Skip to content

Welcome!

I am an experimental particle physicist, specializing in machine learning applications and data analysis of proton-proton collision data of the Large Hadron Collider at CERN.

Currently, my research is targeted at (Run 3) data analysis with the ATLAS experiment (top quark physics (cross section, charge / energy asymmetry) and exotics (ALPs+tops with calorimeter jets)) and R&D of the High Granularity Timing Detector (HGTD) for HL-LHC / Run 4 with a special interest in database development and design / testing of Flex Tails (flexible PCBs).

I enjoy building highly performant and reliable analysis software with comprehensive documentation, as well as user interfaces that my collaborators enjoy. Furthermore, I am a strong advocate of consistency, automation and optimization.

Recent Highlights

Presentations at conferences 2026

Proceedings 2026

Design and Construction of the ATLAS High-Granularity Timing Detector

HGTD Integration Annika Stein

  • Proceedings of 34th International Workshop on Vertex Detectors (Vertex 2026), submitted to Proceedings of Science
    CDS Record: 2972629 | ATL-HGTD-PROC-2026-005
    12 Oct 2026 (submitted) | .

Abstract

The HGTD is a novel detector introduced by ATLAS to augment the new all-silicon Inner Tracker (ITk) in the pseudorapidity range from 2.4 to 4.0, adding the capability to measure charged-particle trajectories in time as well as space. Two double-sided layers of silicon sensors will provide precision timing information for charged particles with a resolution as good as 30 ps per track to help assign each particle to the correct vertex, recovering the ATLAS reconstruction performance at HL-LHC, enhancing the experiment’s pile-up rejection and providing bunch-by-bunch luminosity measurements. Readout cells have a size of 1.3mm x 1.3 mm, leading to a highly granular detector with 3.7 million channels. Low-Gain Avalanche Detectors (LGAD) technology has been chosen as it provides enough gain to reach the large signal over noise ratio needed. The requirements and overall specifications of the HGTD will be presented as well as the technical design and the project status. The R&D efforts on the detector components and construction, supported by laboratory and test beam results, will also be presented.

Data Structures and Algorithms for the Production of the ATLAS High-Granularity Timing Detector

Module Loading Annika Stein

  • Proceedings of 43rd International Conference on High Energy Physics (ICHEP 2026), submitted to Proceedings of Science
    CDS Record: 2971883 | ATL-HGTD-PROC-2026-004
    1 Oct 2026 (submitted) | .



Abstract

The High-Granularity Timing Detector is being built as part of the ATLAS Phase-II upgrade. It is based on Low-Gain-Avalanche-Detector (LGAD) sensors and provides the time of tracks in the forward region. Assembly and testing sites use the Production Database (ProdDB) to track components with properties like measurements and relations. In this presentation, advancements in the production process and how each step interacts with the ProdDB will be explained. The central component type is the module (assembled from a PCB glued on top of two hybrids). Hybrids consist of a sensor bump-bounded to a readout chip (ASIC). We model the processes of module assembly, module loading onto support units, and detector assembly with quality control and automatic reporting. The ProdDB algorithms can guide the production process by suggesting parts matching certain criteria propagated through multiple layers of parent-child relations. When the HGTD is assembled, the tools help e.g. identifying optimal flexible PCB cables (flex tails) to fit certain positions. For this purpose, the HGTD is embedded via the “slot“ table, a data structure for localization of parts and coordinate systems. An example use case is the demonstrator, where the full production and DAQ chain can be tested on a reduced scale.

Contribution: Main developer/analyzer, poster presenter.

Loss surface

  • 14.05.2026: now published in Journal of Physics: Conference Series
    IOP Publishing: DOI
Abstract

In the field of high-energy physics, deep learning algorithms continue to gain in relevance and provide performance improvements over traditional methods, for example when identifying rare signals or finding complex patterns. From an analyst’s perspective, obtaining highest possible performance is desirable, but recently, some attention has been shifted towards studying robustness of models to investigate how well these perform under slight distortions of input features. Especially for tasks that involve many (low-level) inputs, the application of deep neural networks brings new challenges. In the context of jet flavor tagging, adversarial attacks are used to probe a typical classifier‘s vulnerability and can be understood as a model for systematic uncertainties. A corresponding defense strategy, adversarial training, improves robustness, while maintaining high performance. Investigating the loss surface corresponding to the inputs and models in question reveals geometric interpretations of robustness, taking correlations into account.

Contribution: Main analyzer, developed code, interpreted the results and derived new ideas on top of that, designed the poster (see also previous one here) and then wrote the proceedings. Follow-up of this publication.

hgtd-tools

Abstract

Results from pre-production of HGTD parts and their implementation in the Production Database along with the novel user interfaces, especially in "hgtd-tools", are presented.

Contribution: Prepared the software to upload and analyse production data for module assembly, module loading and detector assembly. Calculated / optimized the flex tail categories, setup the monitoring dashboard, reporting, gitlab CI pipeline, invented the algorithms for hybrid clustering and developed the python API client including CLI and GUI.

TopCPToolkit

TopCPToolkit Logo

Abstract

TopCPToolkit is an ntuple production framework for analysing data from LHC Run 2 and Run 3. It is built around common CP and analysis algorithms in release 25 of the central athena software framework of the ATLAS Collaboration.

Contribution: Development and maintenance of software, user support, tutorials as part of top reconstruction convener role, including core team coordination.

Further Navigation / Overview

Badges / Metrics

JOSS Reviews

Last updated: October 23, 2026