Lightnews — Scholar-powered news

Gregor Kasieczka @kasieczka.bsky.social · Feb 28

Two #Jobs with us at @uni-hamburg.de

1) *Permanent* staff position for ultrafast machine learning on FPGAs [1]

2) #PhD position for building & applying unsupervised ML to discover anomalies in data [2]

[1] www.uni-hamburg.de/stellenangeb...

[2] www.dashh.org/application/...

Ausschreibung

www.uni-hamburg.de

1

Gregor Kasieczka @kasieczka.bsky.social · Jan 29

Today we celebrate Erik Buhmann who just defended his excellent doctoral work on generative AI applied to detector simulation!

Congratulations Erik for this #EPiC [1] PhD!

[1] arxiv.org/abs/2301.08128

12

Reposted by Gregor Kasieczka

Steven Lowette @stevenlowette.bsky.social · Jan 28

Hey y'all US friends!
Did you see my postdoc ad to come and work with me on CMS? inspirehep.net/jobs/2864716
We usually get very few American applicants in EU unis (zero again, so far), but maybe at this moment in time it's a nice opportunity?
Deadline is flexible.

INSPIRE

inspirehep.net

6 13

Gregor Kasieczka @kasieczka.bsky.social · Jan 28

🎉🎉🎉 Many congratulations to Manuel for defending his PhD thesis on detecting anomalies in particle physics data!

He did a great job, not only on the CATHODE approach for anomaly detection based on weak supervision, but also applying it to data collected by the CMS experiment at CERN.

#PhDone

12

Reposted by Gregor Kasieczka

Freya Blekman @freyablekman.bsky.social · Jan 27

#CMSPaper 1366 examines a large sample of particle jets for behaviour inconsistent with the standard model, using ambitious unsupervised, weakly supervised and semi-supervised #machinelearning methods. They were usually more sensitive than 'classic' particle physics methods https://buff.ly/3CZwpmr

5 13

Gregor Kasieczka @kasieczka.bsky.social · Jan 17

Interview (in German) I recently gave to the FAZ Vor:Denker on AI, Physics, and how they do & will benefit each other.

vordenker.faz.net/protonen-hab...

»Protonen haben keine Rechte« | vor:denker

Wenn die Teilchenphysiker in der Nähe von Genf 40 Millionen Kollisionen pro Sekunde erzeugen, sind sich Physik und Künstliche Intelligenz ganz nah. Dann generiert der Teilchenbeschleuniger des Europäi...

vordenker.faz.net

1 3

Gregor Kasieczka @kasieczka.bsky.social · Jan 14

Abstract submission for #ACAT2025 is now OPEN!

This will be the 23rd ACAT and we have an exciting theme: "Transforming the Scientific Process: AI at the Heart of Theory, Experiment, and Computation in High-Energy and Nuclear Physics"

See you in Hamburg in September!

cern.ch/acat2025

ACAT 2025

23rd International Workshop on Advanced Computing and Analysis Techniques in Physics Research The 23rd International Workshop on Advanced Computing and Analysis Techniques in Physics Research (ACAT 20...

cern.ch

3

Gregor Kasieczka @kasieczka.bsky.social · Jan 14

Yes, please publish a new "open" dataset on a portal that requires login and only works if you are affiliated with the same university as the authors

a bald man is sitting in a chair with his hand on his chin and looking up .

ALT: a bald man is sitting in a chair with his hand on his chin and looking up .

media.tenor.com

6

Reposted by Gregor Kasieczka

Michael Krämer @mikraemer.bsky.social · Jan 10

With @henkderegt.bsky.social, @danielkostic.bsky.social, @kasieczka.bsky.social and others, we have submitted a white paper that provides a potential roadmap for the development and evaluation of physics-specific AI models: arxiv.org/abs/2501.05382, see also www.lorentzcenter.nl/physics-en-q....

Large Physics Models: Towards a collaborative approach with Large Language Models and Foundation Models

This paper explores ideas and provides a potential roadmap for the development and evaluation of physics-specific large-scale AI models, which we call Large Physics Models (LPMs). These models, based ...

arxiv.org

4 15

Gregor Kasieczka @kasieczka.bsky.social · Jan 6

We show that a hybrid (conserving+breaking) architecture combines the advantages of symmetry-conserving networks (fast learning) and breaking ones (higher maximum performance).

Great work by Seth with @aishikghosh.bsky.social, Ed, @danielwhiteson.bsky.social

Full paper: arxiv.org/abs/2412.18773

1 7

Gregor Kasieczka @kasieczka.bsky.social · Jan 6

Now(ish) on arxiv: Learning Broken Symmetries with Approximate Invariance

Invariances (or, as physicists call them: symmetries) of the data can be baked into a ML model to improve performance or data efficiency.

However, in reality, these symmetries are often broken.

How to deal with that?

1 2 8

Gregor Kasieczka @kasieczka.bsky.social · Dec 18

Realised with the very nice talk by Marie yesterday that I had never posted about our joint paper: "Accurate and robust methods for direct background estimation in resonant anomaly detection"

Check out arxiv.org/abs/2411.00085 for a new idea on how to extract anomalies!

