{"paper":{"title":"Training speedups via batching for geometric learning: an analysis of static and dynamic algorithms","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"Changing the batching algorithm for graph neural networks can speed up training by up to 2.7 times, though the best choice depends on the data, model, batch size, hardware, and training length.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Claudia Draxl, Daniel T. Speckhard, Jonathan Godwin, Sebastian Kehl, Tim Bechtel","submitted_at":"2025-02-02T22:34:17Z","abstract_excerpt":"Graph neural networks (GNN) have shown promising results for several domains such as materials science, chemistry, and the social sciences. GNN models often contain millions of parameters, and like other neural network (NN) models, are often fed only a fraction of the graphs that make up the training dataset in batches to update model parameters. The effect of batching algorithms on training time and model performance has been thoroughly explored for NNs but not yet for GNNs. We analyze two different batching algorithms for graph-based models, namely static and dynamic batching for two dataset"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Changing the batching algorithm can provide up to a 2.7x speedup, but the fastest algorithm depends on the data, model, batch size, hardware, and number of training steps run. For a select number of combinations of batch size, dataset, and model, significant differences in model learning metrics are observed between static and dynamic batching algorithms.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The reported speedups and metric differences arise specifically from the choice of static versus dynamic batching rather than from unaccounted implementation details, hardware variability, or dataset-specific artifacts in the QM9 and AFLOW experiments.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Experiments on QM9 and AFLOW datasets show that static and dynamic batching for GNNs can yield up to 2.7x training speedups depending on data, model, batch size, hardware, and training steps, with occasional differences in learning metrics.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Changing the batching algorithm for graph neural networks can speed up training by up to 2.7 times, though the best choice depends on the data, model, batch size, hardware, and training length.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"97e6b8f7f9b53e3af4f3c296dec4ada09a5f7252e3757e4123b2ff2592aab445"},"source":{"id":"2502.00944","kind":"arxiv","version":4},"verdict":{"id":"30807c02-0f46-42a0-9669-8095c375a638","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-23T04:12:41.055343Z","strongest_claim":"Changing the batching algorithm can provide up to a 2.7x speedup, but the fastest algorithm depends on the data, model, batch size, hardware, and number of training steps run. For a select number of combinations of batch size, dataset, and model, significant differences in model learning metrics are observed between static and dynamic batching algorithms.","one_line_summary":"Experiments on QM9 and AFLOW datasets show that static and dynamic batching for GNNs can yield up to 2.7x training speedups depending on data, model, batch size, hardware, and training steps, with occasional differences in learning metrics.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The reported speedups and metric differences arise specifically from the choice of static versus dynamic batching rather than from unaccounted implementation details, hardware variability, or dataset-specific artifacts in the QM9 and AFLOW experiments.","pith_extraction_headline":"Changing the batching algorithm for graph neural networks can speed up training by up to 2.7 times, though the best choice depends on the data, model, batch size, hardware, and training length."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2502.00944/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":41,"sample":[{"doi":"","year":2022,"title":"ptgnn: A pytorch gnn library, 2022","work_id":"364fb5cd-b01f-49c7-9549-97ea40e2fca5","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"T., GODWIN, J.,ANDDRAXL, C","work_id":"7236449b-072a-4ec3-80fe-864d06a0f2c4","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2006,"title":"BISHOP, C. M.,ANDNASRABADI, N. M.Pattern recognition and machine learning, vol. 4. Springer, 2006","work_id":"4fc0852d-00ea-47d0-aa7b-3fa0339ff7ff","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2007,"title":"The tradeoffs of large scale learning.Advances in neural information processing systems 20(2007)","work_id":"beb6921b-682a-4f58-8037-dfbf0efebe52","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1991,"title":"Stochastic gradient learning in neural networks.Proceedings of Neuro- Nımes 91, 8 (1991), 12","work_id":"c663bece-ec1f-42da-b851-2268c27f6b7f","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":41,"snapshot_sha256":"a6aa3247c841375255fb68bc7aa9064a633482e0c86622ba7f34104a255efaa5","internal_anchors":7},"formal_canon":{"evidence_count":2,"snapshot_sha256":"f2a42a0b2a3bbdaae153c576f41b140f1c55c4a2454094b0d1ae804650908873"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}