CML-HIPEAC-CSW 2022 : Collaborative Machine Learning across IoT, Edge, Fog and Cloud Devices for Improved Privacy and Resilience
Call For Papers
CALL FOR CONTRIBUTIONS
Thematic Session on
"Collaborative Machine Learning across IoT, Edge, Fog and Cloud Devices
for Improved Privacy and Resilience"
Part of HiPEAC Computer System Week Spring 2022, Tampere
April 27, 2022
ABOUT THE SESSION
Machine Learning (ML) and Deep Learning (DL) techniques progressed tremendously
in the last decade and are now well understood, and widely used, but generally
as monoliths. Datasets are centralized prior to being used for training.
Similarly, inference input data is gathered at the monolith entrance, and
inference is generally performed at a single location. However, fundamentally,
both training and inference processes can be divided over multiple devices and/
or locations. The resulting Collaborative ML/DL can provide significant
benefits in terms of privacy (original data remains on the edge/IoT, no big
dataset collections) and resilience (a loss of one or more device or location
can be compensated for). At the same time, Collaborative ML/DL comes with many
additional operational constraints in terms of device maintenance, orchestration
and ML-OPS, in particular due to sharding. Finally, it is unclear whether
Collaborative ML/DL is an obstacle to or rather an enabler of scaling. Impacts
on energy efficiency are also to be assessed, as collaboration certainly induces
extra costs, but also opens room for data-movement optimizations.
This special session intends to gather current and possible future Collaborative
ML/DL practitioners, to share opinions on the viability of the approach, on the
benefits to be claimed in practice, and on the biggest challenges faced in the
One of the targeted objectives of the session to find answers to questions such
- Is federated learning only good at preserving privacy? Or has it other
advantages that could outweigh the complexity of the distribution?
- Should ML/DL models be designed specifically for a collaborative environment?
- How much compute time and energy can be saved through early exiting?
- How to leverage Collaborative ML/DL (or Federated Learning) for best resilience?
- Considering a possible vast amount of IoT devices, how to maintain efficiency
under scalability constraints?
- What are the most promising anticipated research directions?
CALL FOR CONTRIBUTIONS
For the thematic session at HiPEAC Computing System Week, we are seeking
presentations on the following topics:
- Federated ML - distribution of training or inference over multiple items
- Edge/cloud inference division - early exit, optimization
- Model breakup due to IoT hardware limitations
- Training/interference distribution costs modeling
- ML-ops in collaborative environments
- Collaborative machine learning at large - performing ML over multiple-locations
In order to perform a selection, interested contributors must submit either
- A one-page extended abstract detailing a research contribution
- A position paper of up to two pages describing how Collaborative ML is seen as
a challenge or an opportunity
Please send contributions in the form of plain text or pdf via email to the
organizers of the thematic session. Submissions will be reviewed by the TPC of
the thematic session, and accepted contributions will be given the opportunity
to present the proposed content during the thematic session. Furthermore, the
organizers plan to summarize the insights and outcome of this session in the
form of a joint position paper.
Given the short timeframe, anyone considering to contribute or attend the
session is strongly encouraged to contact us (co-organizers below) by email so
that we can orient him/her.
Submission deadline: March 24, 2022
Notification: April 1, 2022
Thematic session: April 27, 2022
Sébastien Rumley, iCoSys institute, HES-SO, Switzerland (email@example.com)
Holger Fröning, Heidelberg University, Germany (firstname.lastname@example.org)
TECHNICAL PROGRAM COMMITTEE
Jean-Frédéric Wagen (HES-SO, Switzerland)
Antonio di Maio (University of Bern, Switzerland)
Gregor Schiele (University of Duisburg-Essen, Germany)
Michaela Blott (XILINX/AMD, Ireland)
Laurent Lefevre (ENS-Lyon, France)
Hari Subramoni (Ohio State University, US)
Denis Trystram (LIG Grenoble, France)
Madeleine Glick (Columbia University, US)