HSF Weekly Meeting #188, 18 June 2020

Present/Contributors: Graeme Stewart, Michel Jouvin, Lukas Heinrich, Benedikt Hegner, Stefan Roiser, Agnieszka Dziurda, Kyle Knoepfel, Caterina Doglioni, Pere Mato, Witek Pokorski, Paul Laycock, Eduardo Rodrigues, Sam Meehan, Antoine Pérus, Chris Jones, Giles Strong, Liz Sexton-Kennedy, Philippe Canal, Sudhir Malik, Daniel Elvira, Vassil Vassilev, Kilian Lieret

News, general matters

Proposal on Differentiable Analysis / Computing

Over the last decade “Deep Learning” has been one of the driving paradigms in Machine Learning and its application in High Energy Physics data preparation and analysis. Neural network architectures have evolved significantly into a growing collection of re-usable building blocks such as recurrent, stateful, or attention units, which can then be freely composed into larger networks that can then be jointly optimized. Optimization often occurs through gradient-based methods, such as backward propagation; the differentiability of such modules is an important characteristic. Based on that, a new paradigm of “differentiable programming” is emerging that utilizes automatic differentiation techniques to formulate optimizable algorithms, with common elements such as control structures and loops. In high-energy physics, activity in this area is gaining increasing consideration based on techniques such as INFERNO, neos, MadMiner, and others. This activity area seeks to provide a forum for research and discussion on techniques and software to implement HEP-specific differentiable building analysis blocks.

Application of these techniques is wider than just analysis, e.g., error propagation in reconstruction.

Mattermost (everyone welcome) to discuss projects:

(this does work for non-CERN people, but you might need to create a lightweight account).

Meeting approved this a new Activity for the HSF, Differentiable Computing. Use the HSF platform to look for cross-cutting topics with, e.g., Analysis and PyHEP WGs.

Technical Note on Best Practices for HSF Projects

Current Draft:

This has had a revamp and update and we would very much like to finish it now.

Comments and suggestions (and PRs!) before the next meeting please!

Project Template Tool

Reminder that we have an HSF project template tool that helps setup some of these best practice points.

If you think of a good way to improve this, please implement it and make a pull request.

Computing and Software for Big Science

One of the editors of the journal contacted us to ask if we would make a call for papers to the journal, which would help them reach their threshold for 2020 (important for the journal being indexed). Papers need to be accepted by November 2020.

IRIS-HEP Blueprint Meeting on Sustainable Software in HEP

July 22: https://indico.cern.ch/event/930127/

Crucial issue for HSF and HEP. Nice set of talks being arranged from other communities, e.g., AstroPy and Carpentries, as well as people inside our community.

LHCC Review of HL-LHC Software and Computing

There are some preliminary conclusions from the review as part of the LHCC minutes. More statements about event generators than any other aspect. They are generally positive about the role the HSF has played.

We do expect more details later from the panel.

Alpaka Training

Dark Matter Test Science Project (ESCAPE) updates

Caterina, Graeme, Xavier, Pere, Tommaso (and others, including Lukas Heinrich and Sam Meehan on the ATLAS side) involved in discussions to make the “ESCAPE Dark Matter Test Science Project” (=review and implementation of software pipelines for different dark matter experiments) part of a proposal for the European Open Science Cloud INFRAEOSC-03-2020.

Working Group Updates

Data Analysis

Detector Simulation

Reconstruction and Software Triggers


Software Training



New Architectures, Portability, and Sustainability

Feedback meeting was yesterday (https://indico.cern.ch/event/925974/), but was not very well attended (~20 people)

Good participation in the survey: 34% of the registered people (75 out of 221)

Next WLCG/HSF event: no possibility for a F2F one this year, date foreseen in September are too early for another virtual workshop

Lund is all “unbooked”, but they will welcome us if we want to come back in 2021!


Website Update

Significant clean-up of broken and outdated links was done on the website. We are almost back to a clean bill of health, except for some strange issues to do with DESY’s Indico.

As part of this cleanup we deprecated the old HSF Knowledge Base

Indico -> HSF Calendar Sync

Significant progress in automating the synchronisation of HSF Indico Events to the HSF Google Calendar thanks to Stefan’s efforts:

Next Meeting