CS/ECE/STAT-861: Theoretical Foundations of Machine Learning
University of Wisconsin-Madison, Fall 2024
Grading overview
Auditing: Students auditing the class need to complete only the scribing portion.
Scribing
Each student will be required to prepare scribed notes for 2 lectures. There will be
around 2 scribes per lecture.
(I may need to change this if the current enrollment falls significantly.)
Scribes should take careful notes in class, and prepare a LaTeX document written in full
prose, understandable to a student who may have missed class.
Scribes should include all steps of the proof, any intuitive explanations, and thoroughly check for typos before submitting.
The latex document and source files should be submitted within two days
of the lecture.
After review, the scribed notes will be posted to the course website.
The LaTeX source files from last year's course are
available in Canvas via
this link. You are encouraged to build on these notes
rather than starting from scratch.
The quality of the scribe notes from last year will be varied.
Pay special attention to areas that are not adequately explained,
mistakes or typos, and
any material that may have changed from last year.
You will
be evaluated on the quality of your submission.
Instructions:
Please use this LaTex style file and
this template for scribing lecture notes.
While not mandatory, I strongly recommended that you use Overleaf to
prepare the notes.
After completing the notes, please invite me (kandasamy@cs{dot}wisc{dot}edu) as a collaborator
on your overleaf project, and send me an email indicating that you have done so.
The scribed lecture notes should be submitted within 2 days of the lecture.
You may sign up in
this spreadsheet either as Scribe 1 or Scribe 2.
If you decide to drop the class before your scribe date, please delete your name from the scribe
slot and email me. If you have enrolled, but are not sure about taking the class,
please sign up for later classes (after Oct 6).
Homeworks
There will be 4 problem sets in addition to homework 0.
Homeworks will be posted on Canvas and will be due by 11.59pm on the due date.
All homeworks will not necessarily have equal credit. Longer/harder homeworks will
be weighted more towards your final homework score.
Homeworks should be typeset using some appropriate software. We will not be accepting written
and scanned homeworks.
LaTeX is a free and excellent system for typesetting.
Students will receive 5 percent extra credit if their solutions are typeset fully using LaTex.
As we will re-use questions in future offerings of this class,
please do not release homework or exam questions outside of class or discuss them in public
forums.
Collaboration: The problem sets will generally be difficult.
Unless explicitly stated otherwise,
students are allowed (and encouraged) to collaborate in groups of size up to three.
However, each student must write up solutions individually and name any collaborators at
the top of each problem.
Please read the university's policy on academic misconduct.
Solutions will be posted within a week of the deadline.
Students will receive zero credit for submissions after solutions are posted.
Late submissions:
You will be allowed three (3) total late days without penalty for the entire semester.
You may be late by 1 day on three different homeworks, late by 2 days on one homework and
by 1 day on another, or by 3 days on a single homework.
Once those days are used, you will be penalized according to the following policy:
Homework is worth full credit if submitted by the deadline.
It is worth half credit for the next 48 hours.
It is worth zero credit after that.
You must turn in all of the homeworks, even if for zero credit, in order to pass the course.
Any submission after the deadline will be counted as a late day, even if it is just a few seconds.
Please plan to submit your homework well ahead of the deadline.
Extensions are extremely unlikely, and will be considered only for documented emergencies.
Take-home exam
The exam will be held from Saturday 11/16/2024 12.01 AM – Sunday 11/24/2024 11.59 PM.
The length of the exam is 48 hours. Students can start the exam at a convenient time within this
period, but must submit their solution by 11.59 PM on 11/24 and within 48 hours of the start time.
You may refer to any of your notes or the scribed lecture notes during the exam.
However, unlike the homeworks, you are not allowed to refer to any other material or collaborate/discuss the questions with
other students in class.
Please read the university's policy on academic misconduct.
Students will receive 5 percent extra credit if their solutions are typeset in LaTex.
If you choose to submit a handwritten exam, please make sure your handwriting is legible. We will
not spend undue effort trying to understand poor handwriting.
Course project
For the course project, you will work in groups of size up to three,
to design a homework question.
You will then have the opportunity to attempt (and evaluate) the problems designed by
your peers.
Each group should submit your problem, along with the solutions, by Friday 11/15/2024.
I will then randomly assign 2 of the submitted questions to each student. The solutions to these
problems will be due by Saturday 12/7/2024.
Along with the solutions, you will also submit an evaluation of the problem that you attempted.
I will take this feedback into consideration when determining the grade for each problem.
Guidelines.
A good homework problem should reinforce class concepts and appropriately challenge your peers. Here are some key guidelines:
Ensure the problem has a coherent focus rather than a collection of unrelated questions.
The problem should allow students to apply concepts learned in class, and not simply recall
information.
Ensure the problem statement is unambiguous, with clear and precise instructions on
what is expected. Clearly state any assumptions or constraints.
The problem should be challenging enough to stimulate critical/creative thinking, but
not overly difficult that it becomes frustrating. Consider including parts that
progressively build on each other.
Ideally, the problem will teach the student (and the instructor!) something new.
You are encouraged to design questions that explores new topics, provided
that you provide necessary background knowledge/information.
If you develop an outstanding homework problem, I will consider including it in future iterations of
this course and will acknowledge your contribution.
You may model the level of difficulty and depth based on the following homework problems (more to be added in the future):
Homework 1, Problem 4 (Sauer's lemma for interval classifiers)
Homework 2, Problem 4 (Upper and lower bounds for density estimation)
Homework 3, Problem 1 (Lower bounds for prediction problems)
Homework 3, Problem 2 (Upper/lower bounds for explore-then-commit)
Homework 4, Problem 4 (Learning in a two-player zero-sum game)
|