Introduction to Machine Learning Jia-Bin Huang Virginia Tech Spring 2019 ECE-5424G / CS-5824
Today’s class • Introduction • A little about us • A little about you • Machine learning • What is machine learning? • Types of machine learning • Example applications • Course logistics
About me • Born and raised in Taiwan
National Chiao-Tung University Microsoft Research Disney Research UIUC B.S. in EE Research Intern Research Intern Ph.D. in ECE 2016
National Chiao-Tung University Microsoft Research Disney Research UIUC B.S. in EE Research Intern Research Intern Ph.D. in ECE 2016
Im Image Completion [S [SIGGRAPH14] - Revealing unseen pixels
Vid ideo Completion [S [SIGGRAPH Asia ia16] - Revealing temporally coherent pixels
Facebook F8 Keynote Talk 2017 Adobe Max 2017
Im Image super-resolution [C [CVPR15] - Revealing unseen high frequency details
Object tracking [ICCV15] Detecting migrating birds [CVPR16] Vis isual Tracking - Locatin ing movin ing objects across vid ideo frames Multi-face tracking [ECCV16]
Weakly supervised localization [CVPR16] Unsupervised feature learning [ICCV17] Learning with weak labels
Teaching Assistant: Chen Gao • 1 st year PhD student in ECE, VT • Email: chengao@vt.edu • Web: https://gaochen315.github.io/ • Office hour: • TBD • Research:
Teaching Assistant: Shih-Yang Su • 1 st year PhD student in ECE, VT • Email: chengao@vt.edu • Web: https://lemonatsu.github.io/ • Office hour: • TBD • Research:
A little about you • Find two persons near you • Introduce yourself • Name? • Department? • Why taking this class? • One interesting fact? • Introduce your neighbors to the class!
What this course is about? Learning to Teach Machine to Learn
Let’s chat! • What is machine learning? • What applications? Discuss with your neighbor
What is machine learning? • Field of study that gives computers the ability to learn without being explicitly programmed Arthur Samuel (1959)
What is machine learning? • A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P if its performance at tasks in T , as measured by P , improves with experience E . Tom Mitchell (1998)
A computer program is said to learn from experie ience E with respect to some class of tasks T and performance measure P if its performance at tasks in T , as measured by P , improves with experience E . Designing a spam filter o Classifying emails as spam or not spam o Watching you label emails as spam or not spam o The number (or fraction) of emails correctly classified as spam/not spam Slide credit: Andrew Ng
A computer program is said to learn from experie ience E with respect to some class of tasks T and performance measure P if its performance at tasks in T , as measured by P , improves with experience E . Designing a spam filter o Classifying emails as spam or not spam Tasks T o Watching you label emails as spam or not spam Experience E o The number (or fraction) of emails correctly classified as spam/not spam Performance measure P Slide credit: Andrew Ng
Types of machine learning algorithms • Supervised learning • Training data includes desired outputs • Unsupervised learning • Training data does not include desired outputs • Weakly or Semi-supervised learning • Training data includes a few desired outputs • Reinforcement learning • Rewards from sequence of actions Slide credit: Dhruv Batra
Machine learning algorithms Supervised Unsupervised Learning Learning Discrete Classification Clustering Dimensionality Continuous Regression reduction
Machine learning algorithms Supervised Unsupervised Learning Learning Discrete Classification Clustering Dimensionality Continuous Regression reduction
Breast cancer (malignant, benign) Classification problem 1 (Yes) Discrete valued output Malignant? e.g., 0 or 1 0 (No) Tumor Size Multi-class classification e.g., 0 or 1 or 2 or 3 Tumor Size Slide credit: Andrew Ng
Multiple features • Clump thickness • Uniformity of cell size • Uniformity of cell shape ? • … Age Tumor Size Slide credit: Andrew Ng
Image classification
Spotting eye disease • Recognize 50 sight-threatening eye diseases • As accurately as world- leading expert doctors https://www.youtube.com/watch?v=MCI0xEGvHx8 Clinically applicable deep learning for diagnosis and referral in retinal disease, Nature Medicine, 2018
Face recognition Facebook auto-tagging
Machine Translation https://www.youtube.com/watch?v=WeByuOD8k1c
Speech Recognition Slide Credit: Carlos Guestrin
Speech recognition http://youtu.be/Nu-nlQqFCKg?t=7m30s
Predicting aftershock patterns Credit: Aflo/REX/Shutterstock Deep learning of aftershock patterns following large earthquakes, Nature, 2018
Machine learning algorithms Supervised Unsupervised Learning Learning Discrete Classification Clustering Dimensionality Continuous Regression reduction
Housing price prediction Price ($) Regression problem in 1000’s Continuous valued 400 output (price) 300 200 100 2000 500 1000 1500 2500 Size in feet^2 Slide credit: Andrew Ng
Stock market Slide credit: Dhruv Batra
Weather prediction Temperature Slide credit: Carlos Guestrin
Human pose estimation DensePose, CVPR 2018
Facial landmark alignment Snapchat filter https://www.youtube.com/watch?v=Pc2aJxnmzh0
Machine learning algorithms Supervised Unsupervised Learning Learning Discrete Classification Clustering Dimensionality Continuous Regression reduction
Supervised Learning Unsupervised Learning 𝑦 2 𝑦 2 𝑦 1 𝑦 1
Google news
Clustering DNA microarray data build groups of genes with related expression patterns (also known as coexpressed genes) Source: Su-In Lee et al.
