/dataflowr

deep learning courses

Primary LanguageJupyter Notebook

Material for Deep Learning hands-on courses.

Largely inspired by fast.ai course: Practical Deep Learning For Coders but with a different focus.

The main goal of the courses is to allow students to understand papers, blog posts and codes available online and to adapt them to their projects as soon as possible. In particular, we avoid the use of any high-level neural networks API and focus on the PyTorch library in Python.

Constructive comments welcome!

two courses:

A 5 days crash course on deep learning for students with a background in python and ML.

currently taught at ENS, under progress... (stay tuned for updates)

main updates:

faster.ai

  • Day 1:

    1. introductory slides
    2. first example: dogs and cats with VGG
    3. making a regression with autograd: intro to pytorch
    4. using colab to compute features, just run the following notebook on colab. Save it in your drive and open it with colab. It should load the features of dogs and cats in your drive.

  • Day 2:

    1. understanding convolutions and your first neural network.
    2. pausing a bit, slides recap on writing a module and BCE loss.
    3. using colab features to overfit . This practical requires the features computed on colab in day 1.
    4. discovering embeddings with collaborative filtering

  • Day 3:

    1. regularization, dropout, batchnorm, residual net, slides to come...
    2. introduction to NLP with sentiment analysis.
    3. optimization, initialization, slides to come...
    4. unsupervised learning with autoencoders

  • Day 4:

    1. Recurrent Neural Networks, slides to come
    2. pytorch tutorial on RNN
    3. understanding my network: class activation map

  • Day 5:

    1. Generative Adversarial Networks, slides
    2. Conditional and info GANs
GAN InfoGAN

Contact

homepage