Introduction to Machine Learning

Virginia Tech, Electrical and Computer Engineering
Spring 2015: ECE 5984

Course Information

Have you ever wondered how Siri understands voice commands? How Netflix recommends movies to watch? How Kinnect recognizes full-body gestures? How Watson defeated Jeopardy champions? The answer is Machine Learning -- the study of algorithms that learn from large quantities of data, identify patterns and make predictions on new instances.

Instructor: Dhruv Batra
Office Hour: Fri 3-4pm, Whittemore 468

TA: Qing Sun (sunqing -at- vt)
Office Hour: Tue 5-6pm, Whittemore 415

Class meets: Mon, Wed 4:00 - 5:15, GBJ 102

Homeworks
Q&A and Discussions: https://scholar.vt.edu/portal/site/s15ece5984; ECE5984 Forum
Class Kaggle Links: http://inclass.kaggle.com/c/VT-ECE-Machine-Learning-HW1; http://inclass.kaggle.com/c/2015-Spring-vt-ece-machine-learning-hw2; http://inclass.kaggle.com/c/2015-Spring-vt-ece-machine-learning-hw3

HW Kaggle Challenge Curators: Faruk Ahmed, Aroma Mahendru,
Senthil Purushwalkam

This course introduces the basics of Machine Learning:

Supervised Learning (Classification/Regression)

Nearest Neighbour
Naive Bayes
Logistic Regression
Support Vector Machine
Neural Networks (Multi-layer Perceptrons, New this year: Convolutional Neural Networks, Recurrrent Neural Nets)
Decision Trees

Unsupervised Learning

Clustering, Kmeans
Dimensionality Reduction (Principal Component Analysis, Linear Discriminant Analysis)

Other Learning Setting and Advanced Topics

Weakly-Supervised and Semi-Supervised Learning
Reinforcement Learning
Learning Theory (VC dimensions, generalization error bounds, PAC Theory)
Structured Learning (Bayes Nets, Markov Random Fields, Structured SVMs)

Best Project Prize Winners

Winning Team:

Latha Pemula.
Low Power Image Recognition. (Contribution: Speeding up Convolutional Neural Networks by using basis filters)

Runners-up:

Kexiong (Curtis) Zeng, Jinshan Liu.
Image-based Localization System for Mobile Devices.

Venkata Pavan Kumar Rangudu, Elaheh Raisi.
Digit Recognition.

Notable Mentions

Here are some great achievements by the students who took this class:

Kexiong (Curtis) Zeng's class project accepted for publication at MobiSys.

Poster: Location Verification and Recovery for Mobile In-Vehicle Applications. Kexiong (Curtis) Zeng, Yanzhi Dou, Yaling Yang and Ranveer Chandra. In Proceedings of the 13th Annual International Conference on Mobile Systems, Applications, and Services (MobiSys), 2015.

If you took this class and have something to share, please email me!

Schedule (Tentative)

