Fork of Lempitsky DL for HSE master students.
Lecture and seminar materials for each week are in ./week* folders
Attention! This is a new iteration of on-campus deeplearning course. For full course materials '2016, go to this branch
- Create cloud jupyter with repo https://beta.mybinder.org/v2/gh/yandexdataschool/Practical_DL/fall17
- Lecture slides are stored in this folder (odp, pdf). You can also view slides from each week's page.
- Any technical issues, ideas, bugs in course materials, contribution ideas - add an issue
HSE classes are happening on wednesdays, 18-10 till 21-00. [room number TBA]
Everyone who wants to attend the course ping [email protected]
- Bookmark repo https://github.com/yandexdataschool/practical_DL
- Join telegram chat https://t.me/dl_hse_fall17
- (only HSE students) Enroll to anytask.org/course/227 with invite code 7pp6jP3
- Join piazza https://piazza.com/cs_hse/fall2017/dl101/home with access code dl101
- Read rules
- 21.09 - for those using TF + keras in week2: if you have any problems please update your notebooks from current repository (new data reading script).
- 06.09 - Course started
-
week0 Recap
- Lecture: Linear models, stochastic optimization, basic neural networks and backprop
- Seminar: Neural networks in numpy, adaptive SGD
- HW due: 17.09.17, 23.59.
- Please get bleeding edge theano+lasagne installed for the next seminar.
- Issue
- Linux Guidelines
- You may choose tensorflow/pytorch version if you prefer 'em
-
week1 Symbolic graphs
- Lecture: Backprop recap. Deep learning frameworks. Some philosophy. DL tricks: dropout, normalization
- Seminar: Symbolic graphs and basic neural networks
- HW due 24.09.17, 23.59
-
week2 Deep learning for computer vision
- Lecture: Convolutional neural networks, data augmentation & hacks.
- Seminar: Convnets for CIFAR
- HW due 1.10.17, 23.59
-
week3 Advanced computer vision
- Lecture: Computer vision beyond image classification. Segmentation, object detection, identification. Model zoo & fine-tuning
- Seminar: Model zoo. Siamese nets for identification.
-
week4 Unsupervised & generative methods
- Lecture: Autoencoders, Generative Adversarial Networks
- Seminar: Generative Adversarial Networks. [hopefully] Art Style Transfer by Dmitry Ulyanov
-
week5 Deep learning for natural language processing 101
- Lecture: NLP problems and applications, bag of words, word embeddings, word2vec, text convolution.
- Seminar: Word embeddings. Text convolutions for salary prediction.
-
week6 Recurrent neural networks
- Lecture: Simple RNN. Why BPTT isn't worth 4 letters. GRU/LSTM. Language modelling. Optimized softmax. Time series applications.
- Seminar: Generating laws for pitiful humans with mighty RNNs.
-
week7 Recurrent neural networks II
- Lecture: Sequence labeling & applications. Seq2seq & applications. Attention. Batchnorm and dropout for RNN.
- Seminar: Image Captioning
-
week8: Deep reinforcement learning
- Lecture: Reinforcement learning applications. Policy gradient. REINFORCE.
- Seminar: REINFORCE agent with deep neural net policy for RL problems
-
week9: Bayesian deep learning
- Lecture: Bayesian vs Frequentist idea of probability. Bayesian methods around you. Variational Autoencoder. Bayesian Neural Network.
- Seminar: Bayesian Neural Nets; Variational autoencoders [Hopefully by Mikhail Khalman]
- One rule to rule them all
- Project rules
- Project examples
- Reducing lateness penalty
- Feedback form (anonymous)
Course materials and teaching performed by
- Fedor Ratnikov - lectures, seminars, hw checkups
- Oleg Vasilev - seminars, hw checkups, technical issue resolution
- Arseniy Ashukha - image captioning, sound processing, week7&9 lectures
- Dmitry Ulyanov - generative models, week8 lecture, week12 homework assignment
- Mikhail Khalman - variational autoencoders, lecture 12
- Vadim Lebedev - week0 & week6 homeworks