Hauptseminar/Master Seminar:
Deep Generative Models SS2019

Team: Prof. Nassir Navab, Dr. Gustavo Carneiro, Azade Farshad, Magda Paschali, Helisa Dhamo, Oliver Scheel, Ashkan Khakzar, Mahdi Saleh, Leslie Casas, Artem Savkin

Type: Advanced Seminar Course Module IN0014
Type: Master Seminar Module IN2107
Type: Seminar Module IN4826
SWS: 2+0
ECTS: 5 Credits
Location: MI 03.13.010
Time: TBD
Course Language: English

Announcements

• 17-01-2019: Motivation Letters are welcomed. You can send them to Azade Farshad (azade.farshad@tum.de).
• 17-01-2019: Preliminary meeting 31.01.2019 at 4:00pm, MI 03.13.010.
• 14-01-2019: The website is up!

Introduction

• The recent advances of generative models in Deep Learning will be studied in the Seminar Course. Variational Autoencoders (VAEs), Generative Adversarial Networks (GANs), adversarial learning and learning with adversaries compose the main topics of the seminar. A set of papers will cover the aforementioned topics. One or two lectures will be also offered to support the seminar.

Registration

• The registration is through the TUM Matching Platform (from 08.02.19 to 13.02.19).
• If you have any questions regarding this seminar, feel free to contact Azade Farshad.

Requirements

In this Seminar (formerly Hauptseminar), each student is asked to select one paper from a list. In order to successfully complete the seminar, participants have to fulfill these requirements:

• Presentation: The selected paper is presented to the other participants (20 minutes presentation 10 minutes questions). Use the CAMP templates for PowerPoint camp-tum-jhu-slides.zip, or Latex: CAMP-latex-template.
• Written Report: A document of maximum 8 pages is written and submitted one week after the presentation. Please download and use the llncs2e.zip Latex template.
• Attendance: Participants have to participate actively in all seminar sessions.

The students are required to attend each seminar presentation which will be held during this course. Each presentation is followed by a discussion and everyone is encouraged to actively participate. Part of the discussions outcome will be included in the report too. The report must include all references used and must be written completely in your own words. Copy and paste will not be tolerated. Both report and presentation have to be done in English.

List of Publications

Nr	Authors	Title	Conference	Link	Tutor	Student
1	Ruixiang Zhang et. al.	MetaGAN?: An Adversarial Approach to Few-Shot Learning	NIPS 2018	https://papers.nips.cc/paper/7504-metagan-an-adversarial-approach-to-few-shot-learning.pdf	Azade	Silvia S.
2	Aaron van den Oord et. al.	Neural Discrete Representation Learning	NIPS 2017	https://arxiv.org/pdf/1711.00937.pdf	Artem	Faizan K.
3	Pouya Samangouei et. al.	ExplainGAN?: Model Explanation via Decision Boundary Crossing Transformations	ECCV 2018	http://openaccess.thecvf.com/content_ECCV_2018/papers/Nathan_Silberman_ExplainGAN_Model_Explanation_ECCV_2018_paper.pdf	Magda	Simon K.
4	Anish Athalye et. al.	Obfuscated Gradients Give a False Sense of Security: Circumventing Defenses to Adversarial Examples	ICML 2018	https://arxiv.org/pdf/1802.00420.pdf	Magda	Volodymyr F.
5	Xingjun Ma et. al.	Characterizing Adversarial Subspaces Using Local Intrinsic Dimensionality	ICLR 2018	https://arxiv.org/pdf/1801.02613.pdf
6	Zelun Luo et. al.	Label Efficient Learning of Transferable Representations across Domains and Tasks	NIPS 2017	https://papers.nips.cc/paper/6621-label-efficient-learning-of-transferable-representations-acrosss-domains-and-tasks.pdf	Mahdi	Jonas B.
7	Xiuming Zhang et. al.	Learning to Reconstruct Shapes from Unseen Classes	NIPS 2018	https://papers.nips.cc/paper/7494-learning-to-reconstruct-shapes-from-unseen-classes.pdf	Leslie	Sergey M.
8	Jiaxuan You et. al.	GraphRNN?: Generating Realistic Graphs with Deep Auto-regressive Models	ICML 2018	http://proceedings.mlr.press/v80/you18a/you18a.pdf	Helisa	Paul M.
9	Ting-Chun Wang et. al.	High-Resolution Image Synthesis and Semantic Manipulation with Conditional GANs	CVPR 2018	http://openaccess.thecvf.com/content_cvpr_2018/papers/Wang_High-Resolution_Image_Synthesis_CVPR_2018_paper.pdf	Helisa	Felix A.
10	Tameem Adel et. al.	Discovering Interpretable Representations for Both Deep Generative and Discriminative Models	ICML 2018	http://proceedings.mlr.press/v80/adel18a/adel18a.pdf	Ashkan	Lukas F.
11	Carlos Florensa et. al.	Automatic Goal Generation for Reinforcement Learning Agents	ICML 2018	http://proceedings.mlr.press/v80/florensa18a/florensa18a.pdf	Oliver	Dominik U.
12	Augustus Odena et. al.	Is Generator Conditioning Causally Related to GAN Performance?	ICML 2018	http://proceedings.mlr.press/v80/odena18a/odena18a.pdf	Oliver	Alessa G.
13	Rithesh Kumar et. al.	Maximum Entropy Generators for Energy-Based Models	arXiv 2019	https://arxiv.org/pdf/1901.08508.pdf	Azade	Shabnam S.
14	Chen et. al.	InfoGAN?: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets	NIPS 2016	https://arxiv.org/pdf/1606.03657.pdf	Ashkan	Hossain S.
15	Dilokthanakul et. al.	Deep Unsupervised Clustering With Gaussian Mixture Variational Autoencoders	arXiv 2017	https://arxiv.org/pdf/1611.02648.pdf
16	Frosst et. al.	Analyzing and Improving Representations with the Soft Nearest Neighbor Loss	arXiv 2019	https://arxiv.org/pdf/1902.01889.pdf	Artem	Alexander S.

Schedule

The schedule for the paper presentations will be announced here during the semester. Each student is required to attend all presentations.

Date	Time	Title	Speaker	Material
31.01.2019	16:00 - 17:00	Preliminary meeting	Azade Farshad	Slides
24.04.2019	16:00 - 18:00	Introductory Lecture	Dr. Gustavo Carneiro	Slides
08.05.2019	16:00 - 18:00	GraphRNN?: Generating Realistic Graphs with Deep Auto-regressive Models Automatic Goal Generation for Reinforcement Learning Agents High-Resolution Image Synthesis and Semantic Manipulation with Conditional GANs Analyzing and Improving Representations with the Soft Nearest Neighbor Loss	Paul M. Dominik U. Felix A. Alexander S.
15.05.2019	16:00 - 18:00	Neural Discrete Representation Learning Maximum Entropy Generators for Energy-Based Models Is Generator Conditioning Causally Related to GAN Performance?	Faizan K. Shabnam S. Alessa G.
22.05.2019	16:00 - 18:00	Discovering Interpretable Representations for Both Deep Generative and Discriminative Models Obfuscated Gradients Give a False Sense of Security: Circumventing Defenses to Adversarial Examples InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets	Lukas F. Volodymyr F. Hossain S.
05.06.2019	16:00 - 18:00	Label Efficient Learning of Transferable Representations across Domains and Tasks Learning to Reconstruct Shapes from Unseen Classes	Jonas B. Sergey M.