{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2021:YPL7GVZFXPHQW4NBPG5OEQ2K4W","short_pith_number":"pith:YPL7GVZF","canonical_record":{"source":{"id":"2105.14491","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2021-05-30T10:17:58Z","cross_cats_sorted":[],"title_canon_sha256":"215510f58299ba0a78cba7aa74025151976b6ef2c24a42b5114174673e589208","abstract_canon_sha256":"044bcd14d1a2616432d49ef12852583c4df320ff08cd76da89b98519a36718e6"},"schema_version":"1.0"},"canonical_sha256":"c3d7f35725bbcf0b71a179bae2434ae587aff5aaf0474de3a8ef794bf0c15a25","source":{"kind":"arxiv","id":"2105.14491","version":3},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2105.14491","created_at":"2026-05-17T23:38:15Z"},{"alias_kind":"arxiv_version","alias_value":"2105.14491v3","created_at":"2026-05-17T23:38:15Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2105.14491","created_at":"2026-05-17T23:38:15Z"},{"alias_kind":"pith_short_12","alias_value":"YPL7GVZFXPHQ","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_16","alias_value":"YPL7GVZFXPHQW4NB","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_8","alias_value":"YPL7GVZF","created_at":"2026-05-18T12:33:33Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2021:YPL7GVZFXPHQW4NBPG5OEQ2K4W","target":"record","payload":{"canonical_record":{"source":{"id":"2105.14491","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2021-05-30T10:17:58Z","cross_cats_sorted":[],"title_canon_sha256":"215510f58299ba0a78cba7aa74025151976b6ef2c24a42b5114174673e589208","abstract_canon_sha256":"044bcd14d1a2616432d49ef12852583c4df320ff08cd76da89b98519a36718e6"},"schema_version":"1.0"},"canonical_sha256":"c3d7f35725bbcf0b71a179bae2434ae587aff5aaf0474de3a8ef794bf0c15a25","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:38:15.390041Z","signature_b64":"GZu4q0LMmgBpV3d8lIW2Rk0t3RgH8T4DiHbrFEf0pNtWbcOzPTvCuynON7Sm6wV5H675A43VKQhjTuf+AEJlDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c3d7f35725bbcf0b71a179bae2434ae587aff5aaf0474de3a8ef794bf0c15a25","last_reissued_at":"2026-05-17T23:38:15.389458Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:38:15.389458Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2105.14491","source_version":3,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:38:15Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"KaJGSS3x7mPHfOLrHHMqf0BUUhVG3Vq9RsvgP8TSKrZNescQczy+TT27sYVSXGi5E0EcYj76sm31NiiXZq/NBA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-24T04:37:00.191975Z"},"content_sha256":"b594ca98282d9160a418f9e893f5086093e261bc9128851b92ecf08e3a8c6f2c","schema_version":"1.0","event_id":"sha256:b594ca98282d9160a418f9e893f5086093e261bc9128851b92ecf08e3a8c6f2c"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2021:YPL7GVZFXPHQW4NBPG5OEQ2K4W","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"How Attentive are Graph Attention Networks?","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Graph Attention Networks use static attention that cannot express simple graph problems, fixed by reordering to create dynamic GATv2.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Eran Yahav, Shaked Brody, Uri Alon","submitted_at":"2021-05-30T10:17:58Z","abstract_excerpt":"Graph Attention Networks (GATs) are one of the most popular GNN architectures and are considered as the state-of-the-art architecture for representation learning with graphs. In GAT, every node attends to its neighbors given its own representation as the query. However, in this paper we show that GAT computes a very limited kind of attention: the ranking of the attention scores is unconditioned on the query node. We formally define this restricted kind of attention as static attention and distinguish it from a strictly more expressive dynamic attention. Because GATs use a static attention mech"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Because GATs use a static attention mechanism, there are simple graph problems that GAT cannot express: in a controlled problem, we show that static attention hinders GAT from even fitting the training data. ... GATv2: a dynamic graph attention variant that is strictly more expressive than GAT.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the controlled synthetic problem is representative of the limitations that matter in real benchmarks, and that reordering the attention operations fully converts static attention into dynamic attention without side effects on optimization or generalization.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"GAT uses static attention where neighbor rankings ignore the query node and thus cannot express some graph problems; GATv2 enables dynamic attention and outperforms GAT on 11 OGB and other benchmarks with equal parameters.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Graph Attention Networks use static attention that cannot express simple graph problems, fixed by reordering to create dynamic GATv2.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"14ea25d6a84064fa403263761241d04840d79cec79fe6fe091be2614de82e8bf"},"source":{"id":"2105.14491","kind":"arxiv","version":3},"verdict":{"id":"6b921b2b-a1e4-47a4-a3d8-484679498b1f","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-17T02:26:02.690351Z","strongest_claim":"Because GATs use a static attention mechanism, there are simple graph problems that GAT cannot express: in a controlled problem, we show that static attention hinders GAT from even fitting the training data. ... GATv2: a dynamic graph attention variant that is strictly more expressive than GAT.","one_line_summary":"GAT uses static attention where neighbor rankings ignore the query node and thus cannot express some graph problems; GATv2 enables dynamic attention and outperforms GAT on 11 OGB and other benchmarks with equal parameters.