Machine Learning ( 機器學習)
Course Introduction, 09/14/2015
Hsuan-Tien Lin (林軒田) htlin@csie.ntu.edu.tw
Department of Computer Science
& Information Engineering
National Taiwan University
(國立台灣大學資訊工程系)
Four Reasons for NOT Taking the Course (1/4) Complicated Contents
• from a Taiwanese student taking MIT ML class (translated):
The professor started writing math equations as if he was using some writing accelerator. After class I always felt feeble. The worst part is: I needed to understand the contents as soon as I can. Otherwise I cannot finish the homework and cannot follow up in the next class.
• NTU ML class: designed to beas good as the best classes in the world
• similar things will happen to you
Four Reasons for NOT Taking the Course (2/4)
Strict Instructor
• Will you give me a second chance if I copy homework from other people? NO.
• Could you let me pass because I will be kicked out by the 1/2 rule?NO.
• Will you change my score from F to C? NO.
• How many will pass? Any, if necessary.
If you do not like astrict instructor, ...
Four Reasons for NOT Taking the Course (3/4) Huge Loads
• from a student taking ML 2010 (posted on BBS):
lxxxxxx9: 作業光一小題就要我們test 100次?( 100*10min = 16hr) 唉 反 覆檢查許多遍 希望是我的code寫壞了 不然出這作業的人真的很沒良 心= =
• our class: four to six times harder than a normal one in NTU
• aroundeight homework sets (and a hard final project)
• homework due within two weeks
• even have homework 0NOW
• already hard
• though no need to submit
Four Reasons for NOT Taking the Course (4/4) Online-learnable
• invited by NTU as two of theMassive Online Open Courses on NTU-Coursera:Machine Learning Foundations and Machine Learning Techniques
• slides teaching
• Mandarin teaching
• MOOC-synced teaching
—https://class.coursera.org/ntumlone-003
• homework setting: multiple-choice problemsplus detailed arguments
• “recorded” teaching mode
• mucheasier to just learn online at home
—you canchoose to only take the online courses instead
If you do not want be restricted by physical class, ...
from a student in ML2013 (final feedback):
人活的好好的，為什麼一定要修Machine
Learning 呢？XD 這是一門體驗各種崩潰、絕望的課 程 ，人生能被課程電成這樣可能也就這麼一回
May the Brave Ones Stay
Basic Information
• instructor: Hsuan-Tien Lin (htlin@csie.ntu.edu.tw)
• office hour: after class or by appointment
• course webpage (CEIBA):
https://ceiba.ntu.edu.tw/1041mlearn announcements, homework, reference handouts, etc.
• mailing list: supported by CEIBA
• discussion forum: supported byML-Foundations on Coursera
—https://class.coursera.org/ntumlone-003
• new: four-hour credits instead of three-hour: as suggested by previous students
update your secondary email address on CEIBA and register for ML-Foundations
Enrollment
• With school policy, students who have taken the three-hour version is NOT allowed to take this class again!
• at most 176 in Room 103 of CSIE Building
• target size 220, which is 125% of 176—assuming that 1/5 will not show up in class
• priority: CSIE (CSIE, INM) > EECS > others
• auditing: welcomed (to sit) only if there is an empty chair
Leave as soon as possible! Give your classmates a chance to be miserable.
Teaching Assistants
Kuan-Hao Huang, You-Lin Tsou, Hong-Min Chu, Hsien-Chun Chiu, Liang-Wei Chen, Yu-An Chung, Meng-Yuan Yang, Yao-Yuan Yang,
Si-An Chen
• forum for course/homework material questions: on Coursera
• email for grading questionsonly: ml2015ta@csie.ntu.edu.tw
• TA Hour for more interactive discussions: to be announced
To save TA loads, questions about course/homework materials will only be answered on the forum and/or TA hours
after the forum starts running.
THE Book
Learning from Data: A Short Course
Y. Abu-Mostafa (Caltech), M. Magdon-Ismail (RPI), H.-T. Lin (NTU)
• idea initiated during 2008
• 5 chapters, closely needed for the first half of the class
• other e-Chapters to be used in the second half of the class
• teaching with the book and suggested reading within the book
Getting the Book to Read
• NTU Library: one reserved copy in the shared course material area
• R536: some shared copies to be read in the room
• Chuan-Hwa Book Company: imported some limited copies of the book
— Ms. Jen Huang (jen@chwa.com.tw) at 0958-008-962
— may or may not offer group discounts
• Amazon: main selling channel in the US, but can be expensive/slow for international shipping
— http://www.amazon.com/gp/product/1600490069
• Bulk order from U.S.: secondary selling channel, usually takes two weeks to arrive — http://amlbook.com
• If the book is not affordable to you but you really want to read it: email me (htlin@csie.ntu.edu.tw) and I’ll see how I can help.
Getting Future Draft Chapters to Read
mechanism: to be announced when needed Your Privileges
• learn from thefirst draft
• obtain the draftfreely Your Responsibilities
• discussactively with me to improve the draft
• do not distribute the draft
enrolling in this class means agreeing to the items above
THE Principle
Taking any unfair advantages over other class members is not allowed. It is everyone’s responsibility to maximize the level of fairness.
• eating? fine, but no smells and no noise
• sleeping? fine, but no snoring
• cellphone? fine, but silent mode, and speak outside
• ...
applies to instructor, TAs, students
Honesty
NO CHEATING NO LYING NO PLAGIARISM
NO PIRATING of THE BOOK
very serious consequences
Grade
• no midterm, no final
• main reference: homework sets, final project
• raw score goes through some order-preserving normalization steps
• raw score 80 with term rank A: possible
• raw score 80 with term rank B: possible
• raw score 60 with term rank F: possible
• raw scores 80, 60 with term scores B, B: possible, but unlikely
• raw scores 80, 60 with term scores F, B:impossible
Collaboration and Open-Book
• homework discussions: encouraged
• but fairness?
write the final solutions alone and understand them fully
• references (books, notes, Internet):
consulted, butnot copied from
• no need to lend/borrow solutions
to maximize fairness (everyone’s responsibility), lending/borrowing not allowed
Collaboration and Open-Book
to maximize fairness (everyone’s responsibility), lending/borrowing not allowed to maximize fairness (everyone’s responsibility),
lending/borrowing not allowed to maximize fairness (everyone’s responsibility),
lending/borrowing not allowed
Deal? If your classmate wants to borrow homework from you, what do you say?
Homework
• students: justify solutions clearly
• TAs: evaluate solutions fairly
• penalty for late parts:
90% of value for 12-hour late, 80% one-day late, ...
late homework should go to a box in R217
Programming Assignments
• about a third or half of the problems
• any programming language, any platforms
• uploadsource code, otherwise:
10% of value only!
• no sophisticated packages
students’ responsibility:
ask TA in advance for what can/cannot be used
Important TODOs
• Update your secondary email address on CEIBA!
• Register ML-Foundations on Coursera!
• Do homework 0; send emails to TAs or post on Coursera forum for questions.
May the Brave Ones Stay
