Sascha Caron

In search of new physics for 20 years. Use knowledge from dark matter, @ATLASexperiment, data and machine learning. + Private opinion.
@Radboud_uni and @_Nikhef

[Sascha Caron @SaschaCaron]

I am no longer on Twitter. Please use emails to reach me.

[Sascha Caron @SaschaCaron]

@freyablekman looks like joseph beuys 7000 eichen ..
https://t.co/N6cEfWQMrn

[arXiv astro-ph @_arXiv_astro_ph]

[2211.09796] Sascha Caron, Christopher Eckner, Luc Hendriks et al.: Mind the gap: The discrepancy between simulation and reality drives interpretations of the Galactic Center Excess https://t.co/QCXayuyYxZ https://t.co/4lh9Rm5Bgh #astro_ph_HE

[Sascha Caron @SaschaCaron]

https://t.co/Nx2PFsOxtf

Galactic Center GeV Excess (GCE):

We find that conclusions (dark matter or not, i.e. f_src=0 or 1, graph below) strongly depend on the type of simulation.

+ we find a gap between all models & reality.

Question: Does this apply to all works on the GCE?

[Sascha Caron @SaschaCaron]

For those interested in Dark Matter and the excess of gamma rays in the Galactic Center, I think tomorrow we will have an interesting article on arxiv. This work actually took 5 years...

[Sascha Caron @SaschaCaron]

Main work done by Rob Verheyen and Zhongyi Zhang. Thanks !

[Sascha Caron @SaschaCaron]

Try to find 4 top events at LHC ! We present a dataset to compare classifiers and comparisons of classifiers.

Best: transformers+particle net with pairwise interactions.

Compared to performance of baseline (BDT) the improvement is 30−50% in run time.

https://t.co/UE43EbCfsm

[Sascha Caron @SaschaCaron]

Tomorrow I give a talk about: "How machine learning and efficient computing can make data-intensive science more sustainable: examples and ideas from particle physics and astronomy". Would be cool to find collaborators (and yes, this all would need funding...)

[Sascha Caron @SaschaCaron]

Before our kids learn anything, they probably need to understand the proof why 1+1 = 2...

https://t.co/IqXxQplELP

[Ivo van Vulpen @IvovanVulpen]

Nice talk by @SaschaCaron today at the @ATLASexperiment exotics workshop, organised at @_nikhef by @fladias_phys. Many topics and food for thought, including work with @MCvBeekveld and @claranellist: https://t.co/mor7A0SJob
PS: you did write it down somewhere right Melissa 🧐😇

[Sascha Caron @SaschaCaron]

We estimated the speed (latency) and power consumption of running a deep anomaly detection network for LHC on GPUs, CPUs, and a novel analog neuromorphic in-memory architecture. Huge speed gain (50 nanoseconds inference) + minimal power consumption

https://t.co/cJ6nHvvHDD

[Sascha Caron @SaschaCaron]

@claranellist @_nikhef Zeer gewaardeerd + we moeten doorgaan: Let's rock the top ;)

[Sascha Caron @SaschaCaron]

My favorite podcast about physics (in German). Strongly recommended.

[Dennis Schmelzer Begabtenschmiede @Begabtenschmie1]

Die neue Episode unseres Physik-Geplänkel Podcasts ist online! Wie immer, überall wo es Podcasts gibt, oder direkt unter https://t.co/LX4jqRzxwD Viel Vergnügen! https://t.co/uBIajENWlI

[Sascha Caron @SaschaCaron]

Source: https://t.co/9RaXspGXv4

Mainly work by Zhongyi Zhang.

[Sascha Caron @SaschaCaron]

We made an addition to the unsupervised (coloured markers) @dark_machines LHC data challenge asking:

Obvious: How well are supervised methods to search for unknowns signals (in orange) ?

Weird: Can we find unknown signals by *Mixing* random Physical Theories (black M's) ?

[Sascha Caron @SaschaCaron]

Conclusion:

Best performers:
- The best "unsupervised" methods
- The "random mixture of Theories"

Bad performance:
- autoencoders and traditional methods
- supervised methods

[Sascha Caron @SaschaCaron]

We investigated the interplay of Dark Matter (DM) and Neutrinos and found some interesting (and I think new) phenomena, e.g. DM Neutrino signals can be peaks and boxes (also in combination) , see today's paper https://t.co/Vil182w4QU (work mainly by Jochem Kip & Zhongyi Zhang)

[Sascha Caron @SaschaCaron]

@bob_stienen I have a bit of data, we have two exams in Particle+Cosmos, the correlation is very high, the variance in grades between exam1 and exam2 (quite uncorrelated because done by different lecturers) is about 0.5-1.

[Sascha Caron @SaschaCaron]

My anomaly detection day ;)
- Very well organised PhyStat workshop
https://t.co/S4nwkPbaKJ
- Nature article on "How the revamped Large Hadron Collider will hunt for new physics"
https://t.co/lR82vmVs2n

[Sascha Caron @SaschaCaron]

I am looking for the first application of a real non-linear multivariate classifier in HEP (beyond Fisher etc.), I know the 1987 paper by B. Denby on NN etc., see https://t.co/iaAGO96fRR, includes also NN triggers in 1987... ). Suggestions ? Thanks.