Most of the research carried out at IP2I relies on large-scale data processing and intensive computing systems, and has done so for many years.

At CERN, for example, in the 1960s the analysis of bubble-chamber photographs required large teams of operators. Later, with the rise of computing and distributed infrastructures, visual and manual analyses were replaced by software running on computing farms such as the CC-IN2P3 in Lyon, or on the Worldwide LHC Computing Grid, which processes data from experiments such as ALICE and CMS.

Today, high-performance computing (HPC), deep learning and big data technologies offer new opportunities across fundamental research. For example:

• HPC and GPU-accelerated computing make it possible to solve the complex differential equations used by theoretical physics teams, for example to describe neutron-star and black-hole mergers, or to model the behaviour of galaxies.
• Deep learning is used to separate signal from background in experiments, whether for the detection of gravitational waves (Virgo/LIGO), searches for as-yet unknown particles (CMS), or particle identification from the signals recorded in detectors (AGATA).
• Big-data techniques inspired by those developed by major internet companies enable interactive analysis of very large datasets (Euclid, CMS).

The aim of the IP2I Scientific Computing Platform is:

• To monitor technological developments so that IP2I can keep pace with worldwide advances and remain at the forefront of these fields.
• To anticipate and provide the computing resources required for the laboratory’s activities (servers, GPUs, etc.).
• To promote knowledge sharing and strengthen links between the laboratory’s teams, as well as with partner organisations, in particular IN2P3/CNRS, UCBL and Université de Lyon.
  • Documentation on the use of IP2I and CC-IN2P3 computing resources.

ACTIVITIES:

• Regular tutorials and discussions on:
  • machine learning and deep learning
  • data science
  • Dask, Kubernetes and Slurm
  • Python and Git

2020

• Deployment of a computing farm with more than 900 CPU cores
• Acquisition of a GPU server equipped with three NVIDIA RTX 6000 cards for prototyping deep-learning algorithms
• Support for research projects:
  • Automatic detection of micro-defects on the surface of mirrors produced by LMA for Virgo
  • Calculation of quantum nuclear wave functions using a neural network
• Machine Learning and Deep Learning training for researchers and faculty members from Université de Lyon, with more than 30 participants
• Creation of The Data Frog, a public-facing blog introducing data science and machine learning with Python.