makci97/MLatURL2018

ML at LaSalle Universitat Ramon Llull 2018

Deep Learning introduction at University Ramon Llull LaSalle, 2018

Day1: Introduction

• logistic regression, neural nets, gradients
• backpropogation, chain rule
• optimization strategy
• weight initialization, activations
• DL platforms, computational graphs, pytorch
• links:
  • https://pytorch.org
  • http://pytorch.org/tutorials/beginner/blitz/autograd_tutorial.html#sphx-glr-beginner-blitz-autograd-tutorial-py
  • https://github.com/spro/practical-pytorch

Day2: Dealing with images

• convolutional layers
• vanishing/exploding activations
• pooling layers
• CIFAR10
• applications: image boxing, segmentation, transformation
• links:
  • http://cs231n.github.io/convolutional-networks/

Day3: Dealing with sequences

• Recurrent Neural Networks
• vanishing/exploding gradients
• LSTM, GRU
• Naming a character
• links:
  • https://karpathy.github.io/2015/05/21/rnn-effectiveness/
  • https://larseidnes.com/2015/10/13/auto-generating-clickbait-with-recurrent-neural-networks/
  • https://jhui.github.io/2017/03/15/RNN-LSTM-GRU/

Day4: Reinforcement learning

• strategy learning
• OpenAI gym
• Pole balancing
• links:
  • https://github.com/yandexdataschool/Practical_RL
  • https://github.com/yandexdataschool/sklearn-deeprl/blob/master/Sklearn_CEM_LunarLander.ipynb

Day5: Generative Adversarial Networks

• Autoencoders, dimensionality reduction
• domain adaptation
• GANs
• Art style transfer
• links:
  • https://arxiv.org/abs/1409.7495
  • https://github.com/soumith/ganhacks

Tons of kudos to Sasha Panin!