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the controlled synthetic problem is representative of the limitations that matter in real benchmarks, and that reordering the attention operations fully converts static attention into dynamic attention without side effects on optimization or generalization.","pith_extraction_headline":"Graph Attention Networks use static attention that cannot express simple graph problems, fixed by reordering to create dynamic GATv2."},"references":{"count":70,"sample":[{"doi":"","year":2018,"title":"Learning to represent programs with graphs","work_id":"7a597620-0dd2-4229-9c8e-eb2089541ab7","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"On the bottleneck of graph neural networks and its practical implications","work_id":"17b47a81-dc95-47f1-84af-50161b9c0a78","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1993,"title":"Diffusion-convolutional neural networks","work_id":"103ff6d2-99ee-456c-9336-2bc6c8dbb6aa","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2014,"title":"Neural Machine Translation by Jointly Learning to Align and Translate","work_id":"d831e763-d530-4029-a65c-ac595d82cb2a","ref_index":4,"cited_arxiv_id":"1409.0473","is_internal_anchor":true},{"doi":"","year":2016,"title":"Interaction networks for learning about objects, relations and physics","work_id":"22f5e757-33a3-4521-b989-944e606e92da","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":70,"snapshot_sha256":"c58a8ffe2b6c93cf7db396f01222836bfdf481e12b17b42f93957fb5baf4c820","internal_anchors":5},"formal_canon":{"evidence_count":2,"snapshot_sha256":"7cc9638c3e4f71f1f8b9702bf538ba143ce315f8a77b572ee72032d5a1ba8971"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"6b921b2b-a1e4-47a4-a3d8-484679498b1f"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:38:15Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"U/L5zO/TEWVXCCxWnXAjUwaOlUxGgp2MrZmhlO3K7QWNz6CZbXmGZAl9mN1u4bnZGKeuYjCdMi1uMiHFsBbCBw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-24T04:37:00.192616Z"},"content_sha256":"2bd048d85644e01393ede7aa8439bcd31ad0377d012ff586e9d387546d96f2df","schema_version":"1.0","event_id":"sha256:2bd048d85644e01393ede7aa8439bcd31ad0377d012ff586e9d387546d96f2df"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/YPL7GVZFXPHQW4NBPG5OEQ2K4W/bundle.json","state_url":"https://pith.science/pith/YPL7GVZFXPHQW4NBPG5OEQ2K4W/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/YPL7GVZFXPHQW4NBPG5OEQ2K4W/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-05-24T04:37:00Z","links":{"resolver":"https://pith.science/pith/YPL7GVZFXPHQW4NBPG5OEQ2K4W","bundle":"https://pith.science/pith/YPL7GVZFXPHQW4NBPG5OEQ2K4W/bundle.json","state":"https://pith.science/pith/YPL7GVZFXPHQW4NBPG5OEQ2K4W/state.json","well_known_bundle":"https://pith.science/.well-known/pith/YPL7GVZFXPHQW4NBPG5OEQ2K4W/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2021:YPL7GVZFXPHQW4NBPG5OEQ2K4W","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"044bcd14d1a2616432d49ef12852583c4df320ff08cd76da89b98519a36718e6","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2021-05-30T10:17:58Z","title_canon_sha256":"215510f58299ba0a78cba7aa74025151976b6ef2c24a42b5114174673e589208"},"schema_version":"1.0","source":{"id":"2105.14491","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2105.14491","created_at":"2026-05-17T23:38:15Z"},{"alias_kind":"arxiv_version","alias_value":"2105.14491v3","created_at":"2026-05-17T23:38:15Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2105.14491","created_at":"2026-05-17T23:38:15Z"},{"alias_kind":"pith_short_12","alias_value":"YPL7GVZFXPHQ","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_16","alias_value":"YPL7GVZFXPHQW4NB","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_8","alias_value":"YPL7GVZF","created_at":"2026-05-18T12:33:33Z"}],"graph_snapshots":[{"event_id":"sha256:2bd048d85644e01393ede7aa8439bcd31ad0377d012ff586e9d387546d96f2df","target":"graph","created_at":"2026-05-17T23:38:15Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Because GATs use a static attention mechanism, there are simple graph problems that GAT cannot express: in a controlled problem, we show that static attention hinders GAT from even fitting the training data. ... GATv2: a dynamic graph attention variant that is strictly more expressive than GAT."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the controlled synthetic problem is representative of the limitations that matter in real benchmarks, and that reordering the attention operations fully converts static attention into dynamic attention without side effects on optimization or generalization."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"GAT uses static attention where neighbor rankings ignore the query node and thus cannot express some graph problems; GATv2 enables dynamic attention and outperforms GAT on 11 OGB and other benchmarks with equal parameters."