Workshop on Machine-learning enabled safety-Critical systems (WMC)

Workshop goals

This workshop seeks to identify some central scientific issues that arise at the interface of machine learning and safety-critical real-time systems, that are of likely interest to the real-time computing community.

The objective is to work towards defining a research agenda that both

(i) builds upon the technical competencies of the real-time systems community to address problems that must be solved so as to enable the safe and effective use of technologies and components based on Deep Learning and related AI technologies; and

(ii) looks to machine learning to further enrich the technical competencies of the real-time systems community.

We plan to form working sessions around the above topics. Each working session will have a keynote and a few (invited/contributed) short talks, followed by open discussions. Through these discussions and the collection of viewpoints, the workshop contributes to the on-going efforts of the real-time systems community on forming such a research agenda. 

Our workshop builds upon the initiative of the ML-RT-Agenda workshop at ECRTS’2024.

Rationale

Machine learning, deep learning, and other AI-related technologies have significantly influenced safety-critical real-time systems in recent years. Their impact is not only evident in the application layer of a real-time system, such as object tracking, avoidance, and navigation, but also throughout the system’s lifecycle, where they aid in the design, monitoring, maintenance, and diagnosis of these systems.

This evolution has been instrumental in the recent research developments within our community, leading to a surge of new challenges and research problems. Consequently, we believe it is an opportune moment to invite the community to contribute to the formation of a research agenda. This agenda aims to highlight past work and provide a vision for future developments in this rapidly expanding field.

To facilitate this, we welcome everyone interested in sharing their perspectives on future research at the intersection of machine learning and real-time safety-critical systems. We will also extend invitations to those who have dedicated substantial time to this field and can share their insights through invited talks and keynotes.

Recognizing that the development of such a research agenda requires open dialogue and the involvement of all interested parties, we plan to allocate significant time during the workshop for in-depth discussions, community building, and opinion gathering. 

Workshop format

A diverse set of activities and contributions are planned including invited keynotes, short talks, working sessions, and open discussions. To make the best of the day, we will primarily focus on two research themes (corresponding to the two enumerated objectives above):

1. How can the expertise that is available in the real-time computing community help solve problems that arise when Learning-Enabled Components are used in safety-critical real-time systems?

2. How can recent developments in Machine Learning be adapted to further enrich the repertoire of techniques that are currently in use in the real-time systems community?

and form working sessions around them. Each working session will have a keynote and a few (invited/ contributed) short talks, followed by open discussions.

Organizers

• Mitra Nasri, Eindhoven University of Technology (TU/e), The Netherlands (m.nasri@tue.nl) 

• Sanjoy Baruah, Washington University in Saint Louis (WashU), USA (baruah@wustl.edu)

Our speakers

We are pleased to announce that Prof. Nicole Megow (University of Bremen, GE) will give a keynote speech at our workshop.

Program

Session 1 (9:00 to 10:00)

• Welcome to WMC 2024 (slides)

• Keynote by Nicole Megow  on "Learning-Augmented Scheduling with Provable Performance Guarantees" (slides)

Session 2 (10:00 to 10:30)

• Hyosung Kim on "Real-Time Scheduling for the AI Era: A Systems Perspective" (slides)

• Zhishan Guo on "When Machine Learning and Neural Networks Marry Real-Time Scheduling" (slides)

• Open discussions

Break (10:30 to 10:45)

Session 3 (10:45 to 12:00)

• Cong Liu and Zexin Li on "Building Robust, Timing-Predictable On-Device Machine Learning Systems" (slides)

• Mirco Theile on "Deep Reinforcement Learning for Real-Time Systems" (slides)

• Michael Yuhas and Arvind Easwaran on "Toward State-Aware Scheduling of Machine-Learning Workloads in Cyber-Physical Systems" (slides)

• Open discussions

Lunch (12:00 to 13:00)

Session 4 (13:00 to 15:15)

• Claire Paggetti  on "What's New in the ML-based System Aeronautical Certification" (slides)

• Yukikazu Nakamoto on "A Propose of Adaptable Real-time Embedded Systems using Unsupervised and Transfer Learning" (slides)

• Alessandro Biondi  on "Towards Safe and Secure Machine Learning for Cyber-Physical Systems" (slides)

• Open discussions (15 minutes)

