{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:YVX5FQR7QSYFKGMS4VXDTFJQ2O","short_pith_number":"pith:YVX5FQR7","schema_version":"1.0","canonical_sha256":"c56fd2c23f84b0551992e56e399530d3b408cfd5822bdee1ea534f15b18616f2","source":{"kind":"arxiv","id":"2205.12411","version":5},"attestation_state":"computed","paper":{"title":"Linear Connectivity Reveals Generalization Strategies","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Jeevesh Juneja, Jo\\~ao Sedoc, Kyunghyun Cho, Naomi Saphra, Rachit Bansal","submitted_at":"2022-05-24T23:43:02Z","abstract_excerpt":"It is widely accepted in the mode connectivity literature that when two neural networks are trained similarly on the same data, they are connected by a path through parameter space over which test set accuracy is maintained. Under some circumstances, including transfer learning from pretrained models, these paths are presumed to be linear. In contrast to existing results, we find that among text classifiers (trained on MNLI, QQP, and CoLA), some pairs of finetuned models have large barriers of increasing loss on the linear paths between them. On each task, we find distinct clusters of models w"},"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":"2205.12411","kind":"arxiv","version":5},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2022-05-24T23:43:02Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"68fb8d9f4a449c4b5d35ac4daf17edd88383a77183b4208fb9d100737dd71b0f","abstract_canon_sha256":"ff934f7524050fde2dd5eb4dab5725729929f219f9741b0f1daf3b9c116e95c4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:34:46.360975Z","signature_b64":"2/y85ofVuZDyve1gmi9u0X4FtwGT+2J7r+LzJKWFVTWaoVzzcDhqg1rFK1nIYxMiFZPBSABrHcQ0rdOADeNoDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c56fd2c23f84b0551992e56e399530d3b408cfd5822bdee1ea534f15b18616f2","last_reissued_at":"2026-07-05T05:34:46.360541Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:34:46.360541Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Linear Connectivity Reveals Generalization Strategies","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Jeevesh Juneja, Jo\\~ao Sedoc, Kyunghyun Cho, Naomi Saphra, Rachit Bansal","submitted_at":"2022-05-24T23:43:02Z","abstract_excerpt":"It is widely accepted in the mode connectivity literature that when two neural networks are trained similarly on the same data, they are connected by a path through parameter space over which test set accuracy is maintained. Under some circumstances, including transfer learning from pretrained models, these paths are presumed to be linear. In contrast to existing results, we find that among text classifiers (trained on MNLI, QQP, and CoLA), some pairs of finetuned models have large barriers of increasing loss on the linear paths between them. On each task, we find distinct clusters of models w"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2205.12411","kind":"arxiv","version":5},"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/2205.12411/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":"2205.12411","created_at":"2026-07-05T05:34:46.360603+00:00"},{"alias_kind":"arxiv_version","alias_value":"2205.12411v5","created_at":"2026-07-05T05:34:46.360603+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2205.12411","created_at":"2026-07-05T05:34:46.360603+00:00"},{"alias_kind":"pith_short_12","alias_value":"YVX5FQR7QSYF","created_at":"2026-07-05T05:34:46.360603+00:00"},{"alias_kind":"pith_short_16","alias_value":"YVX5FQR7QSYFKGMS","created_at":"2026-07-05T05:34:46.360603+00:00"},{"alias_kind":"pith_short_8","alias_value":"YVX5FQR7","created_at":"2026-07-05T05:34:46.360603+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.04754","citing_title":"Beyond Structural Symmetries: Linear Mode Connectivity via Neuron Identifiability","ref_index":86,"is_internal_anchor":false},{"citing_arxiv_id":"2405.07987","citing_title":"The Platonic Representation Hypothesis","ref_index":94,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YVX5FQR7QSYFKGMS4VXDTFJQ2O","json":"https://pith.science/pith/YVX5FQR7QSYFKGMS4VXDTFJQ2O.json","graph_json":"https://pith.science/api/pith-number/YVX5FQR7QSYFKGMS4VXDTFJQ2O/graph.json","events_json":"https://pith.science/api/pith-number/YVX5FQR7QSYFKGMS4VXDTFJQ2O/events.json","paper":"https://pith.science/paper/YVX5FQR7"},"agent_actions":{"view_html":"https://pith.science/pith/YVX5FQR7QSYFKGMS4VXDTFJQ2O","download_json":"https://pith.science/pith/YVX5FQR7QSYFKGMS4VXDTFJQ2O.json","view_paper":"https://pith.science/paper/YVX5FQR7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2205.12411&json=true","fetch_graph":"https://pith.science/api/pith-number/YVX5FQR7QSYFKGMS4VXDTFJQ2O/graph.json","fetch_events":"https://pith.science/api/pith-number/YVX5FQR7QSYFKGMS4VXDTFJQ2O/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YVX5FQR7QSYFKGMS4VXDTFJQ2O/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YVX5FQR7QSYFKGMS4VXDTFJQ2O/action/storage_attestation","attest_author":"https://pith.science/pith/YVX5FQR7QSYFKGMS4VXDTFJQ2O/action/author_attestation","sign_citation":"https://pith.science/pith/YVX5FQR7QSYFKGMS4VXDTFJQ2O/action/citation_signature","submit_replication":"https://pith.science/pith/YVX5FQR7QSYFKGMS4VXDTFJQ2O/action/replication_record"}},"created_at":"2026-07-05T05:34:46.360603+00:00","updated_at":"2026-07-05T05:34:46.360603+00:00"}