{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:JJI62ME3T6PTVDGRYFAXV3BHRV","short_pith_number":"pith:JJI62ME3","canonical_record":{"source":{"id":"2602.04872","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"stat.ML","submitted_at":"2026-02-04T18:57:30Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"3b66e499bb2879ef8b8754f215abc4dab261f169fcd25d3b593e742f91ca5fc0","abstract_canon_sha256":"34aae90f259db7936db6d4d6b26940b47819b3ebe00f0409cfcb35d84b7d68e0"},"schema_version":"1.0"},"canonical_sha256":"4a51ed309b9f9f3a8cd1c1417aec278d4499d3b60ee4b19be1a550c581a83bf5","source":{"kind":"arxiv","id":"2602.04872","version":3},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2602.04872","created_at":"2026-05-20T00:04:24Z"},{"alias_kind":"arxiv_version","alias_value":"2602.04872v3","created_at":"2026-05-20T00:04:24Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.04872","created_at":"2026-05-20T00:04:24Z"},{"alias_kind":"pith_short_12","alias_value":"JJI62ME3T6PT","created_at":"2026-05-20T00:04:24Z"},{"alias_kind":"pith_short_16","alias_value":"JJI62ME3T6PTVDGR","created_at":"2026-05-20T00:04:24Z"},{"alias_kind":"pith_short_8","alias_value":"JJI62ME3","created_at":"2026-05-20T00:04:24Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:JJI62ME3T6PTVDGRYFAXV3BHRV","target":"record","payload":{"canonical_record":{"source":{"id":"2602.04872","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"stat.ML","submitted_at":"2026-02-04T18:57:30Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"3b66e499bb2879ef8b8754f215abc4dab261f169fcd25d3b593e742f91ca5fc0","abstract_canon_sha256":"34aae90f259db7936db6d4d6b26940b47819b3ebe00f0409cfcb35d84b7d68e0"},"schema_version":"1.0"},"canonical_sha256":"4a51ed309b9f9f3a8cd1c1417aec278d4499d3b60ee4b19be1a550c581a83bf5","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:04:24.357395Z","signature_b64":"fpp+M/OD67neAccfs6rXun7eGtRcVz9KQ8hgCO4ajE59Ik2ZIHQRQHjQL0Uu4XDvvdN0p/m59TZCsCueevEODg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4a51ed309b9f9f3a8cd1c1417aec278d4499d3b60ee4b19be1a550c581a83bf5","last_reissued_at":"2026-05-20T00:04:24.356727Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:04:24.356727Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2602.04872","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":"2026-05-20T00:04:24Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"1+vNlXnBVLD5duNwA3Q6mDC0qgRInMSyUDsUR0bFwPt4LLTeSSsn82aSKWcVrppLQeB8ugJRd8gq46mpMNPXBg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-28T14:45:43.246729Z"},"content_sha256":"7fa909dd661f1b7b0425d489a6dba668fef32373b51e67a3a0a05fbcac0ae117","schema_version":"1.0","event_id":"sha256:7fa909dd661f1b7b0425d489a6dba668fef32373b51e67a3a0a05fbcac0ae117"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:JJI62ME3T6PTVDGRYFAXV3BHRV","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Multi-layer Cross-attention is Provably Optimal for Multi-modal In-context Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Multi-layer cross-attention recovers the Bayes-optimal predictor for multi-modal in-context learning when data follows a latent factor model.","cross_cats":["cs.AI","cs.LG"],"primary_cat":"stat.ML","authors_text":"Nicholas Barnfield, Pragya Sur, Subhabrata Sen","submitted_at":"2026-02-04T18:57:30Z","abstract_excerpt":"Recent progress has rapidly advanced our understanding of the mechanisms underlying in-context learning in modern attention-based neural networks. However, existing results focus exclusively on unimodal data; in contrast, the theoretical underpinnings of in-context learning for multi-modal data remain poorly understood. We introduce a mathematically tractable framework for studying multi-modal learning and explore when transformer-like architectures can recover Bayes-optimal performance in-context. To model multi-modal problems, we assume the observed data arises from a latent factor model. Ou"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We show that this cross-attention mechanism is provably Bayes optimal when optimized using gradient flow.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The observed data arises from a latent factor model, and both the number of cross-attention layers and the context length are large.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Multi-layer linearized cross-attention is provably Bayes-optimal for multi-modal in-context learning under a latent factor model when optimized via gradient flow in the large-layer large-context limit.