{"paper":{"title":"Double Metric Learning for Building Directed Graphs with Chain Connections for the ATLAS ITk Detector","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Double Metric Learning resolves contrastive loss conflicts in chain connections by learning two node representations for directed graph construction.","cross_cats":["hep-ex","hep-ph"],"primary_cat":"physics.data-an","authors_text":"Jay Chan","submitted_at":"2026-05-13T21:31:28Z","abstract_excerpt":"Graph construction is an essential step in the Graph Neural Network (GNN) based tracking pipelines. The goal of the graph construction is to construct a graph that contains only the defined true edge connections between nodes (detector hits). A promising approach for the graph construction is through the Metric Learning approach, where a node representation in an embedding space is learned, and nodes are connected according to their distance in the embedding space. The loss function for the metric learning in this case is a contrastive loss encouraging the true pairs of nodes to be close to ea"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We test this idea with the ATLAS ITk detector at the HL-LHC using the ATLAS ITk simulation and show better graph construction performance particularly for particles with high transverse momentum compared to the Simple Metric Learning approach. We also show that Double Metric Learning is able to accurately predict edge direction.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That learning two independent node representations resolves the contrastive loss conflict for chain connections without introducing new overfitting or bias that would degrade overall tracking performance on real data.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Double metric learning learns two embeddings per node to build directed graphs with chain connections, yielding better performance than single metric learning for high-pT particles and accurate edge direction prediction in ATLAS ITk simulations.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Double Metric Learning resolves contrastive loss conflicts in chain connections by learning two node representations for directed graph construction.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"8f0c97df64f947435e32ce5fe7814a94ddb284afae589b304a9999e53dd01699"},"source":{"id":"2605.14131","kind":"arxiv","version":1},"verdict":{"id":"1ac5b16f-c90b-4276-8ef9-cac14ef4f128","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T02:14:42.641074Z","strongest_claim":"We test this idea with the ATLAS ITk detector at the HL-LHC using the ATLAS ITk simulation and show better graph construction performance particularly for particles with high transverse momentum compared to the Simple Metric Learning approach. We also show that Double Metric Learning is able to accurately predict edge direction.","one_line_summary":"Double metric learning learns two embeddings per node to build directed graphs with chain connections, yielding better performance than single metric learning for high-pT particles and accurate edge direction prediction in ATLAS ITk simulations.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That learning two independent node representations resolves the contrastive loss conflict for chain connections without introducing new overfitting or bias that would degrade overall tracking performance on real data.","pith_extraction_headline":"Double Metric Learning resolves contrastive loss conflicts in chain connections by learning two node representations for directed graph construction."},"references":{"count":21,"sample":[{"doi":"10.1103/revmodphys.82.1419","year":2010,"title":"Track and vertex reconstruction: From classical to adaptive methods , author =. Rev. Mod. Phys. , volume =. 2010 , month =. doi:10.1103/RevModPhys.82.1419 , url =","work_id":"3565acb0-c762-4d05-8552-368764eeee2f","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1140/epjc/s10052-017-5225-7","year":2017,"title":"Performance of the ATLAS Track Reconstruction Algorithms in Dense Environments in LHC Run 2","work_id":"3375d343-abe6-449c-8e66-51bd4bde56d8","ref_index":2,"cited_arxiv_id":"1704.07983","is_internal_anchor":true},{"doi":"10.1088/1748-0221/9/10/p10009","year":2014,"title":"Description and performance of track and primary-vertex reconstruction with the CMS tracker","work_id":"2d66f302-5f19-4521-a306-f9e7eed82afc","ref_index":3,"cited_arxiv_id":"1405.6569","is_internal_anchor":true},{"doi":"","year":2025,"title":"Optimizations of the ATLAS ITk GNN reconstruction pipeline. 2025","work_id":"af0163a6-4524-4074-aa03-66100e3ad4aa","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1140/epjc/s10052-021-09675-8","year":2021,"title":"Performance of a geometric deep learning pipeline for HL-LHC particle tracking","work_id":"3631dcd6-38d7-4a77-92de-b42a1e4d8275","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":21,"snapshot_sha256":"9684e27f9a98db0f845d28f6533565f8b4a57ca992fb8596cd53704d90d18cba","internal_anchors":3},"formal_canon":{"evidence_count":2,"snapshot_sha256":"7cf9252485fe19fe2c0b374e11180d75328e2b8a26e2b5cfe8ad02d1c1d16c02"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}