Deep Learning Bootcamp (Hybrid, July 8th - July 26th)
Since 2016, LauzHack has organized hackathons at EPFL in Lausanne, Switzerland. We also organize tech talks during the school year.
Now, we want to do a summer event: a Deep Learning Summer Bootcamp (Hybrid format: lessons on campus with stream to Zoom and recording. You can be fully remote, fully on campus or you can do some days online and some offline, as you wish).
Participation is free. Registration for the event on Luma is required. Language: English
Join our Discord server for announcements and discussions. Time schedule, Zoom link, etc. are there.
Event description:
DL Bootcamp has two stages:
Introduction to DL – for students with little or no DL background. (July 8th - 12th)
Advanced specific topics – for those who already know DL or participated in part 1. (July 15th - July 26th)
The core features of our bootcamp:
You will create a cool DL project from scratch that will improve your portfolio
Many topics to choose from. Each project is a state-of-the-art DL model from different years
Project review from LauzHack mentors
We will discuss best practices of R&D DL coding and provide you with a template
Discord server with all participants and mentors
No matter if you want to do research work or industry work in the future, you can learn all the necessary skills here
Hybrid Format: lessons on campus and Zoom. All lessons will be recorded
Free, no grade, no credits — no stress self-development
LauzHack participation certificate will be given to all participants
Topics:
Introduction: PyTorch, Deep Learning concept, Datasets, Optimizers, Schedulers, Loss Functions, FC, CNN, RNN, Transformer, Python Dev Tools, Git, WandB, Hydra, R&D Coding
Deep Learning in Audio: Voice Biometry (a.k.a. deepfake detection), Neural Vocoders, GANs, GNNs
NLP: Introduction to NLP, Past vs Modern approaches, modeling texts with RNNs and Transformer, seq2seq machine translation, text classification (BERT), text generation (GPT), and text generation at scale (LLMs)
CV: object detection, image segmentation, and 3D vision (basics of volume rendering and implicit surface representations, NeRFs, NeuS, DeepSDF, and their applications)
Guests' Material: DeepRL, XAI (Model interpretation), On-Device Learning, Distributed / Decentralized DL
Prerequisites:
Python knowledge: Classes, functions, for-loops, etc.
NumPy would be helpful. Have a look at this notebook and this lab from COM-202 course to get familiar with the basics. You can also check some of the PyTorch basics here, however, we will go through them.
Lessons format:
Lectures: theory, task description, model architectures, core algorithms for the specific topic, etc.
Seminars: practice, jupyter notebook which will be filled during the class with the mentor, shows how theory looks in practice
Time Schedule:
From 10:00 to 12:00 and from 13:00 to 15:00 three days per week.
The schedule for each week: See pinned message on Discord
Projects:
Most of the advanced topics will be supplemented with a homework project description. The project goal is to reproduce the results from the state-of-the-art paper of a certain year. You will read papers, write your own code, configure experiments, add logging, and train models from scratch. The projects are followed by a tiny report.
The project is optional. It is here to provide you with practical experience and supplement your portfolio.
Each project will have a mentor, who will review your code and report and give some advice/corrections.
Participants may do as many projects as they want, however, only limited number of people for each project will be reviewed due to the limited number of mentors. We will do our best to give at least one review to each of the participants. Nonetheless, the project will be a good experience on its own too.
Compute:
We will show how to use free cloud compute. It will be enough for the projects.
If you have any questions send an email to lauzhack@epfl.ch or contact us in Discord server.