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Multi-layer cross-attention recovers the Bayes-optimal predictor for multi-modal in-context learning when data follows a latent factor model.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"a5953fe48507101dfc9968645cb13884df13533f2822dae986292b670c8670b3"},"source":{"id":"2602.04872","kind":"arxiv","version":3},"verdict":{"id":"4fd1ec84-2754-4e12-b27e-17d44645249f","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T06:36:47.408720Z","strongest_claim":"We show that this cross-attention mechanism is provably Bayes optimal when optimized using gradient flow.","one_line_summary":"Multi-layer linearized cross-attention is provably Bayes-optimal for multi-modal in-context learning under a latent factor model when optimized via gradient flow in the large-layer large-context limit.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The observed data arises from a latent factor model, and both the number of cross-attention layers and the context length are large.","pith_extraction_headline":"Multi-layer cross-attention recovers the Bayes-optimal predictor for multi-modal in-context learning when data follows a latent factor model."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2602.04872/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":2,"snapshot_sha256":"3743b31f769ead8244014d30684a55648c67fcd1d68b40aba93b31d919f12ead"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"4fd1ec84-2754-4e12-b27e-17d44645249f"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-20T00:04:24Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"MSfPtsNhwmh/ChOwP3kWbGMe9Z6HFrRYOfeUe6741bYh3/wUuFMH5pTZl6dwMcIx14ODdX4lj1vp78jKDAcYCA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-28T14:45:43.247189Z"},"content_sha256":"71fe908495a884c11c5c635ddac667fcd00a9255b649f1ab6582a0068655faf2","schema_version":"1.0","event_id":"sha256:71fe908495a884c11c5c635ddac667fcd00a9255b649f1ab6582a0068655faf2"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/JJI62ME3T6PTVDGRYFAXV3BHRV/bundle.json","state_url":"https://pith.science/pith/JJI62ME3T6PTVDGRYFAXV3BHRV/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/JJI62ME3T6PTVDGRYFAXV3BHRV/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":"2026-05-28T14:45:43Z","links":{"resolver":"https://pith.science/pith/JJI62ME3T6PTVDGRYFAXV3BHRV","bundle":"https://pith.science/pith/JJI62ME3T6PTVDGRYFAXV3BHRV/bundle.json","state":"https://pith.science/pith/JJI62ME3T6PTVDGRYFAXV3BHRV/state.json","well_known_bundle":"https://pith.science/.well-known/pith/JJI62ME3T6PTVDGRYFAXV3BHRV/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:JJI62ME3T6PTVDGRYFAXV3BHRV","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":"34aae90f259db7936db6d4d6b26940b47819b3ebe00f0409cfcb35d84b7d68e0","cross_cats_sorted":["cs.AI","cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"stat.ML","submitted_at":"2026-02-04T18:57:30Z","title_canon_sha256":"3b66e499bb2879ef8b8754f215abc4dab261f169fcd25d3b593e742f91ca5fc0"},"schema_version":"1.0","source":{"id":"2602.04872","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2602.04872","created_at":"2026-05-20T00:04:24Z"},{"alias_kind":"arxiv_version","alias_value":"2602.04872v3","created_at":"2026-05-20T00:04:24Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.04872","created_at":"2026-05-20T00:04:24Z"},{"alias_kind":"pith_short_12","alias_value":"JJI62ME3T6PT","created_at":"2026-05-20T00:04:24Z"},{"alias_kind":"pith_short_16","alias_value":"JJI62ME3T6PTVDGR","created_at":"2026-05-20T00:04:24Z"},{"alias_kind":"pith_short_8","alias_value":"JJI62ME3","created_at":"2026-05-20T00:04:24Z"}],"graph_snapshots":[{"event_id":"sha256:71fe908495a884c11c5c635ddac667fcd00a9255b649f1ab6582a0068655faf2","target":"graph","created_at":"2026-05-20T00:04:24Z","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":"We show that this cross-attention mechanism is provably Bayes optimal when optimized using gradient flow."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The observed data arises from a latent factor model, and both the number of cross-attention layers and the context length are large."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Multi-layer linearized cross-attention is provably Bayes-optimal for multi-modal in-context learning under a latent factor model when optimized via gradient flow in the large-layer large-context limit."