3

Gregor Kasieczka @kasieczka.bsky.social · Dec 17

We prepared 180M jets from 2016 CMS data taking to be easily used for machine learning & demonstrate their use for pre-training.

Data at: www.fdr.uni-hamburg.de/record/16505
Paper at: arxiv.org/abs/2412.10504

w/ @ozamram.bsky.social, @joschkabirk.bsky.social, @mikraemer.bsky.social & others

Aspen Open Jets: a real-world ML-ready dataset for jet physics

This dataset contains approximately 180 M boosted jets, derived from open data collected by the CMS experiment at the Large Hadron Collider (LHC) in 2016 — specifically the JetHT datastream — and pres...

www.fdr.uni-hamburg.de

1 4

Gregor Kasieczka @kasieczka.bsky.social · Dec 17

New paper on #arxiv, inspired by discussions this summer at the Aspen Center for Physics

🤔 Foundation models need huge amounts of training data.
🤔 There now exist large volumes of #OpenData by e.g. @cmsexperiment.bsky.social
💡 Why not use that data for training?

1 1 3

Gregor Kasieczka @kasieczka.bsky.social · Dec 16

Greetings from our AI & Physics with friends workshop, this year hosted by Markus @kitkarlsruhe.bsky.social!

Besides great talks by Hamburg & other students, we also got to see the local KATRIN experiment which is trying to measure the neutrino mass

1 7

Gregor Kasieczka @kasieczka.bsky.social · Dec 12

Looking for a short & easy to understand (although in German) introduction to the connections between AI, Physics, & the recent Nobel price? Look no further than the new Physik Konkret!

With many thanks to DPG for inviting me to write about my favourite topics!

www.dpg-physik.de/veroeffentli...

Physiknobelpreis 2024: Physik in der Künstlichen Intelligenz

Ausgabe 73 | Dezember 2024 | „Konzepte aus der Physik haben maschinelles Lernen mitbegründet und leisten heute wichtige Beiträge, KI-Methoden weiter voranzutreiben. Gleichzeitig ist moderne Forschung...

www.dpg-physik.de

1 2

Gregor Kasieczka @kasieczka.bsky.social · Dec 10

Congratulations to Sebastian Bieringer who successfully defended his PhD today!

This was a *truly* interdisciplinary work [1] between particle physics, maths/statistics (thanks to excellent co-supervisor Mathias Trabs) and machine learning!

[1] inspirehep.net/authors/1906...

1 8

Gregor Kasieczka @kasieczka.bsky.social · Dec 10

Finally on #arxiv: Massive @cmsexperiment.bsky.social paper searching for anomalies in data: arxiv.org/abs/2412.03747

No new physics but it shows that ML-based anomaly detection methods can indeed be applied to collider data and offer a broad sensitivity to different potential signals!

2 3 10

Gregor Kasieczka @kasieczka.bsky.social · Nov 20

Wow, did not check in for a while and now this place is BUZZING 🐝🐝🐝

Expect this to be my primary place for posts in the future. Mostly on AI & Physics + the occasional terrible pun

1 10

Gregor Kasieczka @kasieczka.bsky.social · Oct 8

Got interviewed by German national TV today about the very inspired hashtag#NobelPrize decision.

Concepts from physics have been pivotal in developing machine learning & in turn we can use modern AI tools to do amazing physics.

Video here (in German): www.tagesschau.de/multimedia/v...

1 2

Gregor Kasieczka @kasieczka.bsky.social · Oct 2

Look at (&share!) this amazing poster for #ML4Jets this November in Paris

Gregor Kasieczka @kasieczka.bsky.social · Mar 7

Now at #DPG24: Our postdoc #LisaBenato with an awesome overview of searches for long-lived particles! Great to see over the last years the path from wild ideas to published results by @cmsexperiment.bsky.social @atlasexperiment.bsky.social @lhcb.bsky.social & more

indico.desy.de/event/42884/...

2

Gregor Kasieczka @kasieczka.bsky.social · Feb 5

A new #PhD position in my group at Uni Hamburg:
You get to build new anomaly detection tools (e.g. smarter versions of 2109.00546) & run them on data collected by the @cmsexperiment.bsky.social

Details & application at: www.uni-hamburg.de/stellenangeb...

Please consider sharing.

#AcademicJobs

1

Gregor Kasieczka @kasieczka.bsky.social · Dec 15

Had a great time teaching generative models at the COFI school in Puerto Rico

Don't usually post too many photos, but this might be a justified exception

Program & slides at: indico.cern.ch/event/1299889/

With many thanks to the organisers Kevin, Mayda, and Michael!

Gregor Kasieczka @kasieczka.bsky.social · Dec 5

Congratulations to Joerg who just successfully defended his PhD searching for very far-out particles 🎉🎉🎉

In case you want to know more, @cmsexperiment.bsky.social has you covered: cms.cern/news/searchi...

#PhDdone