Overview

Many of today's parallel machine learning algorithms were developed for tightly coupled systems like computing clusters or clouds. However, data volumes generated from machine-to-machine or human-to-machine interaction, such as mobile phones or autonomous vehicles, surpass the amount that can conveniently be centralized. Thus, traditional cloud computing approaches are rendered infeasible.

In order to scale parallel machine learning to these amounts of data, computation needs to be pushed towards the edge, that is, towards the data generating devices. By learning models directly on the data sources - which often have computation power of their own, for example, mobile phones, smart sensors and tablets - network communication is reduced by orders of magnitude. Moreover, it facilitates obtaining a global model without centralizing privacy-sensitive data. This novel form of parallel, distributed, and federated machine learning has gained substantial interest in recent years, both from researchers and practitioners, and may allow for disruptive changes in areas such as smart assistants, machine learning on medical or industrial data, and autonomous driving.

This workshop is the third edition of the successful DMLE workshops at ECMLPKDD 2018 and 2019. We decided to rename the workshop to broaden the scope and integrate novel directions in the field, for example, privacy-preserving federated learning in the medical domain. The workshop aims to foster discussion, discovery, and dissemination of novel ideas and approaches for parallel, distributed, and federated machine learning.

Important Dates:

  • Submission deadline: 23 June 2020 9 June 2020
  • Acceptance notification: 9 July 2020
  • Camera-ready deadline: 7 August 2020 26 July 2020
  • Author registration: 31 July 2020
  • Video submission: 20 August 2020 (Virtual format of the workshop is explained below)
  • Workshop: 14 September 2020

Virtual Format

This year PDFL 2020 will be virtual as the conference ECML/PKDD is virtual given the pandemic situation. All the paper presentations will be pre-recorded. Workshop participants will be able to see the pre-recorded presentations asynchronously. We will also have a live streaming of the pre-recorded presentations on the day of the workshop which will be followed by a live Q&A via Zoom. We will make the schedule for the presentations, Q&A and invited talks available shortly.

At least one author should present the paper and must register for the workshop by July 31.

Instructions for recording and submitting your presentation:

  • DURATION: 15 mins (This is a strict time limit. Please do not go over!)
  • DEADLINE: 20 August 2020
  • MEDIUM: SlidesLive (You will receive recording instructions directly from slideslive after July 24)

Call for Papers

We invite participation in the 3rd Workshop on Parallel, Distributed, and Federated Learning, to be held as part of the ECMLPKDD 2020 conference. This year we invite two types of submissions to the workshop:

  • full length papers (16 pages)
  • short papers (8 pages)

For all accepted papers, we invite the authors for a presentation as a poster. Moreover, for 4-6 papers, we invite the authors for a presentation as a talk during the workshop. We issue a Best Paper Award with certificate and a prize. The main topics are, including, but not limited to:

  • Federated learning
  • Parallel machine learning
  • On-device machine learning
  • Edge computing for machine learning
  • Decentralized deep learning
  • In-situ methods
  • Communication-efficient learning
  • Privacy-preserving learning
  • Black-box machine learning
  • Distributed optimization
  • Theoretical investigations on parallelization
  • Large-scale machine learning, massive data sets
  • Distributed data mining
  • Fairness in Federated Learning
  • Distributed Training of Generative Models
  • Resource constraint machine learning
  • Hardware aspects of distributed learning

Submission Instructions:

Authors should submit a PDF version in Springer LNCS style using the workshop's EasyChair site. The review process is single-blind. Proceedings shall be submitted to Springer ECML PKDD 2020 Workshop Proceedings for publication. Full papers have a page limit of 16 pages, short papers have 8 pages, including bibliography.

Submitting a paper to the workshop means that if the paper is accepted at least one author commits to presenting it at the workshop. Papers not presented at the workshop will not be included in the proceedings.

Please note, that we do allow the submission of previously published work. However, such papers, when accepted, are excluded from the proceedings and when selecting papers for oral presentation, preference is given to original work.

Invited Talks

We are happy to announce that Peter Schlicht, Project Manager for Autonomous Driving at Volkswagen Group Research, and Professor Daniel Keren from Haifa University will give invited talks at the workshop.

