DLDIY (2019)

Cours MAP583 taught at Ecole polytechnique (01-04/2019) with Andrei Bursuc

• Lesson 1:
  1. (slides) introductory slides
  2. (code) a first example on Colab: dogs and cats with VGG (ready for Google Colab)
  3. (code) making a regression with autograd: intro to pytorch (CPU compatible)
  4. (code) using Colab features to overfit (CPU compatible)
• Lesson 2:
  1. (slides) refresher: linear/logistic regressions, classification and PyTorch module.
  2. (slides) neural networks, backpropagation and convolutional networks.
  3. (code) understanding convolutions and your first neural network for a digit recognizer.
• Lesson 3:
  1. (slides) embeddings and dataloader
  2. (code) Collaborative filtering: matrix factorization and recommender system (CPU compatible)
  3. (slides) Convolutions and siamese networks
  4. (code) Siamese networks on MNIST
• Lesson 4:
  1. (slides) optimization for DL
  2. (code) gradient descent optimization algorithms (CPU compatible)
  3. (slides) going deeper
  4. (code) homework unsupervised learning with autoencoder
• Lesson 5:
  1. Playing with Tensorboard
  2. Reccurrent Neural Networks: slides and associated code
  3. (code) PyTorch tutorial on RNN (CPU compatible)
• Lesson 6:
  1. Generative Adversarial Networks, slides
  2. Conditional and info GANs (CPU compatible)
  3. Word2vec (CPU compatible)
• Lesson 7:
  1. Playing with word embedding (CPU compatible)
  2. Packing sequences (CPU compatible)
  3. Structured Self-attentive Sentence Embedding paper code to obtain Glove NLP mini-project
  4. Transfer learning, slides
  5. A project repo

Back to dataflowr