Applied Machine Learning Syllabus and logistics Siamak Ravanbakhsh - - PowerPoint PPT Presentation
Applied Machine Learning Syllabus and logistics Siamak Ravanbakhsh COMP 551 (fall 2020) Admin Live Class: Tuesday & Thursday, 10:05 am - 11:25 am Location: Online Zoom Meeting Instructor: Siamak Ravanbakhsh <siamak@cs.mcgill.ca> Office
Siamak Ravanbakhsh
COMP 551 (fall 2020)
Live Class: Tuesday & Thursday, 10:05 am - 11:25 am Location: Online Zoom Meeting Instructor: Siamak Ravanbakhsh <siamak@cs.mcgill.ca> Office hours: Thursday, 11:30 am - 12:30 pm
COMP 551 | Fall 2020
live classes that are recorded
this may change depending on the outcome of first few lectures
we may also try flipping the classroom:
prerecorded lectures spend more time on the "applied" side
Mostly third year undergraduates 2nd year undergraduates, look out for prerequisites! prior to add-drop
prior to add-drop
Siamak Ravanbakhsh (pronounced almost like see-a-Mac)
Assistant Professor in the School of Computer Science Canada CIFAR AI Chair at Mila
research interest: representation learning
what is the right representation for an AI agent? how do we learn quickly from data and perform inference
two approaches to this problem that I explore
using probabilistic (graphical) models using invariances and symmetries I also collaborate with physicists and cosmologists
COMP 551 | Fall 2020
Amy Zhang (amy.x.zhang@mail.mcgill.ca)
PhD candidate co-supervised by Joelle Pineau and Doina Precup specializing in generalization in reinforcement learning.
Arna Ghosh (arna.ghosh@mail.mcgill.ca)
Second year PhD student supervised by Blake Richards working on brain-inspired AI
Arnab Kumar Mondal
(arnab.mondal@mail.mcgill.ca)
2nd year PhD student supervised Kaleem Siddiqi
Arushi Jain (arushi.jain@mail.mcgill.ca)
I am 2nd year Phd supervised by Doina Precup & Pierre-Luc Bacon. My research area is reinforcement learning and risk-sensitivity.
Samin Yeasar Arnob
(samin.arnob@mail.mcgill.ca)
First year Ph.D. student. Research Intereset: Reinforcement learning, Imitation learning
Tianyu Li (tianyu.li@mail.mcgill.ca)
I'm a 3rd year PhD student supervised by Doina Precup and Guillaume Rabusseau. My research area is primarily
reinforcement learning.
Haque Ishfaq (haque.ishfaq@mail.mcgill.ca)
PhD student supervised by Doina, working on exploration in RL with provable guarantees
Howard Huang (howard.huang@mail.mcgill.ca)
PhD Student supervised by Doina, working on hierarchical reinforcement learning
Practice applying them
Understand the theory behind learning algorithms
lectures readings late midterm exam weekly quizzes code accompanying (some) lectures programming during the class (?) team projects hands on tutorials late midterm and quizzes (?)
late midterm exam (~20%) date TBD, sometime in November team projects (~60%) by teams of 3 students teams change for every project first project will be posted before add-drop try to pick your team members in the same time zone weekly quizzes (~20%) encourage you to review recently covered topics posted on Mondays, available for 24h a practice quiz one before add-drop to a large extend relies on students' honesty
All due dates are 11:59 pm Montreal time (EST). No make-up quizzes will be given. We will ignore your worst quiz, so you can miss one quiz without penalty Maximum of five days late for a project 20% penalty for any late submission
Python-Numpy programming skills
we start using this from the next lecture we will have a Python-Numpy review tutorial very soon
Probability theory
you should know random variables, expectations etc.
Linear algebra
matrix product, determinant, null-space of a matrix, span of a set of vectors ...
limit, derivative, integral etc. we will have a tutorial to review these two topics
COMP 551 | Fall 2020
Python-Numpy (soon) Probability theory and linear algebra review (soon) Scikit-learn (around the end of September) a Python package for ML We implement many methods in this package from scratch PyTorch Tutorial (around the end of October) a Python package for deep learning
Part 1: a short tour of ML
K-Nearest Neighbours Decision Trees Some Basic Concepts Model selection Curse of dimensionality Dimensionality reduction Maximum likelihood and Bayesian reasoning Multivariate Gaussian Expectation Maximization Naive Bayes This is going to change during the semester
Part 2: Linear models, probabilistic interpretations and gradient optimization
linear regression Nonlinear bases Logistic and softmax regression Gradient descent Regularization Bias-variance decomposition Estimating the uncertainty of predictions* This is going to change during the semester (some topics may be dropped depending on our progress)
Part 3: Kernels and more
Perceptrons Support Vector Machines Kernel trick * Gaussian process *
Part 4: Deep learning
Multilayer Perceptrons Gradient computation Automated differentiation and Backpropagation Convolutional neural networks Frontiers
COMP 551 | Fall 2020
Deep Learning (2016) by Ian Goodfellow, Yoshua Bengio, and Aaron Courville
Machine Learning: A Probabilistic Perspective by Kevin Murphy (2012)
No required textbook but slides will cover chapters from the following books, available
Pattern Recognition and Machine Learning by Christopher Bishop (2007) The Elements of Statistical Learning: Data Mining, Inference, and Prediction (2009) by Trevor Hastie, Robert Tibshirani and Jerome Friedman
Is this the right course for me? How can I take the class? I'm on the waiting list...
prerequisites are very important this course needs lots of time and effort on your side see the course webpage from last year to get an idea there are several other ML courses you should consider (anti-requisite to COMP 551)
COMP 451: fundamentals of ML COMP 596: ML for biomedical data ECSE 551: ML for engineers
if none of the courses above are suitable and you can't take this course next semester, then let me know let's revisit close to the end of add-drop period
How can I take the class? I couldn't get on the waiting list...
if you have special circumstance that should be treated as an exception please let me know
COMP 551 | Fall 2020
your answers to quizzes, projects and the exam must be your own work you should not share your solutions with other students Copying (even copying ideas) without giving credit is plagerism you should be respectful in the online course forum and in all other communications you should NOT (re-)post any of the course materials online. This includes: video lectures, codes, and quizzes
Please complete the following poll by Friday (EST) https://forms.gle/852uTec8PVbjZLCbA
your answers so far...
your time zone
your answers so far...