Course TAs: Fredrik Bagge Carlsson, Martin Karlsson
This is a PhD course organized in the form of a reading group where the work is carried out mainly by the participants.
We will use the book Deep Learning by Bengio et al.
To this we will also look at
- Some DL platform(s), Tensorflow (or alternatives)
- Data sets, especially some interesting from biology provided by Magnus Fontes
- Video Lectures
- DL competitions, see kaggle
- Cloud/GPU implementations
Collection of Deep Learning material
Prerequisities: You are supposed to know the material in Ch 1-5 in Bengio beforehand. If you are completly new to machine learning you might want to first follow the ML course given at the math department
Examination: For credits (7.5 ECTS) you should be resposible for one session, complete at least half of the homeworks, and do a smaller deep learning project of your choice.
We will upload our homeworks on this git repository.
Schedule
- Meeting 1, Introduction, 15/9: Thursday Sep 15 at 13.15 in M:2112B (control dept seminar room). Slides from the meeting: BoB, Fredrik BaggeCarlsson, MagnusFontes, KalleÅström. Start reading Goodfellow part 2, pp 167-487. Install Synapse and Slack when you get invitiations. Also install a platform of your choice (e.g. Tensorflow) and run a tutorial of your choice. Watch the video by Bengio: https://youtu.be/JuimBuvEWBg
Meetings: Wednesdays 10.15-12.00 M:2112B from 21/9 onwards
- Meeting 1.5, 21/9 Shorter meeting for planning the schedule.
- Meeting 2, 28/9: Responsible: Fredrik Bagge. Autoencoders.
- Meeting 3, 5/10: Responsible: Kalle Åström. Convolutional Networks. Here are the exercises
- Meeting 4, 12/10: Responsible: Bo B. Structured Probabilistic Models, RBMs.
- Meeting 5, 19/10: Guest lecture Del Bue, Alessio, MH:333 (math building) 10.15, Scene understanding from motion
- Meeting 6, 26/10: Responsible: Martin. Sequence Modeling: Recurrent and Recursive Nets.
From November we change meeting times to Tuesdays 13.15-15 (except guest lecture on November 9)
- Meeting 7, 1/11 13.15: Responsible: Gabriel. Deep Reinforcement Learning.
- Meeting 8, 8/11 13.15: Responsible: Jacob. Tensorflow jam session. Code example.
- Meeting 8.5, 9/11 14.00: Guest lecture by Michael Felsberg Linköping
- Meeting 9, 15/11 13.15: Responsible: Mattias. Deep Learning using GPU.
- Meeting 10, 22/11 13.15: Responsible: Najmeh. Improving Imputation using stacked denoising autoencoders
- Meeting 11, 29/11 13.15: Responsible: Lianhao, DL for Natural Language Processing
- Meeting 12, 6/12 13.15: Responsible: Johan B. Deconvolution networks
- Meeting 13, NOTE: Time changed to Thursday 15 Dec at 13.15: Responsible: Carl Åkerlindh, Practical overview of optimization of Deep Networks. Also have a look at this blog, and this video about Bayesian optimization of hyperparameters
- Ideas for a DL project course during the spring: 15/12 at 14.15: (Bo Eliasson, BoB, all) Data sets and future challenges. Planning of DL project course. Files from Bo Eliasson's presentation
You should upload your homework on this git repository and fill in this google doc (you need 6 finished home works. They can overlap, and some cooperation is allowed, but state who did what). Send also a presentation about your mini project