Week/Day	Date	Topic	Notes
1 W	Jan 21	Overview of Machine Learning Slides (pptx), Slides (pdf), Video	HW0 is out. Due: Jan 23. Readings: Barber Chap 1, 13.1. [Optional] Video: Sam Roweis -- What is Machine Learning?
2 M	Jan 26	Supervised Learning General Setup: learning from data Your First Classifier: 1/k Nearest Neighbour Slides (pptx), Slides (pdf), Notes	Reading: Barber Chap 14. [Optional] Video: Pedro Domingos -- The K-NN Algorithm
2 W	Jan 28	(Finish) 1/k Nearest Neighbour Kernel regression/classification Slides (pptx), Slides (pdf), Notes
3 M	Feb 2	(Finish) Kernel regression/classification Curse of Dimensionality Slides (pptx), Slides (pdf), Notes	HW1 is out. Due: Feb 15.
3 W	Feb 4	Project Details Supervised Statistical Learning Probability Review Slides (pptx), Slides (pdf), Notes	Readings: Barber 8.1, 8.2. [Optional] Videos: Probability Primer. [Optional] Video: Michael Jordon -- Bayesian or Frequentist: Which Are You?
4 M	Feb 9	Statistical Estimation (MLE, MAP, Bayesian) Slides (pptx), Slides (pdf), Notes	[Optional] Video: Daphne Koller -- Coursera: Probabilistic Graphical Models, MLE Lecture, MAP Lecture. Readings: Barber 8.6, 8.7.
4 W	Feb 11	Gaussians Linear Models Regression Slides (pptx), Slides (pdf), Notes	Readings: Barber 8.4, 17.1, 17.2.
5 M	Feb 16	VT closed due to inclement weather. No class.
5 W	Feb 18	Regression Slides (pptx), Slides (pdf), Notes	HW2 is out. Due: Mar 6. Readings: 17.1, 17.2.
6 M	Feb 23	(Finish) Regression Model Selection Error Decomposition Cross-validation Slides (pptx), Slides (pdf), Notes	Project proposal is due: Feb 24.
6 W	Feb 25	(Finish) Error Decomposition Bias-Variance Tradeoff Classification: Naive Bayes Slides (pptx), Slides (pdf),	Readings: Barber 10.1-3. [Optional] Video: Andrew Ng -- Naive Bayes
7 M	Mar 2	(Finish) Naive Bayes Classification: Logistic Regression Slides (pptx), Slides (pdf), Notes
7 W	Mar 4	(Finish) Logistic Regression NB & LR connections Slides (pptx), Slides (pdf), Notes	Readings: Barber 17.4. [Optional] Video: Andrew Ng -- Logistic Regression. Reading: Tom Mitchell -- Book Chapter: Naive Bayes and Logistic Regression
8 M	Mar 9	Spring Break: No Class.
8 W	Mar 11	Spring Break: No Class.
9 M	Mar 16	No Class. Convex Optimization video lectures.	Readings: Boyd and Vandenberghe Chap 2 (Convex Sets): 2.1, 2.2, 2.3; Chap 3 (Convex Functions): 3.1, 3.2. [Optional] Video: Stephen Boyd -- Convex Optimization.
9 W	Mar 18	In class Mid-Term
10 M	Mar 23	Support Vector Machines Linear SVMs Hard & soft-margin SVMs Slides (pptx), Slides (pdf), Notes	Readings: Barber 17.5. [Optional] Video: Andrew Ng -- KKT Conditions and SVM Duality. [Optional] Video: Stephen Boyd -- Lagrangian Duality.
10 W	Mar 25	Support Vector Machines Lagrangian Duality Slides (pptx), Slides (pdf), Notes
11 M	Mar 30	Support Vector Machines SVM dual Kernel Trick Slides (pptx), Slides (pdf), Notes	HW3 is out.
11 W	Apr 1	(Finish) Support Vector Machines Multi-class SVM Neural Networks Perceptron Multilayer Perceptron Slides (pptx), Slides (pdf), Notes	Reading: Murphy 16.5. [Optional] Reading: Hastie, Tibshirani, Friedman -- Chap 11. [Optional] Video: Andrew Ng -- Coursera: Machine Learning, Neural Networks lecture, Backpropagation lecture.
12 M	Apr 6	Neural Networks Backprop Autoencoders Convolutional Nets Slides (pptx), Slides (pdf), Notes	Reading: Murphy 16.5. [Optional] Reading: Hastie, Tibshirani, Friedman -- Chap 11. [Optional] Reading: Yann LeCun et al. -- Efficient BackProp
12 W	Apr 8	Decision Trees Mutual Information Slides (pptx), Slides (pdf), Notes	Reading: Murphy 16.1-16.2. [Optional] Reading: Hastie, Tibshirani, Friedman -- Chap 9.2. [Optional] Video: Pedro Domingos -- Coursera: Machine Learning, Decision Trees lecture,
13 M	Apr 13	(Finish) Decision Trees Ensemble Methods Bagging Slides (pptx), Slides (pdf)
13 W	Apr 15	Ensemble Methods (Finish) Bagging Boosting Learning Theory PAC-Learning Generalization bounds Slides (pptx), Slides (pdf), Notes	Reading: Murphy 16.4. [Optional] Video: Robert Schapire -- Boosting [Optional] Reading: Tom Dietterich -- Ensemble Methods in Machine Learning.
14 M	Apr 20	In-class project presentations.	HW4 is out.
14 W	Apr 22	ICCV deadline. No class.
15 M	Apr 27	Unsupervised Learning Clustering K-Means Gaussian Mixture Models Slides (pptx), Slides (pdf) Notes	Readings: Barber 20.1-20.3. [Optional] Video: Andrew Ng -- Clustering, GMM
15 W	Apr 29	EM for GMMs EM for general unsupervised learning Slides (pptx), Slides (pdf) Notes
16 M	May 4	(Finish) EM for general unsupervised learning Factor Analysis PCA SVD Dimensionality Reduction Slides (pptx), Slides (pdf), Notes	Readings: Barber 15.1-15.4. [Optional] Video: Andrew Ng -- PCA
16 W	May 6	Overview of Advanced Topics Probabilistic Graphical Models Bayes Nets Hidden Markov Models (HMMs) Online Learning Active Learning Semi-Supervised Learning Reinforcement Learning	[Optional] Reading: Pedro Domingo -- A Few Useful Things to Know about Machine Learning [Optional] Video: Andrew Ng -- Advice for Applying Machine Learning
	May 8	Project Poster+Demo Presentation	310 Kelly Hall; 1:30 - 4:00 pm
	May 11	Final Exam	In class (GBJ 102); 7:45 - 9:45 am

