Welcome

Welcome to the webpage of the Deep Learning course. This course is taught by Professor Stéphane Gaïffas. This course is for students from the Masters 2 programs MIDS and M2MO. On this webpage you will find all the teaching material (mainly lecture slides, jupyter notebooks and code for the labs).

Important links

Zoom

The course takes place on Mondays 8:30 - 12:30 at Halle aux Farines room 027C, the agenda is here https://m2mids.github.io/m2mids/edt/

Slack

Go and log in the following Slack space (ideally install Slack on your laptop):

• https://join.slack.com/t/m2midsunivdeparis/shared_invite/zt-10zt2lk02-qeLXCHaMjlGSNpu1kHO3Jg

if you haven’t already. You can ask your question in the #cours-deep-learning channel.

Agenda for the course

• Week 1 Lecture: Course logistics and the success to deep learning. Lab: Polynomial logistic regression versus multi-layer perceptron on toy datasets.
• Week 1 Lecture: Machine learning recap, history of neural networks and the main building blocks. Lab: Quick description of PyTorch tensors.
• Week 2 Lecture: Some hyperparameters, regularization techniques and practical recommendations. Lab: Using pre-trained neural networks for complex tasks.
• Week 2 Lecture: About optimization and optimizers for deep learning. Lab: Homemade perceptron on toy-data and multi-layer feedforward neural net on CIFAR-10 using PyTorch.
• Week 3 Lecture: Convolutional neural networks or how to make a neural net see ? Lab: Convolutional neural networks in PyTorch and PyTorch-Lightning
• Week 3 Lecture: A description of some computer vision tasks and architectures to solve them, such as semantic segmentation with U-nets and other architectures.
• Week 5 Lecture: We will explain how to learn from sequence using neural networks, including architectures such as RNN, LSTM and GRU.
• Week 5 Lecture: We will finish this course with some advanced topics in deep learning’s current state-of-the-art, include transformer architectures and self-supervised learning

References

Here is a list of learning ressources that can be useful for this course, among many others.

• Deep learning book about the main concepts
• Francois Chollet’s book about tensorflow and keras programming
• Aurélien Géron’s book about general machine learning with scikit-learn and tensorflow / keras
• Jeremy Howard’s book and GitHub repo https://github.com/fastai/fastbook

Technologies

The course will focus mainly pytorch and tensorflow for deep learning, and we will use the regular Python stack for data-science, namely scikit-learn and pandas, among some others stuff.