Hands-On Deep Learning (HS 2024)

Overview

When: Monday 12:15-16:00
Place: HG G1
Language: English
Register: for this course on myStudies
Head TA: Till Aczél
TAs: TBA

This lab introduces deep learning through the PyTorch framework in a series of hands-on exercises, exploring topics in computer vision, natural language processing, audio processing, graph neural networks, and representation learning. You'll explore common and advanced neural architectures, understand network structures, and apply them to solve examples and challenges.

Prerequisites

This course has two prerequsites:

Students must have some familiarity with the ideas behind deep learning. Any course that teaches the fundamentals of deep learning, e.g. Computational Thinking (Chapters 5 and 6), is sufficient, and so is an adequate amount of self-study.
Students must know an imperative programming language, e.g., C, C++, Java, Javascript, or Python. The course uses Python as a programming language, so we expect students to know the basics of Python (e.g., lists, control structures, functions). Concretely, understanding the concepts of this Python Cheat Sheet is necessary (but also sufficient) to attend the course.

Structure

This course is organized into specific topics, with each topic spanning two weeks.

Session week: Includes a session, the notebook, and challenge submission.
Presentation week: You present your work.

The detailed schedule can be seen below.

Session

In each session, students work independently or in small groups, following a Python notebook, to explore various aspects of the topic of the given session. There are going to be TAs present at every session, answering students' questions and providing guidance where necessary.

Bring your laptop. Some machines will also be available in the room. The exercises require GPUs. You have two alternatives:

Snowflake cluster (recommended) A GPU cluster using SLURM, where you can start an interactive session. Login details are sent via email. Instructions on how to start the jupyter notebook.
Google Colab: A user-friendly interface offering free access to computational resources for a limited duration each day. Colab Tips and Tricks: Save the notebook in your drive before you begin working on it to avoid losing your progress. If you run out of GPU hours, you may switch to an alternative Google account.

Submission

The submission consists of 2 parts:

Notebook: You must submit the filled-in notebook after each session.
Challenge: The last part of each notebook is a challenge. You must submit the output of the challenge solution separately. The challenge cell in your notebook needs to generate your uploaded challenge CSV. The top 3 submissions for each challenge receive a small award!

The submission deadline for each topic is Friday of the session week. Collaboration with colleagues is encouraged, but everyone must write their own programs. Do not copy other people's code or solutions!

Presentation

To demonstrate your understanding of the topic and receive feedback on your work, you will present to a TA. Each presentation lasts approximately 10 minutes, during which the TA will ask you questions about your notebook and the challenge. Two students will be teamed up in a presentation. You need to sign up to a presentation slot.

Passing Requirements

To pass the lab, you need to earn 18 credits. You can earn a maximum 4 credits for each topic:

1 credit for attending the session.
1 credit for solving the notebook and challenge and submitting them on time.
2 credits for the presentation.

If you have a valid reason for missing a session or presentation (e.g., doctor's note, military service), please email Till Aczél. If you miss a presentation with a valid excuse, we will review your submission and award you 2 presentation credits.

Schedule

Topic	Topic TA	Session	Submission	Presentation
Introduction to Deep Learning	Till Aczél	September 23	September 27	September 30
Computer Vision and Audio	Luca Lanzendörfer	Oktober 7	Oktober 11	Oktober 14
Graph Neural Networks	Joël Mathys	Oktober 21	Oktober 25	Oktober 28
Natural Language Processing	Frédéric Berdoz	November 4	November 8	November 11
Reinforcement Learning	Saku Peltonen	November 18	November 22	November 25
Generative Computer Vision	Benjamin Estermann	December 2	December 6	December 9