Grading

40% Homework
25% Project
10% Mid Term
20% Final
5% Class Participation

Late policy for deliverables

No penalties for medical reasons (bring doctor's note) or emergencies.
Every student has 5 free late days (5 x 24hour-chunks) for this course; these cannot be used for exams (midterm or final) or the poster session.
After all free late days are used up, penalty is 25% for each additional late day.

Collaboration Policy

Study groups are allowed. You can discuss the questions and potential directions for solving them. However, your discussions should be collaborative and not involve one student simply telling the other student the solutions.
If you worked in a group, please write the names of the other participants on the first page of your assignment.
Important: you need to write your solutions and code independently, and not as a group activity.
We often reuse problem set questions from various other sources. We expect students to want to learn and not google for answers. The purpose of problem sets in this class is to help you think about the material, not just give us the right answers. We take Honor Code violations seriously.

Project

Team project: 1-3 people. Description and grading policy (proposal, midsem presentation, final poster + demo presentation).

Textbooks

None required. Reference Books:

[On Library Reserve]
Machine Learning: A Probabilistic Perspective, Kevin Murphy

[Free PDF from the book webpage]
The Elements of Statistical Learning, Hastie, Tibshirani, and Friedman

[Free PDF from author's webpage]
Bayesian Reasoning and Machine Learning, David Barber

[Available in the Library]
Pattern Recognition and Machine Learning, Chris Bishop

Prerequisites

Probability and Statistics

Distributions, densities, marginalization, moments, typical distributions (Gaussian, Laplacian, etc)

Calculus and Linear Algebra

Matrix multiplication, eigenvalues, positive semi-definiteness, multivariate derivates.

Algorithms

Dynamic programming, basic data structures, complexity (NP-hardness).

Programming

Matlab for HWs. Your language of choice for project.
No coding / compilation / debugging support.

Ability to deal with abstract mathematical concepts.

Related Classes / Online Resources

10-701 Machine Learning, Carnegie Mellon University
CIS 520 Machine Learning, UPenn
CS 229 Machine Learning, Stanford
CSE 546 Machine Learning, University of Washington
Machine Learning, Coursera

Note to people outside VT

Feel free to use the slides and materials available online here.
If you use our slides, an appropriate attribution is requested.
Please email the instructor with any corrections or improvements.

Webpage CSS courtesy Bootstrap and Polo Chau.