A conceptual and practical introduction to modern machine learning:

- machine learning as an engineering approach, how it relates to other approaches to solving problems or "artificial intelligence", including a historical overview of machine learning, and how machine learning relates to computational complexity and cryptography.  

- core concepts in supervised learning, form a conceptual, theoretical and practical perspective: approximation, estimation, empirical and structural risk minimization, complexity control, and the relationship to parameter estimation, probabilistic and Bayesian approaches,   The role of statistics, information and computation in machine learning.

- methods and models for supervised learning: Decision Trees, Linear Prediction, Support Vector Machines (SVMs) and Kernel methods, L1-methods, Boosting*, Nearest Neighbor methods, Differentiable Learning, Feedforward Neural Networks (The Multilayer Perceptron), Modern Deep Learning Architectures*

- supervised learning in practice: training, testing, validation and cross-validation, significance and working with data.

 - beyond supervised learning*: transfer and semi-supervised learning, active learning, exploration and reinforcement learning.

*topics mentioned but not covered in depth.

 

Meets: Mondays and Wednesdays 3-4:20PM in TTIC 530.

Recitations: Fridays 3-4:20PM and Thursday 11AM-12:20PM (two identical recitations each week, need only attend one).

In accordance with the University's calendar adjustment, the first class will be on Monday, January 10th, with the first two weeks available online.

 

 

Instructor: Nati Srebro. Office hours Monday and Wednesday 4:30-5pm at TTIC 503.

TAs:

- Anmol Kabra (office hours Wednesdays 1:45-2:45pm at TTIC 428)

- Kavya Ravichandran (office hours Thursday 2:30 PM - 4 PM at TTIC 428 -- this is the open area on the 4th floor near the foot of the atrium stairs.)

Email course staff at: introml-winter22-staff@ttic.edu

 

Whom should I contact and where should I ask questions?

Questions about the course material, including questions about class, the slides or other reference material --> ask on Piazza, in person during TA or instructor office hours, or during lectures or recitations.  Do not ask by direct email (we will not answer such questions via email, and instead refer you to Piazza).

Clarifications and technical problems with the homework, or general questions about course logistics (not specific to you) --> ask on Piazza (preferred) or during TA office hours.  Please post on Piazza instead of emailing us since others might have also encountered the same problem.

Help with the homework --> seek help during TA office hours.

Questions and concerns about grading, including mistakes in grading, late submissions, extensions and other special arrangements or concerns that are specific to you --> email us at introml-winter22-staff@ttic.edu. Please use this email address instead of emailing a staff person individually.

Questions about registration, consent, auditing, dropping, canvas and pizza access --> email us at introml-winter22-staff@ttic.edu.

General comments or feedback  --> via Piazza, privately via Piazza (only staff can see private posts), email introml-winter22-staff@ttic.edu (preferred) or directly to the relevant staff. 

Personally sensitive issue --> you may contact any staff member you are comfortable contacting. 

 

Recommended textbook:

We will not be following any particular text, and there are many others you might find useful.  The approach taken in this course is most similar to the following textbook

Understanding Machine Learning: From Theory to Algorithms

See here for additional resources.

 

Assignments and grading:

The course will include eight weekly assignments, generally due on Thursdays (the first two assignments will have automatic extensions due to the switch to remote learning).  About half of the assignments will be programming and experimentation assignments and half will be written assignments. 

The assignments are a required and integral part of the course.  Some topics will be introduced primarily through assignments, and many exam questions will be heavily based on the assignments.  Passing grades on at least six (out of eight) assignments is required in order to receive a passing grade in the course.  Each assignment counts for 5% of the final grade, but if the final exam score is higher than any of the HW scores, the exam score can replace two HW scores (automatically), and possibly partially replace additional HW scores, if mastery of the HW material was demonstrated on the final (at the discretion of the instructor).

Late HW policy: For HW 3 through the last HW, students have a total budget of 3 late days to use at their discretion.  You can decide if and how you want to allocate your budget across the homeworks. Each day after a HW deadline counts as a late day. 

 

Prerequisites: 

1) All the following core undergraduate courses: Probability, Linear Algebra, Multi-variate Calculus, Introduction to Algorithms, and Introduction to Programming

AND

2) One of the following: TTIC 31150 Mathematical Toolkit (recommended), or 25300/35300 Mathematical Foundations of Machine Learning, or STAT 34300 Applied Linear Statistical Methods.  Instructor consent without these pre-requisites might be granted to students with a high degree of mathematical maturity (eg graduate students in math-related areas and undergraduates that did well in honors analysis) and to graduate students with background in machine learning.

AND

3) Passing a short online entrance quiz, covering basic probability, linear algebra, calculating gradients, understanding algorithms and runtime analysis of algorithms. 

The entrance quiz:  The quiz does not require calculations (beyond simple arithmetic without needing a calculator), touches only on core concepts, does not require creativity or coming up with solution approaches, and should take under 40 minutes to complete.  If you are comfortable with the pre-perquisites, you should not need to study for the quiz.  If the quiz takes you over an hour to complete, you are struggling with some of the questions, or you need to study in order to answer the questions, then you should reconsider applying to take the class.

 

 

Consent Request:

To request consent to take the course you must complete the following online form and quiz: https://forms.gle/6dEUZjBBtnK2RP7k7 

• You may complete the form once, and you must take the quiz immediately after completing the first part of the request form.  Completing the form will take you 10-15 minutes and completing the quiz another hour---only click the link above once you are ready to complete the form and take the quiz.
• You must use a @uchicago.edu google account in order to access the form.  Please contact us if you are not a TTIC or UChicago student and do not have a @uchicago.edu google account.
• Anyone wishing to take the course, or participate it in any way, must complete the form and quiz.  Even if you contacted us by email, you must still complete the form and include all relevant information in the form.  I generally discourage auditing vs registering for the course, but in any case you must first complete the form and take the quiz before we can discuss further.

We will communicate with you the decision regarding registration consent after the quiz is submitted, based on prerequisites, the quiz, seat availability and program of study.