{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:635KPATCUVX35O5JAWU6IORPW7","short_pith_number":"pith:635KPATC","schema_version":"1.0","canonical_sha256":"f6faa78262a56fbebba905a9e43a2fb7e7f19c6ecdf887ecfa3ed51bc1913f6c","source":{"kind":"arxiv","id":"2409.03463","version":3},"attestation_state":"computed","paper":{"title":"Massive Activations in Graph Neural Networks: Decoding Attention for Domain-Dependent Interpretability","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Lorenzo Bini, Marco Sorbi, Stephane Marchand-Maillet","submitted_at":"2024-09-05T12:19:07Z","abstract_excerpt":"Graph Neural Networks (GNNs) have become increasingly popular for effectively modeling graph-structured data, and attention mechanisms have been pivotal in enabling these models to capture complex patterns. In our study, we reveal a critical yet underexplored consequence of integrating attention into edge-featured GNNs: the emergence of Massive Activations (MAs) within attention layers. By developing a novel method for detecting MAs on edge features, we show that these extreme activations are not only activation anomalies but encode domain-relevant signals. Our post-hoc interpretability analys"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2409.03463","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-09-05T12:19:07Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"c84ccbfa40a4fd84f5c58b1ba419447592daf48a1b0d245e80f536e3603cd53c","abstract_canon_sha256":"ebb2d43aef22a8140f3921227c432b9f6cd9c07162f35b7bf3f27a954a338223"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:26:07.043008Z","signature_b64":"MR8/FoAVutQf30HohVhQdM0NlXfJ57HFmQ2cXTDj0cC/z/8bGDMj2OWZPriu8J+p14cMZyAMY9TU8p03bWnqBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f6faa78262a56fbebba905a9e43a2fb7e7f19c6ecdf887ecfa3ed51bc1913f6c","last_reissued_at":"2026-07-05T10:26:07.041959Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:26:07.041959Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Massive Activations in Graph Neural Networks: Decoding Attention for Domain-Dependent Interpretability","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Lorenzo Bini, Marco Sorbi, Stephane Marchand-Maillet","submitted_at":"2024-09-05T12:19:07Z","abstract_excerpt":"Graph Neural Networks (GNNs) have become increasingly popular for effectively modeling graph-structured data, and attention mechanisms have been pivotal in enabling these models to capture complex patterns. In our study, we reveal a critical yet underexplored consequence of integrating attention into edge-featured GNNs: the emergence of Massive Activations (MAs) within attention layers. By developing a novel method for detecting MAs on edge features, we show that these extreme activations are not only activation anomalies but encode domain-relevant signals. Our post-hoc interpretability analys"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2409.03463","kind":"arxiv","version":3},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2409.03463/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2409.03463","created_at":"2026-07-05T10:26:07.042100+00:00"},{"alias_kind":"arxiv_version","alias_value":"2409.03463v3","created_at":"2026-07-05T10:26:07.042100+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2409.03463","created_at":"2026-07-05T10:26:07.042100+00:00"},{"alias_kind":"pith_short_12","alias_value":"635KPATCUVX3","created_at":"2026-07-05T10:26:07.042100+00:00"},{"alias_kind":"pith_short_16","alias_value":"635KPATCUVX35O5J","created_at":"2026-07-05T10:26:07.042100+00:00"},{"alias_kind":"pith_short_8","alias_value":"635KPATC","created_at":"2026-07-05T10:26:07.042100+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2505.21670","citing_title":"Rethinking the Outlier Distribution in Large Language Models: An In-depth Study","ref_index":4,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/635KPATCUVX35O5JAWU6IORPW7","json":"https://pith.science/pith/635KPATCUVX35O5JAWU6IORPW7.json","graph_json":"https://pith.science/api/pith-number/635KPATCUVX35O5JAWU6IORPW7/graph.json","events_json":"https://pith.science/api/pith-number/635KPATCUVX35O5JAWU6IORPW7/events.json","paper":"https://pith.science/paper/635KPATC"},"agent_actions":{"view_html":"https://pith.science/pith/635KPATCUVX35O5JAWU6IORPW7","download_json":"https://pith.science/pith/635KPATCUVX35O5JAWU6IORPW7.json","view_paper":"https://pith.science/paper/635KPATC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2409.03463&json=true","fetch_graph":"https://pith.science/api/pith-number/635KPATCUVX35O5JAWU6IORPW7/graph.json","fetch_events":"https://pith.science/api/pith-number/635KPATCUVX35O5JAWU6IORPW7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/635KPATCUVX35O5JAWU6IORPW7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/635KPATCUVX35O5JAWU6IORPW7/action/storage_attestation","attest_author":"https://pith.science/pith/635KPATCUVX35O5JAWU6IORPW7/action/author_attestation","sign_citation":"https://pith.science/pith/635KPATCUVX35O5JAWU6IORPW7/action/citation_signature","submit_replication":"https://pith.science/pith/635KPATCUVX35O5JAWU6IORPW7/action/replication_record"}},"created_at":"2026-07-05T10:26:07.042100+00:00","updated_at":"2026-07-05T10:26:07.042100+00:00"}