Slide credit: Andrew Ng
Machine learning algorithms Supervised Unsupervised Learning Learning Discrete Classification Clustering Dimensionality Continuous Regression reduction
Dimensionality reduction 𝑦 2 𝑦 1
3D face modeling A morphable model for the synthesis of 3D faces, SIGGRAPH 1999
Shape modeling SMPL: Skinned multi-person linear model, SIGGRAPH Asia 2015
Cocktail party problem
Source: https://hbr.org/2016/11/the-competitive-landscape-for-machine-intelligence
Course Overview
General information • Course title: Advanced Machine Learning • Not really… this is an introductory machine learning course • ECE-5424 / CS-5824 • Mon and Wed 2:30 PM – 3:45 PM • Surge Space Building 118C • Office hours - Jia-Bin • Mon 3:45 – 4:45 PM • Office hours - Chen, Shih-Yang • TBD. Survey on Piazza/Canvas
Useful links • Course webpage: http://bit.ly/vt-machine-learning-spring-2019 • Download lecture slides • Piazza discussion forum: https://piazza.com/class/jr6vbmqyvwy3wk • All communications go through piazza. No emails please. • HW submission: https://canvas.vt.edu/ • Start early! • Anonymous course feedback: https://goo.gl/forms/nSz66NogxKXnXLBD2
Textbooks (optional)
Course work • Homework assignments (50%) • Six main homework assignments + HW0 • Late policy: Up to six free late days. After that, a penalty of 10% per day. • Midterm exam (10%) Grading [0-60] F, [60-62] D-, [63-66] D, • Final exam (15%) [67-69] D+, [70-72] C-, [73-76] C, [77-79] C+, [80-82] B-, [83-86] B, [87-89] B+, [90-92] A-, [93-100] A • Final project (25%) • Proposal, project report, and spotlight video • Work in a team of 2-3 students
Request • Homework extension request • Only for medical/family emergency (please send me email with doctor’s note) • No “I have an interview this week”, “I have a midterm exam”, “I am busy recently.” • Homework regrade request • One week after the grade release date • Final grading change request • No “I need to get an B+ to graduate”, “Can I can a grade upgrade?”
Academic Integrity • Can discuss HW with peers, but cannot copy and/or share code • Carefully document any sources within HW hand-in • Do not use code from Internet unless you have permission • If you’re not sure, ask • Do not use your published work as your final project • Plagiarism. Zero tolerance. We are required to report it to the university.
Course enrollment • Classroom capacity 140 • (70 ECE session + 70 CS session) • A long waiting list • Drop the class if you are not able to commit your time • Policy: no force-add students to a full class. • Sit in • Please leave room for students who registered the class
Prerequisites • Linear algebra, basic calculus • Review: http://cs229.stanford.edu/section/cs229-linalg.pdf • Probability and statistics • Review: https://see.stanford.edu/materials/aimlcs229/cs229-prob.pdf • Python (NumPy) • http://web.stanford.edu/class/cs224n/readings/python-review.pdf • Review: Python review session by TAs
Recommend
More recommend
