ELLIOT & MINERVA Workshop on AI and HPC

Name: ELLIOT & MINERVA Workshop on AI and HPC
Start: 2026-03-13T11:00:00+01:00
Location: Auditorium of the Technopole, Modena

Date: 13 March 2026
Time: 11:00
Type: Workshop
Location: Auditorium of the Technopole, Modena
Status: Past event

On 13 March 2026 the Modena ELLIS Unit hosted the ELLIOT & MINERVA Workshop on AI and HPC, a public, hybrid workshop co-organised by the Horizon Europe projects ELLIOT and MINERVA. The workshop took place in the Auditorium of the Modena Technopole, alongside the 3rd plenary meeting of the ELLIOT project, which the Unit hosted at the new UNIMORE AI Center on 11–12 March.

The workshop followed the inauguration of the UNIMORE AI Center, held the same morning.

The workshop brought together researchers and practitioners working at the frontier of large-scale AI and high-performance computing (HPC). As foundation models and other large AI systems continue to scale, the field faces a growing set of methodological and computational challenges, from efficient training strategies and system-level constraints to reproducibility, performance optimisation and the responsible use of compute resources.

Topics

AI model scaling and practical strategies for training and evaluation at scale
Methodological challenges, including optimisation, robustness and reproducibility in large experiments
Computational aspects of large-scale AI, such as performance, efficiency and resource management
Future research directions at the intersection of advanced AI and HPC

Programme

11:00 – 11:30	Keynote: Document Understanding in the Era of Foundation ModelsDimosthenis Karatzas (CVC-CERCA)
11:30 – 11:50	Introduction to MINERVA servicesClaudio Domenico Arlandini (CINECA, Coordinator of the MINERVA project)
11:50 – 12:10	Any-to-any modeling of tens of diverse modalitiesAmir Zamir (EPFL), online
12:10 – 12:30	Cross-modal understanding and generation of multimodal contentNicu Sebe (University of Trento)
12:30 – 12:50	Advancing Physical AI: The AMD Open Ecosystem for AutonomyPier Luigi Dovesi (AMD / Silo AI)
12:50 – 13:30	Open discussion and future challenges
13:30 – 14:00	Lunch

Highlights

The workshop gathered more than 70 participants on site and over 40 online, from academia and industry. Across five sessions, the speakers discussed how AI is moving towards more integrated, scalable and real-world applications, from foundation models to physical AI.

Keynote: Document understanding in the era of foundation models. Dimosthenis Karatzas (CVC-CERCA) opened the workshop with a keynote on how foundation models are reshaping the analysis and understanding of documents.

Introduction to MINERVA services. Claudio Domenico Arlandini (CINECA) presented MINERVA as Europe’s support centre for scalable AI research and deployment, helping research teams bring their models to HPC environments. He described its services, from porting and optimisation to pre-training, fine-tuning and responsible AI, and its engagement formats, including embedded expert support for ambitious large-scale projects.

Any-to-any modeling of tens of diverse modalities. Amir Zamir (EPFL) presented any-to-any modelling, the ability to predict any modality from any other. Building on the 4M framework for massively multimodal masked modelling, covering RGB, depth, segmentation, captions, metadata and feature maps, he extended the idea to 3D, video and variable-length tokenisation.

Cross-modal understanding and generation of multimodal content. Nicu Sebe (University of Trento) showed how AI can move from passive analysis towards interactive and controllable systems, with work on learnable game engines, continual video customisation and bias in text-to-image models.

Advancing Physical AI: the AMD open ecosystem for autonomy. Pier Luigi Dovesi (AMD / Silo AI) outlined an open ecosystem for autonomy, robotics and simulation built on ROCm-native, GPU-agnostic development: synthetic data generation, perception data engines, world models and evaluation pipelines, co-developed with industry and academia.