{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2020:FPOP7OXYMMLXKGCYR2G62ROY4T","short_pith_number":"pith:FPOP7OXY","schema_version":"1.0","canonical_sha256":"2bdcffbaf863177518588e8ded45d8e4da3ab1d009254b6b0f98ff537ae930da","source":{"kind":"arxiv","id":"2011.12428","version":2},"attestation_state":"computed","paper":{"title":"Align, then memorise: the dynamics of learning with feedback alignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cond-mat.dis-nn","cs.LG","cs.NE"],"primary_cat":"stat.ML","authors_text":"Maria Refinetti, Ruben Ohana, Sebastian Goldt, St\\'ephane d'Ascoli","submitted_at":"2020-11-24T22:21:27Z","abstract_excerpt":"Direct Feedback Alignment (DFA) is emerging as an efficient and biologically plausible alternative to the ubiquitous backpropagation algorithm for training deep neural networks. Despite relying on random feedback weights for the backward pass, DFA successfully trains state-of-the-art models such as Transformers. On the other hand, it notoriously fails to train convolutional networks. An understanding of the inner workings of DFA to explain these diverging results remains elusive. Here, we propose a theory for the success of DFA. We first show that learning in shallow networks proceeds in two s"},"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":"2011.12428","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"stat.ML","submitted_at":"2020-11-24T22:21:27Z","cross_cats_sorted":["cond-mat.dis-nn","cs.LG","cs.NE"],"title_canon_sha256":"a5ef9cdf7746be6b17f7e632a7c48331801cd70d79f0f7475b950eec54d9105e","abstract_canon_sha256":"51bf2c7e2f374bc0b6f02c75435205bdd4fb3e037987450faa383d5aa84eff2f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:48:09.845251Z","signature_b64":"2XK/2zijsqmMDiaMklB0s4rp3pWmL5CorUkCMaqeQMonIshp0J/YeYt9dgSIeZO1qiXFe0QG6yQJMVj4n3hADQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2bdcffbaf863177518588e8ded45d8e4da3ab1d009254b6b0f98ff537ae930da","last_reissued_at":"2026-07-05T02:48:09.844853Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:48:09.844853Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Align, then memorise: the dynamics of learning with feedback alignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cond-mat.dis-nn","cs.LG","cs.NE"],"primary_cat":"stat.ML","authors_text":"Maria Refinetti, Ruben Ohana, Sebastian Goldt, St\\'ephane d'Ascoli","submitted_at":"2020-11-24T22:21:27Z","abstract_excerpt":"Direct Feedback Alignment (DFA) is emerging as an efficient and biologically plausible alternative to the ubiquitous backpropagation algorithm for training deep neural networks. Despite relying on random feedback weights for the backward pass, DFA successfully trains state-of-the-art models such as Transformers. On the other hand, it notoriously fails to train convolutional networks. An understanding of the inner workings of DFA to explain these diverging results remains elusive. Here, we propose a theory for the success of DFA. We first show that learning in shallow networks proceeds in two s"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2011.12428","kind":"arxiv","version":2},"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/2011.12428/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":"2011.12428","created_at":"2026-07-05T02:48:09.844905+00:00"},{"alias_kind":"arxiv_version","alias_value":"2011.12428v2","created_at":"2026-07-05T02:48:09.844905+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2011.12428","created_at":"2026-07-05T02:48:09.844905+00:00"},{"alias_kind":"pith_short_12","alias_value":"FPOP7OXYMMLX","created_at":"2026-07-05T02:48:09.844905+00:00"},{"alias_kind":"pith_short_16","alias_value":"FPOP7OXYMMLXKGCY","created_at":"2026-07-05T02:48:09.844905+00:00"},{"alias_kind":"pith_short_8","alias_value":"FPOP7OXY","created_at":"2026-07-05T02:48:09.844905+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.15622","citing_title":"Position: Zeroth-Order Optimization in Deep Learning Is Underexplored, Not Underpowered","ref_index":67,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/FPOP7OXYMMLXKGCYR2G62ROY4T","json":"https://pith.science/pith/FPOP7OXYMMLXKGCYR2G62ROY4T.json","graph_json":"https://pith.science/api/pith-number/FPOP7OXYMMLXKGCYR2G62ROY4T/graph.json","events_json":"https://pith.science/api/pith-number/FPOP7OXYMMLXKGCYR2G62ROY4T/events.json","paper":"https://pith.science/paper/FPOP7OXY"},"agent_actions":{"view_html":"https://pith.science/pith/FPOP7OXYMMLXKGCYR2G62ROY4T","download_json":"https://pith.science/pith/FPOP7OXYMMLXKGCYR2G62ROY4T.json","view_paper":"https://pith.science/paper/FPOP7OXY","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2011.12428&json=true","fetch_graph":"https://pith.science/api/pith-number/FPOP7OXYMMLXKGCYR2G62ROY4T/graph.json","fetch_events":"https://pith.science/api/pith-number/FPOP7OXYMMLXKGCYR2G62ROY4T/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/FPOP7OXYMMLXKGCYR2G62ROY4T/action/timestamp_anchor","attest_storage":"https://pith.science/pith/FPOP7OXYMMLXKGCYR2G62ROY4T/action/storage_attestation","attest_author":"https://pith.science/pith/FPOP7OXYMMLXKGCYR2G62ROY4T/action/author_attestation","sign_citation":"https://pith.science/pith/FPOP7OXYMMLXKGCYR2G62ROY4T/action/citation_signature","submit_replication":"https://pith.science/pith/FPOP7OXYMMLXKGCYR2G62ROY4T/action/replication_record"}},"created_at":"2026-07-05T02:48:09.844905+00:00","updated_at":"2026-07-05T02:48:09.844905+00:00"}