Course Info
Deep learning is a powerful and relatively-new branch of machine learning. In recent years it has been successfully applied to some of the most challenging problems in the broad field of AI, such as recognizing objects in an image, converting speech to text or playing games. In many such tasks, the state of the art performance today is attained by deep-learning algorithms, in some cases surpassing human-level performance.
This course will focus on the theory and algorithms behind deep learning, as well as on hardware and software interfaces that allow efficient training of deep learning algorithms. It will provide both the necessary theoretical background and the hands-on experience required to be an effective deep learning practitioner, or to start on the path towards deep learning research.
Learning Outcomes
At the end of the course, the student will:
- Understand the key notions of deep learning, such as learning regimes, model types, optimization and training methodologies.
- Be able to apply deep learning algorithms to real-world data and problems.
- Know how to effectively use python and deep-learning frameworks to implement models and algorithms from the recent literature.
- Know how to leverage GPUs and write custom computational kernels to accelerate both training and inference.
- Perform a small research project using the studied notions and techniques.
Administration
Evaluation: 100% Homework assignments.
Language: The course will be taught in English.
Credits: 3.0.
Prerequisites
- A good background of linear algebra, probability and calculus. See the supplemental material page if you need a refresher on one of these.
- Programming competency. The course will be very hands-on; much programming will be required. We will use Python exclusively, so it’s recommended to have experience with it.
- An introductory course about machine learning and/or signal/image processing.
Collaboration Policy and Honor Code
By enrolling in this course, you agree that you will strictly follow our collaboration policy as specified below. Any violation of this policy will result in an immediate failure in the course, and treatment by the Technion regulations committee.
- Submission of assignments is in singles or pairs. You are free to form study groups and discuss homeworks with other students. However, you must implement all required code independently of other groups (only with your submission partner).
- Submitted work must only be your own. You must do your own thinking, coding, debugging and write all answers yourself. We will run automatic plagiarism-detection software on your submissions to enforce this policy.
- You may not use any solutions from previous semesters’ homeworks.
- You may not share your solutions with other students.
- You may not upload your homework solutions to any public website, such as github. Private repos are OK, but they must remain so even after course completion.
Course Piazza (Q&A Site)
We will be using Piazza for class discussion. The system is highly catered to getting you help fast and efficiently from classmates and TAs.
Please refrain from using email for non-personal questions and all questions relating to course material. Such questions should be posted on Piazza only and they will be addressed by the course TAs.
Notes about posting on Piazza:
- Select the proper post type (at the top of the “New Post” page). Select “Question” if you need an answer.
- Select the appropriate folder for your question, e.g. “hw1”, “lectures”, etc.
Course Staff
Lecturers
TAs
Checkers
Literature
The course does not follow any specific book. For your own reference, the following material may be useful.
-
Deep Learning
Ian Goodfellow, Yoshua Bengio, Aaron Courville
MIT Press, 2016
Detailed Syllabus
Due to the COVID-19 pandemic, this semester the course will be given using a remote-learning approach, over Zoom only.
The lectures will follow a flipped-classroom approach: Students will be requested to watch recorded video lectures as a mandatory course requirement. We provide videos and written material, on the course Lectures page, to facilitate self-learning of the core topics. The in-class (zoom) lectures will be short (1h), optional, and cover more advanced material, such as state of the art approaches from the latest research.
The Tutorials are based on detailed and self-contained Jupyter notebooks, which guide you through a full implementation of one or more models and techniques for solving a specific task. They are meant to teach you the technical aspects of implementing deep learning systems. The in-class (zoom) tutorials will cover all this material - no pre-requisite viewing required.
The course also includes hands-on homework assignments in which you’ll implement working real-world models and run them on GPUs on the course servers. Performing the assignment in full is a crucial aspect of the course, which will provide you with many of the technical skills required to be effective with Deep Learning.
This semester’s syllabus is provided below.
| # | Date | Lecture (video, mandatory) | Supplemental (zoom, optional) | Tutorial | Homework |
|---|---|---|---|---|---|
| 1 | 22/10/2020 |
Introduction (🔗) | Introduction | Env setup, numpy, torch | |
| 2 | 29/10/2020 |
Supervised learning (🔗) | Supervised learning | Supervised learning, PyTorch basics I | HW1 |
| 3 | 05/11/2020 |
Neural networks, CNNs (🔗) | CNNs | MLP, PyTorch basics II | |
| 4 | 12/11/2020 |
Training (🔗) | Advanced training | CNNs, ResNets | |
| 5 | 19/11/2020 |
- | Hardware aspects of training | Optimization | HW2 |
| 6 | 26/11/2020 |
Sequence models (🔗) | RNNs | Sequence modeling, RNNs, TCNs | |
| 7 | 03/12/2020 |
- | Attention and Transformers | Attention | |
| 8 | 10/12/2020 |
Unsupervised learning (🔗) | Unsupervised learning I | Transfer learning, domain adaptation | HW3 |
| 9 | 17/12/2020 |
NO CLASS | |||
| 10 | 24/12/2020 |
Deep reinforcement learning (🔗) | Unsupervised learning II | Deep reinforcement learning | |
| 11 | 31/12/2020 |
Non-euclidean domains (🔗) | Geometric deep learning | Geometric deep learning | |
| 12 | 07/01/2021 |
- | Adversarial robustness | Matrix Completion | HW4 |
| 13 | 14/01/2021 |
Parallel architectures I (🔗) | DNN Compression | ||
| 14 | 21/01/2021 |
Parallel architectures II (🔗) | DNN Hardware | CUDA |