Peter Schlicht
Peter Schlicht studied mathematics with a minor in computer science in Göttingen and received his doctorate in mathematics from the University of Leipzig. After a two-year research stay at the Ecole Polytechnique Fédérale (EPFL) in Lausanne (Switzerland), he joined Volkswagen Group Research in 2016 as an AI architect. There he deals with research questions on artificial intelligence technologies for automatic driving. He is particularly interested in distributed machine learning for autonomous driving, and methods for monitoring, explaining and robotizing deep neural networks, as well as securing them. Peter Schlicht is member of the Plattform Lernende Systeme working group "Mobilität und intelligente Verkehrssysteme" and project manager for AI-technologies for autonomous driving.
Daniel Keren
Daniel Keren (PhD 1991, Hebrew University in Jerusalem) is currently the chairman of the computer science department in Haifa University, Haifa, Israel. Prof. Keren's main fields of research are geometry and probability. Since 2003, he has been working closely with Prof. Assaf Schuster’s group in the Technion, in the area of distributed monitoring. His main contribution is in the mathematical aspects of the research such as object modeling, learning, optimization, and probability. A main novelty of the joint research is the incorporation of such mathematical tools into the research paradigm; this allowed to develop entirely new methodologies, based on geometry, to monitor general functions. He applied these methodologies to distributed stream monitoring, distributed online learning, and parallel machine learning. Prof. Keren’s goal is to continue developing the mathematical tools used so far, as well as to develop new ones, to improve and extend the applications of these tools to monitor, mine, and learn from large, distributed data sets.

Accepted Papers

  • Knowledge Discovery on Blockchains: Challenges and Opportunities for Distributed Event Detection under Constraints

    Thomas Liebig and Cedric Sanders

  • Resource-Constrained On-Device Learning by Dynamic Averaging

    Lukas Heppe, Michael Kamp, Linara Adilova, Nico Piatkowski, Danny Heinrich and Katharina Morik

Program

This workshop aims to provide interesting talks - both by expert invited speakers and great participants - and lively discussions. To achieve that, we have an invited talk with dedicated time for discussions. Moreover, we propose two sessions with talks for selected papers, again with time for discussions. We will issue a best paper award; the winner will receive a certificate and a prize. All accepted papers will be presented at a poster session in the end.

09:00 - 09:15 Opening Remarks, Yamuna Krishnamurthy
09:15 - 09:45 Invited Talk: Distributed Learning in Automated Driving, Peter Schlicht
09:45 - 10:00 Discussion
10:00 - 10:30 Knowledge Discovery on Blockchains: Challenges and Opportunities for Distributed Event Detection under Constraints, Thomas Liebig
10:30 - 11:00 Coffee Break
11:00 - 11:30 Invited Talk: Learning over Distributed Data Streams: Regression and Classification, Daniel Keren
11:30 - 11:45 Discussion
11:45 - 12:15 Resource-Constrained On-Device Learning by Dynamic Averaging, Lukas Heppe
12:15 - 12:30 Best Paper Award

Venue:

Please, read about the venue in the ECMLPKDD venues web page. You will find a description of the venue and a map.

Program Committee

We would like to thank our amazing program committee.

  • Stefan Wrobel, Fraunhofer IAIS
  • Katharina Morik, TU Dortmund
  • Tamas Horvath, University of Bonn
  • Jochen Garke, Fraunhofer SCAI
  • Mario Boley, Monash University
  • Christian Bauckhage, Fraunhofer IAIS
  • Janis Keuper, HS Offenburg
  • Mark Jelasity, University of Szeged
  • Dino Oglic, King’s College, London
  • Henning Petzka, Lund University
  • Michael Mock, Fraunhofer IAIS
  • Daniel Paurat, Deutsche Telekom Technik
  • Xiaoxiao Li, Yale University
  • Sven Giesselbach, Fraunhofer IAIS
  • Tim Wirtz, Fraunhofer IAIS
  • Pascal Welke, University of Bonn
  • Nico Piatkowski, Fraunhofer IAIS
  • Rafet Sifa, Fraunhofer IAIS
  • Dorina Weichert, Fraunhofer IAIS

Organizers

Linara Adilova

Linara
Adilova

Fraunhofer IAIS
Fraunhofer Center for Machine Learning

linara.adilova[at]iais.fraunhofer.de

Michael Kamp

Michael
Kamp

Monash University

info[at]michaelkamp.org