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Graph Attention Networks use static attention that cannot express simple graph problems, fixed by reordering to create dynamic GATv2."}],"snapshot_sha256":"14ea25d6a84064fa403263761241d04840d79cec79fe6fe091be2614de82e8bf"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"7cc9638c3e4f71f1f8b9702bf538ba143ce315f8a77b572ee72032d5a1ba8971"},"paper":{"abstract_excerpt":"Graph Attention Networks (GATs) are one of the most popular GNN architectures and are considered as the state-of-the-art architecture for representation learning with graphs. In GAT, every node attends to its neighbors given its own representation as the query. However, in this paper we show that GAT computes a very limited kind of attention: the ranking of the attention scores is unconditioned on the query node. We formally define this restricted kind of attention as static attention and distinguish it from a strictly more expressive dynamic attention. Because GATs use a static attention mech","authors_text":"Eran Yahav, Shaked Brody, Uri Alon","cross_cats":[],"headline":"Graph Attention Networks use static attention that cannot express simple graph problems, fixed by reordering to create dynamic GATv2.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2021-05-30T10:17:58Z","title":"How Attentive are Graph Attention Networks?"},"references":{"count":70,"internal_anchors":5,"resolved_work":70,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Learning to represent programs with graphs","work_id":"7a597620-0dd2-4229-9c8e-eb2089541ab7","year":2018},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"On the bottleneck of graph neural networks and its practical implications","work_id":"17b47a81-dc95-47f1-84af-50161b9c0a78","year":2021},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Diffusion-convolutional neural networks","work_id":"103ff6d2-99ee-456c-9336-2bc6c8dbb6aa","year":1993},{"cited_arxiv_id":"1409.0473","doi":"","is_internal_anchor":true,"ref_index":4,"title":"Neural Machine Translation by Jointly Learning to Align and Translate","work_id":"d831e763-d530-4029-a65c-ac595d82cb2a","year":2014},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Interaction networks for learning about objects, relations and physics","work_id":"22f5e757-33a3-4521-b989-944e606e92da","year":2016}],"snapshot_sha256":"c58a8ffe2b6c93cf7db396f01222836bfdf481e12b17b42f93957fb5baf4c820"},"source":{"id":"2105.14491","kind":"arxiv","version":3},"verdict":{"created_at":"2026-05-17T02:26:02.690351Z","id":"6b921b2b-a1e4-47a4-a3d8-484679498b1f","model_set":{"reader":"grok-4.3"},"one_line_summary":"GAT uses static attention where neighbor rankings ignore the query node and thus cannot express some graph problems; GATv2 enables dynamic attention and outperforms GAT on 11 OGB and other benchmarks with equal parameters.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Graph Attention Networks use static attention that cannot express simple graph problems, fixed by reordering to create dynamic GATv2.","strongest_claim":"Because GATs use a static attention mechanism, there are simple graph problems that GAT cannot express: in a controlled problem, we show that static attention hinders GAT from even fitting the training data. ... GATv2: a dynamic graph attention variant that is strictly more expressive than GAT.","weakest_assumption":"That the controlled synthetic problem is representative of the limitations that matter in real benchmarks, and that reordering the attention operations fully converts static attention into dynamic attention without side effects on optimization or generalization."}},"verdict_id":"6b921b2b-a1e4-47a4-a3d8-484679498b1f"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:b594ca98282d9160a418f9e893f5086093e261bc9128851b92ecf08e3a8c6f2c","target":"record","created_at":"2026-05-17T23:38:15Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"044bcd14d1a2616432d49ef12852583c4df320ff08cd76da89b98519a36718e6","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2021-05-30T10:17:58Z","title_canon_sha256":"215510f58299ba0a78cba7aa74025151976b6ef2c24a42b5114174673e589208"},"schema_version":"1.0","source":{"id":"2105.14491","kind":"arxiv","version":3}},"canonical_sha256":"c3d7f35725bbcf0b71a179bae2434ae587aff5aaf0474de3a8ef794bf0c15a25","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"c3d7f35725bbcf0b71a179bae2434ae587aff5aaf0474de3a8ef794bf0c15a25","first_computed_at":"2026-05-17T23:38:15.389458Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:38:15.389458Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"GZu4q0LMmgBpV3d8lIW2Rk0t3RgH8T4DiHbrFEf0pNtWbcOzPTvCuynON7Sm6wV5H675A43VKQhjTuf+AEJlDA==","signature_status":"signed_v1","signed_at":"2026-05-17T23:38:15.390041Z","signed_message":"canonical_sha256_bytes"},"source_id":"2105.14491","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:b594ca98282d9160a418f9e893f5086093e261bc9128851b92ecf08e3a8c6f2c","sha256:2bd048d85644e01393ede7aa8439bcd31ad0377d012ff586e9d387546d96f2df"],"state_sha256":"b401703c2255776e3aaa0702a44c85353c8e3822649fff7ff04d1e7c1a9f1fcd"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"dq7brbEw0RV399ntooL25wd/rkF6cezrEMztxXGNagkFEnIb6vVAo5lsoEFDKwZHPeSZ8IyGL4SGIB0W83lXAw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-24T04:37:00.197023Z","bundle_sha256":"e27b35ad070266221cda7be96e9185fd1feda2ff753de210f103c4e7f7eac92d"}}