• Angelegt von Unbekannter Benutzer (ga74few), zuletzt geändert am 14. Juni 2021

Instructors: Prof. Dr. Nassir Navab, Anees Kazi, Roger Soberanis, Mahsa Ghorbani

Announcements

  • 26.04.2021 - Access details for the course are available in the course Moodle.
  • 12.04.2021 - Assignments have been released! Check the date for your presentation and contact your tutor on time.
  • 06.04.2021 - [Optional] Let us know your four preferred papers by email before Fri. 09.04
  • 06.04.2021 - List of topics has been published. 
  • 12.02.2021 - Registrations are open from 11.02.2021 to 16.02.2021 through the TUM Matching Platform. Additionally, let us know your interest through our application form.
  • 03.02.2021 - Students interested in participating in the seminar should attend the preliminary meeting on Feb 08, 2021. 4:00 pm. Due to the pandemic situation, the meeting will be online. 
  • 03.02.2021 - Course Wiki is up!
  • 02.02.2021 - Contact information. If you have any questions about the seminar, feel free to contact Roger Soberanis (roger.soberanis@tum.de) or Anees Kazi (anees.kazi@tum.de)

Introduction 

Graph Deep Learning is a new exotic branch in many fields like computer Vision and Medical Imaging. Many real-world medical and non-medical datasets can be represented in the form of graphs, providing a powerful source of information for machine learning models. This graph-based data, combined with the success of the convolutional neural networks, has motivated to translate the key ingredients of deep learning models into the graph domain. Many communities such as healthcare, social media, and computer vision are moving towards analyzing the data using Graph Convolutions. This seminar provides a space for discussion of the recent scientific publications on GCN with a focus on their medical applications and others.

Registration 

  • Interested students should attend the introductory meeting to enlist in the course.
  • Students can only register through TUM Matching Platform themselves if the maximum number of participants hasn't been reached (please pay attention to the Deadlines).
  • A maximum number of participants: 20.

Requirements and Course Structure

For this course, previous background in Machine/Deep Learning is required. 

During the course, students will select and present a scientific publication from the given list. Additionally, students will also contribute to the course blog post with their own summary of the selected article. Requirements are described in the image below: 

List of Topics


TopicNumberTitleJournal/ConferenceTutorStudentLink
Medical
Applications		1Multi-View Graph Convolutional Network and Its Applications on Neuroimage Analysis for Parkinson’s DiseaseAMIA Annu Symp Proc. 2018ShahroozCizevskijhttps://www.ncbi.nlm.nih.gov/pmc/articles/PMC6371363/

2Cgc-net: Cell graph convolutional network for grading of colorectal cancer histology imagesICCV 2019ArioHayranhttp://openaccess.thecvf.com/content_ICCVW_2019/html/VRMI/Zhou_CGC-Net_Cell_Graph_Convolutional_Network_for_Grading_of_Colorectal_Cancer_ICCVW_2019_paper.html

3Uncertainty-based Graph Convolutional Networks for Organ Segmentation RefinementMIDL 2020TimoKhattabhttp://proceedings.mlr.press/v121/soberanis-mukul20a/soberanis-mukul20a.pdf

4Latent-Graph Learning for Disease PredictionMICCAI 2020TimoCuturilohttps://link.springer.com/chapter/10.1007/978-3-030-59713-9_62

5A joint 3D UNet-Graph Neural Network-based method for Airway Segmentation from chest CTMLMI 2019RogerGrafhttps://www.springerprofessional.de/en/a-joint-3d-unet-graph-neural-network-based-method-for-airway-seg/17256006
Node-level
Applications		6Graph Random Neural Networks for Semi-Supervised Learning on GraphsNeurIPS 2020MahsaImdahlhttps://paperswithcode.com/paper/graph-random-neural-network

7Recovering Brain Structural Connectivity from Functional Connectivity via Multi-GCN Based Generative Adversarial NetworkMICCAI 2020RogerIvarssonhttps://papers.nips.cc/paper/2020/file/968c9b4f09cbb7d7925f38aea3484111-Paper.pdf

Graph-level

Applications

8Scalable Deep Generative Modeling for Sparse GraphsICML 2020Mahsa
https://paperswithcode.com/paper/scalable-deep-generative-modeling-for-sparse

