New paper: Smart Data Management and ML-Based Workflow Prediction for Cognitive Compute Continuum in SPICE
- Conference paper
- Authors
- Michał Orzechowski, Piotr Kica, Mikołaj Zasada, Łukasz Opioła, Bartosz Kryza, Jakub Liput, Agnieszka Raczek, Jakub Karczewski, Wojciech Szmelich, Darin Nikolow, Łukasz Dutka, Bartosz Baliś, Renata G. Słota, Jacek Kitowski
- Published in
- KU KDM 2026 : eighteenth ACC Cyfronet AGH HPC users conference : Zakopane, 13–15 April 2026 : proceedings, p. 37-38, ISBN 978-83-61433-51-4
Summary
This paper presents an approach to smart data management and ML-based workflow prediction for the cognitive compute continuum developed within the SPICE project. The solution combines the Onedata platform with machine learning techniques to support efficient processing of large-scale, distributed data across heterogeneous edge, cloud, and HPC infrastructures. Onedata provides decentralized data access, trust-driven sharing, smart data movement, and semantic data discovery, while the ML-based pipeline predictor forecasts workloads and resource consumption to enable proactive runtime optimization. A cognitive mapper uses these predictions for adaptive scheduling, balancing throughput, cost, and data locality. The approach targets demanding industrial, energy, and medical use cases.