• Heechul Yun on "Anytime Perception for Intelligent Cyber-Physical Systems" (slides)

• Arpan Gujarati  on "Working with NVIDIA Holoscan Middleware for Real-Time AI: Lessons Learned and Future Challenges" (slides)

• Joshua Bakita on "Unlocking Simple and Efficient Real-Time AI Inference on COTS Hardware via Better GPU Models" (slides)

• Open discussions

Break (15:15 to 15:30)

Session 5 (15:30 to 17:00)

• Sathish Gopalakrishnan on "Combining Statistical and Worst-Case Guarantees for Scheduling Policies" (slides)

• Jinkyu Lee on "Timing Guarantees for Inference of AI Models in Embedded Systems" (slides)

• Daniel Cassini on "AI for RT: Research Directions on Using AI to Design Embedded Real-Time Systems" (slides)

• Open discussions

• Conclusions and follow-ups (slides)

Intended outcome

The outcome of the workshop will be a report capturing a short summary of the event and discussions, the keynotes and talks, key challenges identified during the working sessions, and a draft of an initial research roadmap for the themes that are considered in the workshop.

Join our community

We plan to form a mailing list to keep touch with whomever is interested in the crossroad of machine learning and real-time systems. On the mailing list, we share our plans, future events, interesting achievements, and relevant news to the topic.

If you would like to join our community, please fill the following form. 

An expression of WMC'24

What an incredible experience at WMC'24! With over 60 attendees—some even standing in the hallway to catch the talks—it was a vibrant hub of discussions, ideas, and action points. Who knows, this might even lead to a Dagstuhl seminar! A huge thank you to all who participated so attentively and to our amazing speakers and keynote—this success was truly because of you! 

Previous editions of the workshop

The current workshop format has been adopted from the ML-RT-Agenda workshop at ECRTS'2024. 

Our workshop is a sequel of the WMC workshop initiated in 2013. Until 2023, the focus of the workshop was solely on mixed-criticality systems, but in 2024, the steering committee has decided to open up the floor to learning-enabled real-time systems. Therefore, the name has changed to the workshop for machine-learning safety-critical systems but the acronym has been retained (WMC) for continuity.

WMC ran for the first time at RTSS in 2013, and subsequently at RTSS in 2014, 2015, 2016, 2017, 2018, 2019, 2020, 2022, and 2023. The workshop was a resounding success on each occasion: there were over 30 participants that participated in lively and animated discussions throughout the day. In 2017 and 2018 there were more than 40 participants each time. In 2020, we hit 55+ participants given it was held in hybrid.

The websites of prior editions have the program, proceedings, and presentations:

• WMC2013: https://www-users.cs.york.ac.uk/~robdavis/wmc2013/

• WMC2014: http://www-users.cs.york.ac.uk/~robdavis/wmc2014/

• WMC2015: http://www-users.cs.york.ac.uk/~robdavis/wmc/

• WMC2016: https://gsathish.github.io/wmc2016/

• WMC2017: https://cps-research-group.github.io/WMC2017/

• WMC2018: https://sites.google.com/njit.edu/wmc2018/

• WMC2019*: https://sites.google.com/njit.edu/wmc2019 (*no presentation)

• WMC2020: http://2020.rtss.org/wmc2020/

• WMC2022: https://wmc2022.github.io/

• WMC2023: https://sites.google.com/view/wmc-2023/home

The formal steering committee for WMC also organized a Dagstuhl Seminar on Mixed-Criticality Systems in March 2015: http://www.dagstuhl.de/en/program/calendar/semhp/?semnr=15121. This prestigious event was hugely successful and shows there is considerable interest in MCS across a broad spectrum of real-time systems research. A second successful event in this series took place in March 2017: http://www.dagstuhl.de/en/program/calendar/semhp/?semnr=17131.

Steering committee of the WMC workshop

• Iain Bate, Real-Time Systems Research Group, University of York, UK (iain.bate@york.ac.uk)

• Arvind Easwaran, College of Computing and Data Science, Nanyang Technological University, Singapore (arvinde@ntu.edu.sg)

• Zhishan Guo, Department of Computer Science, North Carolina State University, US (zguo32@ncsu.edu)

• Jing Li, Department of Computer Science, New Jersey Institute of Technology, US (jingli@njit.edu)