Examples

This section provides worked examples demonstrating the library's capabilities in real-world deployments — conference talks, tutorials, and teaching.

For an exhaustive list of publications, see Publications. The examples below are the ones with associated talks, slides, or external materials.

Featured Examples

Example 1: Influencer Loss — End-to-end Geometric Representation Learning for Track Reconstruction

const talk = await daniel.present({
  title: 'Influencer Loss: End-to-end Geometric Representation Learning for Track Reconstruction',
  venue: 'CHEP 2024 — Computing in High Energy Physics',
  date: '2024',
  format: 'contributed',
  proceedings: 'EPJ Web of Conferences 295, 09016',
  doi: 'https://doi.org/10.1051/epjconf/202429509016'
});

Abstract: A novel loss function for end-to-end geometric representation learning, designed to produce embeddings that respect the topology of charged particle trajectories. Tested on the ATLAS ITk reconstruction problem.

---

Example 2: Physics Performance of the ATLAS GNN4ITk Track Reconstruction Chain

const talk = await daniel.present({
  title: 'Physics Performance of the ATLAS GNN4ITk Track Reconstruction Chain',
  venue: 'CHEP 2024',
  date: '2024',
  format: 'contributed',
  proceedings: 'EPJ Web of Conferences 295, 03030',
  doi: 'https://doi.org/10.1051/epjconf/202429503030',
  collaborators: ['S. Caillou', 'P. Calafiura', 'X. Ju', 'T.Q. Pham', 'C. Rougier', 'J. Stark', 'A. Vallier']
});

Abstract: Production-grade physics performance of the GNN-based tracking pipeline being prepared for the ATLAS Inner Tracker upgrade. End-to-end results on simulated HL-LHC events, compared against the classical CKF baseline.

---

Example 3: A Language Model for Particle Tracking

const talk = await daniel.present({
  title: 'A Language Model for Particle Tracking',
  venue: 'ML4PS / arXiv:2402.10239',
  date: '2024-02-14',
  format: 'preprint + talk',
  link: 'https://arxiv.org/abs/2402.10239'
});

Abstract: Reframing particle tracking as a sequence modeling problem. Tokenizing detector hits and training a transformer to produce track candidates — an early experiment in physics language models.

---

Tutorials & Workshops

Graph Neural Networks for Charged Particle Tracking

const tutorial = await daniel.teach({
  title: 'Graph Neural Networks for Charged Particle Tracking',
  event: 'CERN summer student lectures and various ML4HEP workshops',
  materials: 'https://github.com/murnanedaniel',
  level: 'intermediate'
});

Description: Hands-on tutorial covering the full GNN tracking pipeline: graph construction from detector hits, edge classification, segment connection, and physics-performance evaluation. Includes runnable notebooks built on the Exa.TrkX framework.

---

Conference Proceedings

A selection of the most relevant proceedings with associated talks. The full list lives on the Publications page.

Year	Title	Venue
2024	Influencer Loss: End-to-end Geometric Representation Learning for Track Reconstruction	CHEP 2024 (EPJ Web 295, 09016)
2024	Physics Performance of the ATLAS GNN4ITk Track Reconstruction Chain	CHEP 2024 (EPJ Web 295, 03030)
2024	A Language Model for Particle Tracking	arXiv:2402.10239
2024	Learning to Reconstruct Quirky Tracks	arXiv:2410.00269

How to Request a Demo

Interested in a live demonstration, invited talk, or tutorial? See the Contributing Guide for booking protocols, or call daniel.schedule() directly.