Learn PyTorch deep learning, it may only take 5 days.
This course, titled "Hands-on tour to deep learning with PyTorch" by French deep learning researcher Marc Lelarge, allows you to quickly understand deep learning and learn how to apply open-source deep learning projects within 5 days.
This course not only introduces the theoretical foundations of deep learning, but also includes a lot of practical content, including examples such as classification, GAN, word embeddings, as well as code and Colab, which are very convenient and practical.
To prevent you from getting bored, this tutorial even includes emojis, making it very user-friendly.
After completing the course, you will be able to quickly understand neural networks and apply various new projects and resources shared by others to your own projects.
What will be learned in the five days?
What exactly will be learned in the 5-day PyTorch deep learning course?
Course schedule ↓
Day 1#
A very beginner-friendly introduction to deep learning:
Using CNN to distinguish between cats and dogs, with Colab included:
Then learn about PyTorch.
Day 2#
Now that you know what deep learning and PyTorch are, you can delve into more complex concepts.
First, brush up on your math knowledge, you should know about logistic regression, convolution, and other concepts.
You should also have an understanding of various modules in PyTorch:
You can start learning about embeddings, variational autoencoders, and other topics.
Day 3#
Now, you can learn more in-depth theory, such as Bayesian methods and backpropagation, which are covered in today's classes.
To prevent you from getting too tired, the PPT even includes emojis:
In addition, you can also learn about GAN through examples:
Day 4#
Congratulations, you can start learning about RNN, and in addition to GAN, you can also delve into NLP.
For example, using word2vec for word embeddings:
There is also code and a Colab version.
Day 5#
The last day is about taking it to the next level, discussing serious issues such as the black box nature of neural networks.
In addition, there are also topics such as class activation maps and adversarial research, with code and examples. Interested students can continue their research.
The dataflowr course by a French doctor
Finally, this 5-day PyTorch deep learning course is part of a series of courses called dataflowr.
The core author of dataflowr, Marc Lelarge, is a researcher at the French Institute of Digital Science, a PhD in Applied Mathematics from the Paris Institute of Technology, and a part-time professor at the same school. He has been teaching deep learning courses recently.
Inspired by fast.ai, he developed dataflowr, taking a practical approach to deep learning, and named his course "faster.ai", as if you can learn everything in 5 days.
Therefore, the courses in dataflowr are relatively simple and easy to grasp, without highly advanced and complex APIs.
In addition to Lelarge, there are 7 other teachers in this course:
Andrei Bursuc, a research scientist at Valeo.ai;
Alexandre Défossez, Timothée Lacroix, Pierre Stock, and Alexandre Sablayrolles from Facebook AI Research;
Nicolas Prost, a PhD in Machine Learning from the French Institute of Digital Science;
Stéphane d'Ascoli, an AI scientist at Snips.
Although the core team is French, don't worry, the course is not taught in French~
Links#
Course:
https://mlelarge.github.io/dataflowr-web/cea_edf_inria.html
dataflowr:
https://mlelarge.github.io/dataflowr-web/