course. You will have to pass the placement exam in order to take CSE517A.

PLACEMENT EXAM

There will be an entry exam given to everyone conditionally enrolled or waitlisted who does not meet the CSE417T prerequisite. The exam and further instructions will be available in the first week of January 2020.

  • Download the take-home entry exam HERE.
  • Download the stub files HERE.
  • Download the data HERE.
  • You have one week to complete this exam. It is due on the first day of class (TUE Jan 14 2020) at 10am.
  • If you do not pass the entry test, you will be unenrolled or removed from the waitlist.

In addition to this placement exam you should get a copy of the CSE417T course book Learning from Data (LFD) by Abu-Mostafa, Magdon-Ismail, and Lin and makeup all CSE417T homework assignments before starting CSE517a.

We will not be able to provide help/support for conceptual questions on course materials from prerequisite courses nor general SVN and MATLAB/Python (basic programming) issues. 

Pre-requisite knowledge and content from CSE417T

– supervised learning setup
– training, testing, validation, generalization
– training error, testing error, generalization error
– loss functions for regression, classification
– perceptron algorithm (analysis and implementation in MATLAB/Python)
– linear regression (least squares model)
– linear classification (logistic regression)
– gradient descent
– non-linear feature space transformation
– (hyper)parameter selection, model selection
– cross validation
– regularization, structural risk minimization
– bias-variance decomposition of the error
– parametric vs non-parametric models
– multi-class classification
– k-NN model (2-optimal, implementation in MATLAB/Python)
– KD-trees, Ball-trees
– decision trees: training, pruning, and prediction (analysis and implementation in MATLAB/Python)
– bagging, random forests (analysis and implementation in MATLAB/Python)
– boosting, Adaboost (theoretical analysis and implementation in MATLAB/Python)
– support vector machines (primal and dual optimization, slack variables, kernel SVM)
– neural networks (back propagation algorithm)

Spring 2020

All course information and contents will be available on Canvas.