FLH — Winter 2020/21 | Albarqouni Lab

Organizers: Dr. Shadi Albarqouni, Helmholtz AI and TU Munich, and Prof. Nassir Navab, TU Munich.

Announcements

- 17-12-2020: The deadline to submit the blog post is moved to 1st. Feb. 2021.

- 05-11-2020: Our seminar will be held online via Zoom. Password is communicated to the participants through TUMonline. - 30-09-2020: Your name has been assigned to one of our sessions. Please pin the date on your calendar.
- 30-09-2020: Two invited talks were just arranged. Thanks to the invited speakers!
- 21-09-2020: We are still working on the schedule.
- 21-09-2020: We extended the time slots of the presentations from 20 to 30 mins (See the requirements)
- 21-09-2020: Papers are assigned to the students according to their preferences (See the list of Topics and Material Table).
- 22-07-2020: Registration is done via the matching system. You need to send a motivation letter with the subject "FLH_Motivatoon" to Dr. Shadi Albarqouni in order to get a higher ranking in the matching system from our side
- 14-07-2020: Register in advance for this meeting here. After registering, you will receive a confirmation email containing information about joining the meeting.
- 08-07-2020: Preliminary meeting: ~~Thursday, 16.07.2020 (10:00-11:00)~~ Friday, 17.07.2020 (11:00-12:00) in virtual meeting room (zoom).
- 08-07-2020: Contact information-If you have any question about this seminar, please feel free to contact Dr. Shadi Albarqouni
- 08-07-2020: Website is up!

Introduction

Following the great success of our on-going seminar on Deep Learning for Medical Applications, we would like to discuss advanced topics that are quite relevant to Federated Learning which becomes an interesting and hot research direction in the community. In simple words, Federated Learning enables training models at the client-side while preserving their privacy, and aggregates the knowledge from the nodes to learn a global model. The interesting part here that the data are kept private and not transmitted to any other nodes. Instead, the characteristics (e.g. parameters) of the global model are shared with the clients, and once the training is done locally, the characteristics are sent back to the global one for aggregation. This learning paradigm has been received quite nicely in the community, in particular, for sensitive domains, e.g. Healthcare. To push this momentum, we proposed, together with our academia and industry partners, a workshop on Federated, Collaborative, and Distributed Learning in the International Conference on Medical Image Computing and Computer-Aided Intervention (MICCAI) to attract significant contributions attacking the challenges in Medical Imaging and Healthcare. In this seminar, we will be discussing the relevant papers on Federated Learning with an emphasis on the papers tackling the common challenges in Medical Imaging, e.g. data heterogeneity, domain shift, and non-iid distributed data.

Registration

Interested students should attend the preliminary 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: 12.

Requirements

In this Master Seminar, each student is asked to send three preferences from the list, then he will be assigned one paper. In order to successfully complete the seminar, participants have to fulfill these requirements:

Presentation: The selected paper is presented to the other participants.
Blog Post: A blog post of 1000-1500 words excluding references should be submitted before the deadline.
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. The blog post must include all references used and must be written completely in your own words. Copy and paste will not be tolerated. Both the blog post and presentation have to be written in English.

Submission Deadline: You have to submit both the presentation two weeks right after your presentation session. The deadline of the blog post is moved to 1st Feb. 2021.

Guidelines: I could not find better than this guidelines to prepare for your presentation. The only difference is that you need to plan for 30 minutes for 1-4, and 10 minutes for 5). Nevertheless, I have prepared a few slides acting as a guidelines for your presentation and blog posts.

Schedule

Date	Session: Topic	Speakers / Presenters	Slides
17.07.2020 (11:00 AM)	Preliminary Meeting		Slides
Online	Paper Assignment		Guidelines
09.11.2020 @ 11:00 AM	Federated Learning; Challenges, Methods, and Future I	Invited Talk: Federated Learning: Collaborative AI without Exposing Patient Data –Nicola Rieke from NVIDIA
16.11.2020	Federated Learning; Challenges, Methods, and Future II	Ünay, Sánchez Clemente
23.11.2020	Data Heterogeneity I	Lin, Raether
30.11.2020	Data Heterogeneity II	Stoican, Schwarz
07.12.2020	System Heterogeneity and Privacy Issues I	Invited Talk: Secure, privacy-preserving and federated machine learning in medical imaging –George Kaissis from Klinikum rechts der Isar
14.12.2020	System Heterogeneity and Privacy Issues II	Qian, Heidmann
21.12.2020	Data Heterogeneity III	Spannagl, Arfaoui
11.01.2021	Federated Learning with Medical Imaging I	Guest speaker (NeurIPS'20 Paper), Cosmin (MICCAIW'20 Paper)
18.01.2021	Federated Learning with Medical Imaging II	Boysen
25.01.2021	Federated Learning with Medical Imaging III	Invited Talk: Federated Learning with Heterogeneous data in Healthcare –Jean Ogier du Terrail from Owkin

List of Topics and Papers

Topic	No	Title	Conference/Journal	Tutor	Student (Last name)	Link
Intro. to FL	1	FedAvg: Communication-Efficient Learning of Deep Networks from Decentralized Data	AISTATS, 2016		Ünay	arXiv
	2	The Future of Digital Health with Federated Learning	arXiv, 2020		Invited Speaker	arXiv
Challenges	3	Federated Learning: Challenges, Methods, and Future Directions	IEEE Signal Processing Magazine, 2020		Sánchez Clemente	arXiv
	4	On the Convergence of FedAvg on Non-IID Data	ICLR 2020			PDF
Data Heterogeneity	5	FedMA: Federated Learning with Matched Averaging	ICLR 2020		Lin	PDF
	6	Federated Adversarial Domain Adaptation	ICLR 2020		Spannagl	PDF
	7	Federated optimization in heterogeneous networks	MLSys 2020		Raether	PDF
	8	FedAwS: Federated Learning with Only Positive Labels	ICML 2020			PDF
	9	SCAFFOLD: Stochastic Controlled Averaging for Federated Learning	ICML 2020		Stoican	PDF
	10	Federated Visual Classification with Real-World Data Distribution	CVPR 2020		Schwarz	PDF
System Heterogeneity	11	Federated Multi-Task Learning	NeurIPS 2017		Qian	PDF
	12	Variational Federated Multi-Task Learning	arXiv 2019		Heidmann	arXiv
Privacy-Issues	13	Secure, privacy-preserving and federated machine learning in medical imaging	Nature MI		Invited Speaker	HTML
	14	Differentially Private Meta-Learning	ICLR 2020			PDF
Explainability and Robustness	15	The Non-IID Data Quagmire of Decentralized Machine Learning	ICML 2020		Arfaoui	PDF
	16	DBA: Distributed Backdoor Attacks against Federated Learning	ICLR 2020			PDF
Open Problems in FL	–	Advances and Open Problems in Federated Learning	arXiv	–	–	PDF
Federated Learning with Medical Imaging	17	Privacy-preserving Federated Brain Tumour Segmentation	MICCAIW 2019		Boysen	HTML
		Multi-institutional Deep Learning Modeling Without Sharing Patient Data: A Feasibility Study on Brain Tumor Segmentation	MICCAIW 2019		Boysen	HTML
	18	Federated Learning in Distributed Medical Databases: Meta-Analysis of Large-Scale Subcortical Brain Data	ISBI 2019		Hofmann	HTML
		Inverse Distance Aggregation for Federated Learning with Non-IID Data	MICCAIW 2020		Hofmann	PDF