March 24, 2024 Presentation Open Access

AI Platform for INFN Scientific Use Cases

Mauro Gattari

Talk given at the International Symposium on Grids & Clouds (ISGC), Taipei 24-29/03/2024

Abstract

Researchers at INFN (National Institute for Nuclear Physics) face challenges from basic to hard science
use cases (e.g., big-data latest generation experiments) in many areas: HEP (High Energy Physics),
Astrophysics, Quantum Computing, Genomics, etc.
Machine Learning (ML) adoption is ubiquitous in these areas, requiring researchers to solve problems
related to the specificity of applications (e.g., tailored models and intricate domain knowledge), but also
requiring solving general infrastructure-level and ML-workflow related problems.
As the demand for ML solutions continues to rise across the diverse research domains, there exists a
critical need for an innovative approach to accelerate ML adoption.
In this regard we propose an AI platform designed as an application-agnostic MLaaS (Machine Learning
as a Service) solution, which provides a paradigm shift by offering a flexible and generalized
infrastructure that decouples the ML development process from specific use cases.
The AI platform is implemented as a software layer on top of our cloud platform, the INFN Cloud, which
offers composable, scalable, and open-source solutions on a dedicated geographically distributed
infrastructure. The INFN Cloud core mission is to facilitate resource sharing and enhance accessibility for
INFN users, encompassing a wide range of resources, including GPUs and storage.
The AI platform leverages INFN Cloud resources and principles, gathering and orchestrating technologies
to support end-to-end scalable ML solutions: Kubernetes, Kubeflow, KServe, KNative, Kueue, Horovod,
etc., ensuring support for many ML frameworks: TensorFlow, PyTorch, Apache MXNet, XGBoost, etc.
This contribution will describe the platform’s design and principles, as well as some selected use cases
from NLP and HEP domains that benefitted from the “aaS” approach.
The platform's agnostic nature extends beyond model compatibility to address the practical challenges
associated with deploying ML solutions in real hard science scenarios: streaming services, exabyte-scale
storage solutions, high-bandwidth networking, support for native HEP data (e.g., CERN ROOT data
format), etc.
Furthermore, the AI platform promotes transfer learning and model reuse, to accelerate the ML
development lifecycle. Developers can leverage pre-trained models and share knowledge across
different applications, reducing the time and resources required for training new models from scratch.
This collaborative aspect not only enhances efficiency but also promotes a collective learning
environment within the research community.
In conclusion, the application-agnostic AI platform serves as a unified ecosystem where developers, data
scientists, and domain experts can collaborate seamlessly. By providing a standardized framework for
ML model development, training, and deployment, the platform eliminates the need for extensive
domain expertise in every application area. This democratization of ML empowers a broader audience
to leverage the benefits of machine learning, breaking down barriers and fostering innovation across
diverse research domains.