9Diffusion Improves Graph LearningNeurIPS 2019TBA
https://paperswithcode.com/paper/diffusion-improves-graph-learning-1

10Temporal Graph Networks For Deep Learning on Dynamic Graphs
Shun-Cheng WuBani-Harounihttps://arxiv.org/pdf/2006.10637.pdf

11Graph Meta Learning via Local SubgraphsNeurIPS 2020AzadeKayalihttps://arxiv.org/pdf/2006.07889.pdf

12Learning Graph Embeddings for Compositional Zero-shot LearningCVPR 2021AzadeGabsihttps://arxiv.org/pdf/2102.01987.pdf

Interpretability, Robustness, and Improving

13Simplifying Graph Convolutional NetworksICML 2019AneesMüllerhttp://proceedings.mlr.press/v97/wu19e.html

14Interpreting Graph Neural Networks For NLP With Differentiable Edge MaskingICLR 2021AneesKinihttps://openreview.net/forum?id=WznmQa42ZAx

15On The Bottleneck of Graph Neural Networks and Its Practical ImplicationsICLR 2021AshkanKistolhttps://openreview.net/forum?id=i80OPhOCVH2

16Attacking Graph Convolutional Networks via RewiringICLR 2021MehrdadKiesgenhttps://openreview.net/forum?id=B1eXygBFPH

17Distilling Knowledge from Graph Convolutional Networks
MehrdadKadrihttps://arxiv.org/pdf/2003.10477.pdf

Schedule

Sessions will take place Tuesdays from 12:00 to 14:00 hrs, in a virtual zoom meeting. A group of Two to Three students will present their papers during each session. 

The list of papers will be published during the second half of March. You will receive an email for selecting your preferences. The list of assigned papers is planned to be available early in April.

On 27.04.2021, there will be an introductory lecture on Graph Deep Learning by the organizers. To keep a fair preparation time after this introductory session, the first group will have one week to finish their blogpost and two weeks for preparing their presentation after the introductory lecture. 

Topics for each session to be announced.

Course Schedule (Tentative):

DateTimePlaceTopicAdditional Info
08.02.202116:00-16:30Zoom MeetingPreliminary Meeting

Slides

05.04 - 09.04 2021

Papers assignments to be released (contact your corresponding tutor once you know your paper)
27.04.202112:00 - 14:00Zoom MeetingIntro to Graph Deep Learning

Slides

04.05.2021



11.05.2021



18.05.202112:00 - 14:00Zoom Meeting

Presentations I:

  • Multi-View Graph Convolutional Network and Its Applications on Neuroimage Analysis for Parkinson’s Disease
  • Cgc-net: Cell graph convolutional network for grading of colorectal cancer histology images 
  • Uncertainty-based Graph Convolutional Networks for Organ Segmentation Refinement

25.05.202112:00 - 14:00Zoom Meeting

Presentations II:

  • Latent-Graph Learning for Disease Prediction
  • A joint 3D UNet-Graph Neural Network-based method for Airway Segmentation from chest CT
  • Graph Random Neural Networks for Semi-Supervised Learning on Graphs

01.06.202112:00 - 14:00Zoom Meeting

Presentations III:

  • Interpreting Graph Neural Networks For NLP With Differentiable Edge Masking
  • Simplifying Graph Convolutional Networks

08.06.202112:00 - 14:00Zoom Meeting

Presentations IV:

  • Temporal Graph Networks For Deep Learning on Dynamic Graphs
  • Graph Meta Learning via Local Subgraphs
  • Recovering Brain Structural Connectivity from Functional Connectivity via Multi-GCN Based Generative Adversarial Network

15.06.202112:00 - 14:00Zoom Meeting

Presentations V:

  • Learning Graph Embeddings for Compositional Zero-shot Learning
  • Distilling Knowledge from Graph Convolutional Networks
  • On The Bottleneck of Graph Neural Networks and Its Practical Implications

22.06.202112:00 - 14:00Zoom MeetingPresentations VI
08.02.2021



08.02.2021




Resources and Material

An introductory lecture to GCNs by Prof. Xavier Bresson on Alfredo Canziani Youtube channel:


If you find more interesting material you would like to share in this section, feel free to contact us. 

  • Keine Stichwörter