April 24, 2025

We are thrilled to announce that we have successfully deployed our first transatlantic testbed for AI inference based on the OSCAR serverless platform, thanks to the integration of the  Infrastructure Manager (IM) with Chameleon Cloud, a USA-based configurable experimental environment for large-scale edge to cloud research, managed by the University of Chicago.

This marks a significant milestone in AI4EOSC’s mission to facilitate large-scale resource orchestration for AI-based scientific computing. This integration demonstrates our ability to manage and deploy resources on Chameleon Cloud, leveraging IM’s advanced automation and orchestration capabilities. We validated that our platform can efficiently interact with Chameleon Cloud’s OpenStack-based infrastructure, by deploying two Kubernetes-based OSCAR clusters (one in NCG-INGRID-PT, managed by LIP – Laboratory of Instrumentation and Experimental Particle Physics at Portugal, and another at KVM@TACC, managed by TACC – Texas Advanced Computer Center. 

Two OSCAR services were deployed based on the OBSEA Fish Detection algorithm, an AI-powered fish detection and classification model leveraging YOLO. The model has been fine-tuned to identify and classify fish in images captured by OBSEA, a cabled underwater observatory off the coast of Vilanova i la Geltrú, Spain, providing real-time oceanographic and environmental data for scientific research. This model has been trained using the AI4EOSC platform, based on the AI4OS open-source software stack, as part of the iMagine European project, which offers a portfolio of services for AI model development, training, and deployment, to be adopted by researchers in aquatic sciences. 

For the deployment of services in each cluster, an OSCAR service replica architecture has been used. This allows that AI inference requests made to the service deployed in OSCAR Cluster @EU can be offloaded to the corresponding service in the OSCAR Cluster @US in case of lack of available computing resources.  The execution results are consolidated back into the storage area of the corresponding service in the OSCAR Cluster @EU.

AI4EOSC will leverage this integration to enhance its AI-driven ecosystem for open science, ensuring robust and adaptable cloud-based AI inference workflows. Meanwhile, iMagine can gain improved access to computational resources, supporting its mission to advance imaging-based research in various scientific domains.

We are excited to share this milestone, as it strengthens AI4EOSC’s position as a key AI player, not only in the EOSC ecosystem, but across top-notch geographical distributed infrastructures thanks to the computing capabilities of the Chameleon Cloud. 

Acknowledgements

This work was supported by the project AI4EOSC ‘‘Artificial Intelligence for the European Open Science Cloud’’ that has received funding from the European Union’s Horizon Europe Research and Innovation Programme under Grant 101058593. Also, by the project iMagine ‘‘AI-based image data analysis tools for aquatic research’’ that has received funding from the European Union’s Horizon Europe Research and Innovation Programme under Grant 101058625. Results presented were obtained using the Chameleon testbed supported by the National Science Foundation. Also, Grant PID2020-113126RB-I00 funded by MICIU/AEI/10.13039/501100011033.