Distributed Computing Library (dislib)

A classical machine learning library developed on top of PyCOMPSs.

Institution:

Technology

Research Group:

Workflows and Distributed Computing

Researcher/s:

Rosa Maria Badia, Eduardo Iraola, Fernando Vázquez Novoa, Francesc Lordan, Raül Sirvent, Javier Conejero

Description:

dislib wraps classical machine learning and deep learning algorithms in a distributed array abstraction; at runtime, PyCOMPSs partitions data and dispatches tasks over heterogeneous HPC or cloud resources, transparently handling datasets that exceed per-node memory budgets. This delivers the “write once, run anywhere” portability highlighted by the programming-model trend of HPC.

Value Proposition:

Parallel ML library that stretches Python workloads to full clusters.

Aplication areas:

Astrophysics analytics, molecular-dynamics workflows, urgent-computing for natural hazards, manufacturing digital-twins, large-scale feature engineering, distributed neural-network training.

Target market:

National & regional super-computing centres needing user-friendly ML stacks; Cloud-HPC providers bundling managed data-science services; Industrial R&D (automative, pharma, energy) building AI+simulation pipelines; Independent Software Vendors adding distributed ML to existing products.

Technology Readiness Level (1-9): 7

Protection:

Open source (apache v2)

More information

if you want to know more about this project do not hesitate to contact us

Contact us