Lecture 00: Introduction - cs.uwaterloo.cay328yu/mycourses/480/lectures/lec00.pdf · Textbook •No...
Transcript of Lecture 00: Introduction - cs.uwaterloo.cay328yu/mycourses/480/lectures/lec00.pdf · Textbook •No...
-
CS480/680: Intro to MLLecture 00: Introduction
9/5/2019 Yao-Liang Yu1
-
Outl ine
• Course Logistics
• Course Overview
9/5/2019 Yao-Liang Yu2
-
Outl ine
• Course Logistics
• Course Overview
9/5/2019 Yao-Liang Yu3
-
Course Info
• Instructor: Yao-Liang Yu ([email protected])• Office hours: DC 3617, Th 11:30 – 12:30
• TAs: Amir Farrag, Jingjing Wang, Theo Hu
• Website: cs.uwaterloo.ca/~y328yu/mycourses/480
• slides, notes, policy, etc.
• Piazza: piazza.com/uwaterloo.ca/fall2019/cs480cs680/
• Announcements, questions, discussions, etc.
• Learn: https://learn.uwaterloo.ca• Assignments, solutions, grades, etc.
9/5/2019 Yao-Liang Yu4
https://piazza.com/uwaterloo.ca/fall2019/cs480cs680/homehttps://piazza.com/uwaterloo.ca/fall2017/cs489cs698/homehttps://learn.uwaterloo.ca/
-
Prerequisi tes
• CS341, STAT 206 or 231 or 241
• Basic probability, statistics, algorithms, linear algebra, calculus
9/5/2019 Yao-Liang Yu5
• Mathematical maturity
• Programming
https://julialang.org/https://www.python.org/
https://julialang.org/https://julialang.org/https://www.python.org/
-
What to expect
9/5/2019 Yao-Liang Yu6
-
Textbook
• No required textbook
• Lecture notes or slides will be posted on course web
• Some fine textbooks:
9/5/2019 Yao-Liang Yu7
-
Assignments
• 4 assignments with the following tentative plan:
• Submit on LEARN. Submit early and often.• Typeset using LaTeX is recommended
9/5/2019 Yao-Liang Yu8
Out date Due date CS480 CS680
A1 Sep. 05 Sep. 26 15% 15%
A2 Sep. 26 Oct. 22 15% 15%
A3 Oct. 22 Nov. 12 15% 15%
A4 Nov. 12 Dec. 03 15% 15%
Proposal Oct. 31 5% 5%
Report Dec. 15 20% 20%
https://en.wikipedia.org/wiki/LaTeX
-
Exam
• Midterm: 15%, Oct 22, in class
• Final exam: 25%, date TBA• Open book
• No electronics
9/5/2019 Yao-Liang Yu9
-
Policy
• Do your own work independently and individually• Discussion is fine, but no sharing of text or code
• Explicitly acknowledge any source that helped you
• Ignorance is no excuse!• Good online discussion, more on course website
• Serious offence will result in expulsion…
• NO late submissions!• Except hospitalization, family urgency, …
• Appeal within two weeks, otherwise final
9/5/2019 Yao-Liang Yu10
http://www.math.uwaterloo.ca/navigation/Current/cheating_policy.shtmlhttps://cs.uwaterloo.ca/~y328yu/mycourses/489/policy.html
-
Project
• Optional: can trade for final. Need TA / Instr. approval.
• You project should• Relate to machine learning (obviously)
• Allow you to learn something new (and hopefully significant)
• Be interesting and nontrivial (publishable)
• 2-4 pages proposal and
-
Enrol lment
• If you already enrolled• Good for you!
• If you are not enrolled yet• Permission numbers will be sent early next week
9/5/2019 Yao-Liang Yu12
-
Questions?
9/5/2019 Yao-Liang Yu13
-
Outl ine
• Course Logistics
• Course Overview
9/5/2019 Yao-Liang Yu14
-
What is Machine Learning (ML)?
• “Machine learning is the field of study that gives computers the ability to learn without being explicitly programmed.” --- Arthur Samuel (1959).
9/5/2019 Yao-Liang Yu15
• “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)
https://en.wikipedia.org/wiki/Arthur_Samuelhttp://www.cs.cmu.edu/~tom/
-
Why is ML important for YOU?
• First off, you use ML everyday
9/5/2019 Yao-Liang Yu16
• Lots of cool applications
• Excellent for job-hunting
-
Learning Categories
9/5/2019 Yao-Liang Yu17
Supervised
• Classification
• Regression
• Ranking
Reinforcement
• Control
• Pricing
• Gaming
Unsupervised
• Clustering
• Visualization
• Representation
Teacher provides answer Teacher provides motivation Surprise, surprise
-
Supervised learning
9/5/2019 Yao-Liang Yu18
-
Formally
• Given a training set of pairs of examples (xi, yi)
• Return a function f: X Y
• On an unseen test example x, output f(x)
• The goal is to do well on unseen test data• Usually do not care about performance on training data
9/5/2019 Yao-Liang Yu19
-
Reinforcement learning
• Not in this course…
9/5/2019 Yao-Liang Yu20
• CS486/686
Silver et al., Nature’16Abbeel et al., NIPS’06
-
Unsupervised learning
• Let the data speak for itself!
9/5/2019 Yao-Liang Yu21
Le et al., ICML’12
-
Focus of ML research
• Representation and Interpretation• How to represent the data? How to interpret result?
• Generalization• How well can we do on test data? On a different domain?
• Complexity• How much time and space?
• Efficiency• How many samples?
• Applications
9/5/2019 Yao-Liang Yu22
-
Course Overview
9/5/2019 Yao-Liang Yu23
-
Classic
9/5/2019 Yao-Liang Yu24
-
Neural Nets
9/5/2019 Yao-Liang Yu25
-
Generative Models
9/5/2019 Yao-Liang Yu26
-
Exotic
9/5/2019 Yao-Liang Yu27
-
Questions?
9/5/2019 Yao-Liang Yu28