Registration

Welcome Address

• Paul Alexander (University of Cambridge)

WP-3 OBELICS overview

• Jayesh WAGH (Programme Coordinator-Obelics-Asterics)
• Giovanni Lamanna (LAPP/IN2P3/CNRS)

Panel discussion: OBELICS review and conclusion

Welcome Reception

Machine Learning applications in Gravitational Wave research

• Elena Cuoco

Noise of non-astrophysical origin contaminates science data taken by the gravitational-wave detectors. Characterization of instrumental and environmental noise has proven critical in identifying false positives in the first observing runs. In this context the application of different machine learning methods can help in achieving, for example, a fast classification of transient events to disentangle noise from gravitational signals helping a fast real time analysis. Moreover, these approaches could be used to disentangle Gravitational signals from noise. Deep Learning techniques could even be used to the aim of non linear noise cancellation to condition the data before any detection algorithm will be used and there are promising on-going studies on simulated data.

Time-domain Machine Learning - Opportunities and Challenges for the SKA

• Robert Lyon (University of Manchester)

To harness the discovery potential of data collected by the SKA, we require efficient and effective automated data processing methods. Machine learning tools have the potential to deliver this capability, as evidenced via their successful application to similar problems in the astronomy domain. This talk introduces the machine learning required for successful time-domain data processing (pulsar / transient discovery), and the infrastructure required to support it. Here the overriding aim is to increase awareness of what is required to facilitate the execution of automated learning methods, which we’ll need if we are to achieve the SKA's ambitious science goals.

Applications of Machine Learning to Deblending in LSST

• Fred Moolekamp

It is estimated that at least 63% of sources observed in the 10 year LSST survey will have at least 2% of their flux blended with another object. Achieving many of the LSST science goals requires a working deblender to separate the flux from overlapping stars and galaxies by extracting morphological and spectral data for each source. The primary focus of this talk will be on scarlet, the python package I have been developing with Peter Melchior that will soon be implemented in the HSC and LSST software pipelines. Time permitted I will also mention more ambitious efforts by other members of the Dark Energy Science Collaboration (DESC) using neural networks to solve the blending problem.

Machine learning in the Cherenkov Telescope Array

• Daniel Nieto

The sensitivity of ground-based gamma-ray telescopes based on the imaging atmospheric Cherenkov technique (IACTs) is driven by, among other factors, our ability to reconstruct the primary particles that originate the extended atmospheric showers that are imaged by the telescopes: this particle reconstruction enables us to classify gamma-ray events from the much more frequent background of cosmic-ray events. Supervised machine learning algorithms, like random forest or boosted decision trees, have been successfully applied to the task of event reconstruction by current generation IACTs, substantially improving their sensitivity. In this talk we will briefly review the state-of-the-art of machine-learning based event reconstruction for current-generation IACTs and will present an overview of the novel approaches, like deep learning, currently being explored for the Cherenkov Telescope Array, the next-generation gamma-ray observatory.

Machine learning in High Energy Physics

• Michael Kagan

High Energy Physics probes the mysteries of the universe using some of the worlds largest experiments and datasets. To interpret and analyze this data in search of new physical phenomena, a wide array of domain specific algorithms have been developed. At the same time, recent advances in deep learning have seen great success in the realms of computer vision, natural language processing, and broadly in data science. By connecting the challenges in the HEP domain with those in deep learning, new and powerful approaches to analyzing HEP data are being developed. In this talk, I will discuss developments in the application of machine learning techniques to the analysis and interpretation of High Energy Physics data, with a focus on the Large Hadron Collider.

Machine learning in KM3NeT

• Chiara De Sio (University of Salerno)

Machine learning in solar physics

• Sebastian Hoch

Bayesian statistics

• Will Handley

A summary of the workshop "AI at CERN and SKA"

• Vlad Stolyarov

The two-day workshop was held in the Alan Turing Institute in London (17-18.09.2018) where ATI staff members, AI researches from industry as well as CERN and SKA scientists discussed the application of Artificial Intelligence (AI) and Machine Learning (ML) for scientific discovery in High Energy Physics, astrophysics, cosmology and radio astronomy. An overview of the presentations will be given in this summary talk.

Social Dinner

DATA LAKE - WLCG computing model evolution

• Ian Bird (CERN)

EOSC overview and paths

• Brian Matthews

ASTERICS-OBELICS software and services contribution to EOSC needs

• Giovanni Lamanna (LAPP/IN2P3/CNRS)

EOSC-hub: project status and plans, presentation of service catalogue

• Matthew Viljoen (EGI Foundation)
• Tiziana Ferrari (EGI Foundation)

Coffee Break

Rules of participation (state of play) for service providers: EOSCpilot outcome and EOSC-hub experience

• Matthew Viljoen (EGI Foundation)

Interactive group discussion - Shaping the European Open Science Cloud service roadmap

• Bjorn Backeberg
• J B Raymond Oonk
• Matthew Viljoen (EGI Foundation)
• Tiziana Ferrari (EGI Foundation)

During this session representatives from the European Open Science Cloud (EOSC) will introduce EOSC and provide an overview of existing services and then engage with the Asterics meeting participants in an interactive world cafe-style session on prospective future service provisioning, access models and concrete services that the RI would like to bring into EOSC. EOSC is an ambitious initiative aiming at the federation of existing and planned digital infrastructures for research. It seeks to remove barriers among disciplines and countries and make it easier for researchers to share and access the digital resources they need. Looking forward to the evolution of EOSC, it is essential to understand and prioritise what services are most needed and that should be added to the future EOSC service portfolio and importantly, what criteria are to be used for uptake in the service portfolio. The goal of this session is to take advantage of the collective knowledge of the audience to extract high-level needs for services and identify priorities for the coming years to develop a service roadmap. The outcome of the session will help us draw a more detailed picture of the European Open Science Cloud roadmap, and the services that ESFRI RIs, such as Asterics, would need and bring into it.

Lunch Break

H2020-HNSCiCloud project update and future directions

• Tony Wildish

H2020 AENEAS results and objectives

• Rob van der Meer (ASTRON)

Coffee Break

Panel Discussion