CEng 793 - Advanced Deep Learning - Spring 2017
[Image generated by http://deepdreamgenerator.com/]
- Instructors: Asst. Prof. Dr. Emre Akbas and Assoc. Prof. Dr. Sinan Kalkan; E-mails: {emre|skalkan} at ceng dot metu.edu.tr; Offices: B-202 & B-207; Office hours by apppointment
- Lectures: Thursdays 13:40-16:30 at BMB4
- Online communication: (e-mail list, forum, homework submissions) https://odtuclass.metu.edu.tr/
- Syllabus: pdf
Course conduct
This is an advanced graduate class that builds upon CEng 783 (Fall 2016; Spring 2017). The emphasis will be on paper discussion and term project. In the first four or five weeks, we will give an overview of various topics in deep learning, then in the rest of the course, we will be discussing 2-3 papers per week. Each student will present but most importantly, each student will be expected to read the weekly assigned papers before coming to class.
Announcements
- 02/02/17: Course website is up.
- 02/02/17: First lecture is on Thursday, Feb 23, 2017.
- 17/02/17: Students who want to take this course should come to the first lecture.
Detailed weekly schedule & lecture notes:
Week 1 (Feb 23)
- Introduction.
Week 2 (Mar 2)
- Review of machine learning and optimization basics, MLPs and auto-encoders.
Week 3 (Mar 9)
- Important announcements.
- Review of CNN and RNN.
Week 4 (Mar 16)
- Project proposals were reviewed and discussed in class.
- Next week’s (March 23) papers are:
- Generative Adversarial Networks [Presenter: 1881739]
- Generative Adversarial Text to Image Synthesis [Presenter: 2137297]
Week 5 (Mar 23)
- Generative Adversarial Networks presentation by İlke Çuğu.
- Generative Adversarial Text to Image Synthesis presentation by Ezgi Ekiz.
- Next week’s (March 30) papers are:
- Google’s Multilingual Neural Machine Translation System: Enabling Zero-Shot Translation [Presenter: 1560101]
- Pixel Recurrent Neural Networks [Presenter: 1893122]
Week 6 (Mar 30)
- Pixel recurrent neural networks presentation by Osman Tursun.
- Next week’s (April 6) papers are:
- How to Construct Deep Recurrent Neural Networks [Presenter: 1901735]
- HyperNetworks [Presenter: 1819630]
Week 7 (Apr 6)
- HyperNetworks presentation by Engin Deniz Usta.
- Next week’s (April 13) papers are:
- End to end memory networks [Presenter: 2204451]
- Tracking the World State with Recurrent Entity Networks [Presenter: 1902717]
Week 8 (Apr 13)
- Next week’s (April 20) papers are:
- Neural Random-Access Machines [Presenter: 1881887]
- Neural Programmer-Interpreters [Presenter: 1819242]
Week 9 (Apr 20)
- Neural Programmer-Interpreters presentation by Fethiye Irmak Dogan.
- Next week (April 27): project progress demos
Week 10 (Apr 27)
- Project progress demos.
- Next week’s (May 4) papers are:
- Human-level control through deep reinforcement learning [Presenter: 2270981]
- Continuous control with deep reinforcement learning [Presenter: 1631027]
Week 11 (May 4)
- Human-level control through deep reinforcement learning presentation by Muhammed Kocabaş.
- Next week’s (May 11) papers are:
- Neural architecture search with reinforcement learning [Presenter: 2194538]
- FeUdal Networks for Hierarchical Reinforcement Learning [Presenter: 1631126]
Week 12 (May 11)
- Neural architecture search with reinforcement learning presentation by Erkut Akdağ.
- FeUdal Networks for Hierarchical Reinforcement Learning presentation by Çağlar Seylan.
- Next week’s (May 18) papers are:
- Why and when can deep-but not shallow-networks avoid the curse of dimensionality: A review [Presenter: 1560325]
- Understanding deep learning requires rethinking generalization [Presenter: 1881093]
Week 13 (May 18)
- Next week’s (May 25) paper is Building Machines That Learn and Think Like People [Presenter: 1818889]
Week 14 (May 25)
- Discussion on open issues.