MLSysOps Hackathon: A Successful Two-Day Challenge

Over 60 students and mentors from across Europe came together for two days of collaboration at the MLSysOps Hackathon, using MLSysOps’ framework to design and apply different policies for cloud-edge continuum computing.

Red Team took the top prize as overall winner, while Purple Team impressed as Best Presenter. All 7 teams showcased creativity, technical skill, and collaborative problem-solving.

Read the full recap and see the highlights at the main Hackathon page!

READ MORE

MLSysOps @ the Workshop on AI Edge Cloud Computing Continuum (HaDEA)

On September 19, 2025, MLSysOps was invited to participate in the Workshop on AI Edge Cloud Computing Continuum organized by the European Health and Digital Executive Agency (HaDEA). The event brought together experts from EU-funded Horizon projects to discuss challenges and opportunities in deploying Artificial Intelligence across the cloud–edge continuum, with a particular focus on sustainability, efficiency, and future research directions.

Nikos Bellas of UTH represented MLSysOps and gave a five-minute presentation on the use of Machine Learning for Autonomic System Operation in the Heterogeneous Edge-Cloud Continuum. The talk described the project’s vision, ongoing research, and outcomes on ML-based resource optimization and application deployment/execution in heterogeneous Edge-Cloud environments.  He also participated in a panel of experts on Energy Efficiency & Sustainability, which explored ways to balance performance scalability with energy efficiency in AI-driven and data-intensive infrastructures, exploiting the synergy between hardware innovation and software optimization.

READ MORE

Software Framework Now Available

We are excited to announce that the MLSysOps Framework—the open-source outcome of the MLSysOps Project—is now available.

Join us for a hands-on session on July 18, 2025 where we will introduce the framework and the motivation behind its development. During the session, we will demonstrate how set up a testbed from scratch using provided scripts, deploy the system, and run a real-world example. Additionally, we will highlight our policy API and offer a sneak peek of how ML blends in. The session will conclude with a discussion on next steps and ways to get involved.

For more information about the live event as well as access to the framework, please visit our Software Framework page.

READ MORE

Award at DTW25 Ignite

NTT Data is proud to announce that the work of the Italian teams of Telco Innovation & EDGE Network Engineering and System engineering was awarded Best Moonshot Catalyst for TelcoMonetization at the DTW25-Ignite forum in Copenhagen!

Monetizing Federated Connectivity for Automotive OEMs” is a project focused on Telco monetization and distributed connectivity with Edge/MEC. It was developed in collaboration with leading partners from the telecommunications and automotive sectors and presented at the TM Forum’s flagship event. Participation in the MLSysOps Project contributed to shaping the proposed solution.

 

READ MORE

INRIA Shares Latest Advances at Recent and Upcoming Conferences

INRIA highlighted its latest research at two key conferences this June:

Aya Moheddine and Valeria Loscri presented their work on “Identifying and Exploiting a Denial-of-Service Vulnerability in the NGAP Protocol in 5G Networks at the EuCNC & 6G Summit which took place on 3-6 June at Poznan, Poland.

Jiali Xu and Valeria Loscrì highlighted their research on “Leveraging UE-Level Collaborative Intelligence for Scalable Jamming Detection in 5G Networks” at DISCOLI 2025 which took place on 9-11 Jun 2025, in Lucca, Italy, in conjunction with IEEE DCOSS-IoT 2025.

Looking ahead, INRIA is excited to announce its participation in the 20th International Conference on Availability, Reliability and Security (ARES) on Aug 2025 in Ghent, BE, where the team will present their latest work on SHIELD: Scalable and Holistic Evaluation Framework for ML-Based 5G Jamming Detection.”

READ MORE

MLSysOps @ DataWeek 2025

Christos Antonopoulos of UTH presented the MLSysOps Project at Data Week 2025 during the session “MLOps, Continuous Learning, and Resource Management in the Edge-Cloud Continuum“. His talk, titled “ML-based Autonomic System Management in the Edge-Cloud Continuum” emphasized the project’s contributions to ML-based automated resource management, resource-aware workload deployment, and the ML-assisted orchestration of complex application workflows.

Following the presentations, Christos Antonopoulos participated in a panel alongside experts from academia and industry. The discussion explored the practical challenges of integrating ML-driven methods into production environments, including the interfacing with legacy infrastructure, the availability and annotation of training data, and the deployment of resilient MLOps pipelines across the edge-cloud continuum. Christos contributed insights from the MLSysOps project’s ongoing efforts to build dependable and intelligent systems at the edge-cloud frontier, driven by the requirements of real-world applications.

 

READ MORE

Presentation at ICS

Excited to share that our paper, “TMModel: Modeling Texture Memory and Mobile GPU Performance to Accelerate DNN Computations,” will be presented by Jiexiong Guan (UTH,  University of William & Mary) at the International Conference of Supercomputing, which will take place in Salt Lake City, U.S.A, on June 8-11, 2025. This work is a collaboration between UTH, the University of William & Mary, and the University of Georgia.

Mainstream mobile GPUs (such as Qualcomm’s Adreno) usually have a 2.5D L1 texture cache that offers throughput superior to that of on-chip memory. However, to date, there is limited understanding of the performance features of such a 2.5D cache, which limits their optimization potential. TMModel introduces a novel performance modeling framework for mobile GPUs that combines micro-benchmarking, an analytical performance model, and a lightweight compiler to optimize DNN execution based on access patterns and GPU parameters. TMModel delivers up to 66× speedup for end-to-end on-device DNN training with significantly lower tuning cost than existing frameworks. As mobile devices grow more powerful, this work is a step towards efficient, real-time deep learning training directly on such devices.

READ MORE
Skip to content