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Multi-layer cross-attention recovers the Bayes-optimal predictor for multi-modal in-context learning when data follows a latent factor model."}],"snapshot_sha256":"a5953fe48507101dfc9968645cb13884df13533f2822dae986292b670c8670b3"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"3743b31f769ead8244014d30684a55648c67fcd1d68b40aba93b31d919f12ead"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2602.04872/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Recent progress has rapidly advanced our understanding of the mechanisms underlying in-context learning in modern attention-based neural networks. However, existing results focus exclusively on unimodal data; in contrast, the theoretical underpinnings of in-context learning for multi-modal data remain poorly understood. We introduce a mathematically tractable framework for studying multi-modal learning and explore when transformer-like architectures can recover Bayes-optimal performance in-context. To model multi-modal problems, we assume the observed data arises from a latent factor model. Ou","authors_text":"Nicholas Barnfield, Pragya Sur, Subhabrata Sen","cross_cats":["cs.AI","cs.LG"],"headline":"Multi-layer cross-attention recovers the Bayes-optimal predictor for multi-modal in-context learning when data follows a latent factor model.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"stat.ML","submitted_at":"2026-02-04T18:57:30Z","title":"Multi-layer Cross-attention is Provably Optimal for Multi-modal In-context Learning"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.04872","kind":"arxiv","version":3},"verdict":{"created_at":"2026-05-16T06:36:47.408720Z","id":"4fd1ec84-2754-4e12-b27e-17d44645249f","model_set":{"reader":"grok-4.3"},"one_line_summary":"Multi-layer linearized cross-attention is provably Bayes-optimal for multi-modal in-context learning under a latent factor model when optimized via gradient flow in the large-layer large-context limit.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Multi-layer cross-attention recovers the Bayes-optimal predictor for multi-modal in-context learning when data follows a latent factor model.","strongest_claim":"We show that this cross-attention mechanism is provably Bayes optimal when optimized using gradient flow.","weakest_assumption":"The observed data arises from a latent factor model, and both the number of cross-attention layers and the context length are large."}},"verdict_id":"4fd1ec84-2754-4e12-b27e-17d44645249f"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:7fa909dd661f1b7b0425d489a6dba668fef32373b51e67a3a0a05fbcac0ae117","target":"record","created_at":"2026-05-20T00:04:24Z","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":"34aae90f259db7936db6d4d6b26940b47819b3ebe00f0409cfcb35d84b7d68e0","cross_cats_sorted":["cs.AI","cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"stat.ML","submitted_at":"2026-02-04T18:57:30Z","title_canon_sha256":"3b66e499bb2879ef8b8754f215abc4dab261f169fcd25d3b593e742f91ca5fc0"},"schema_version":"1.0","source":{"id":"2602.04872","kind":"arxiv","version":3}},"canonical_sha256":"4a51ed309b9f9f3a8cd1c1417aec278d4499d3b60ee4b19be1a550c581a83bf5","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"4a51ed309b9f9f3a8cd1c1417aec278d4499d3b60ee4b19be1a550c581a83bf5","first_computed_at":"2026-05-20T00:04:24.356727Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-20T00:04:24.356727Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"fpp+M/OD67neAccfs6rXun7eGtRcVz9KQ8hgCO4ajE59Ik2ZIHQRQHjQL0Uu4XDvvdN0p/m59TZCsCueevEODg==","signature_status":"signed_v1","signed_at":"2026-05-20T00:04:24.357395Z","signed_message":"canonical_sha256_bytes"},"source_id":"2602.04872","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:7fa909dd661f1b7b0425d489a6dba668fef32373b51e67a3a0a05fbcac0ae117","sha256:71fe908495a884c11c5c635ddac667fcd00a9255b649f1ab6582a0068655faf2"],"state_sha256":"d96042e900656292b43590de3d329b96c8af39c3265b203072c553ef0832f249"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"EjivZ+MDjAUJcQawhH2fYxwy32+f7bDxR/xDieqnaVIr6mbiZHl/xWSsc3VkObbaNssrSABj1p9Vdid7bBNkCg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-28T14:45:43.249470Z","bundle_sha256":"f09e98623411d6d1b5ae1ce41e010e31e5730ac61cfc3eabbc824f8bc282a132"}}