Hands-On Deep Learning (HS 2026)
Time: Tuesday 12:15-16:00
Place: HG G1
Language: English
Register: for this course on myStudies
Contact Person: Susann Staub
Student TAs: Anselm Ivanovas, Basil Rohner, Joshua Durrant, Julian Isser, Julius Wendland, Mattia Loszach, Máté Tirpák, Sascha Pucillo, Simon Meier, Yves Berger
Head TA: David Jenny
This lab offers hands-on deep learning exercises using PyTorch, covering computer vision, audio processing, graph neural networks, natural language processing, reinforcement learning, and generative AI. The material is organized into six topics, each spanning two weeks. The course is conducted on CodeExpert, sign up here. Important: you need to sign up in myStudies before signing up in CodeExpert!
Prerequisites
This course has two prerequisites:
- Students should be familiar with deep learning concepts, such as those covered in Computational Thinking (Chapters 5 and 6) or through self-study.
- Students must know an imperative programming language (e.g., C, C++, Java, JavaScript, or Python). Python is used in this course, and understanding the Python Cheat Sheet is required.
Passing Requirements
To pass the lab, you must earn 150 points out of a possible 180 points. Each topic is worth 30 points, so aim to earn at least 25 points per topic. You can track your progress on CodeExpert. The point distribution per topic is as follows:
| Task | Points | Grading Type | Comments |
|---|---|---|---|
| Session | 4 | Point Scale | Two attendance codes per session (start and end), each worth 2 points. |
| Notebook | 6 | Pass/Fail | A solved notebook gets the full 6 points. |
| Challenge | 6 | Point Scale | Minimal solution: 1 point; Excellent solution: 6 points. |
| Discussion | 14 | Pass/Fail | All discussions receive the full 14 points. |
If you miss a session or discussion for a valid reason (e.g., doctor's note, military service, exam), please email Susann Staub. Missed sessions with a valid, documented excuse will earn session points. Missed discussions with a valid excuse can be re-taken during a session. Not valid reasons: attending other lectures, courses, programs, volunteering, etc. Students are allowed to re-take at most one discussion due to an unexcused absence, which must also be done during a session.
Deregistration
We follow the standard ETH rules. For the first few weeks you can remove this course unit yourself in myStudies. Note that staying registered and not participating counts as a fail, so please deregister in time if you decide not to take the course. If you want to request a late deregistration, please contact Susann Staub, and your own study administration.
Session
In each session, students work independently or in small groups to solve the notebook and challenge. Most students will need a few extra hours to finish everything after the session. TAs are available to answer questions and provide guidance. Bring your laptop to the session. A limited number of machines will also be available in the room.
Notebook
Each topic is explored using a Python notebook. While collaboration with peers is encouraged, you are required to write and submit your own solutions. The notebook links are listed below.
The notebooks require GPUs, and you have two options for accessing them:
- Google Colab: A user-friendly web-based interface offering free access to computational resources for a limited duration each day. Most students find this the easier option, unless you need your free credits for another course, or for some other reason do not wish to use Google Colab. Colab Tips and Tricks: Save a copy of the notebook in your own 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, or switch the runtime to CPU for the parts that do not need a GPU.
- Snowflake cluster: A GPU cluster using SLURM hosted by us, where you can start an interactive session. Login details are sent via email. Snowflake instructions.
Both options are fully supported and the choice is yours. Running the notebooks on your own machine is not officially supported; the TAs will help if they have capacity, but students using Colab or Snowflake are helped first.
Challenge
The challenge builds on the material covered in the notebook and is available on CodeExpert. Remember, do not copy anyone else’s code or solutions. After submitting your solution, you receive feedback on the number of points earned immediately after your model finishes training. The top three submissions for each challenge will receive a small prize!
Discussion
The goal of the discussion is to help you better understand the topic and receive feedback from a TA. Each discussion will take about 15 minutes, and you'll be paired with another student. Please ensure that you sign up in CodeExpert for a discussion slot in advance.
Schedule
Sessions and discussions alternate every week. The notebook and challenge must be submitted latest on the day before the discussion. An extra repeat discussion will be held on TBA. Slots open one week before the discussion.
| Topic | Topic TA | Notebook Link | Date / Deadline | ||
|---|---|---|---|---|---|
| Session | Notebook and Challenge | Discussion | |||
| Introduction to Deep Learning | TBA | TBA | 22.09.2026 | 28.09.2026 | 29.09.2026 |
| Computer Vision and Audio | TBA | TBA | 06.10.2026 | 12.10.2026 | 13.10.2026 |
| Natural Language Processing | TBA | TBA | 20.10.2026 | 26.10.2026 | 27.10.2026 |
| Graph Neural Networks | TBA | TBA | 03.11.2026 | 09.11.2026 | 10.11.2026 |
| Reinforcement Learning | TBA | TBA | 17.11.2026 | 23.11.2026 | 24.11.2026 |
| Generative Computer Vision | TBA | TBA | 01.12.2026 | 07.12.2026 | 08.12.2026 |
Updates