{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:JY2T5G5DNIZKCVUDOA4ENRN7YX","short_pith_number":"pith:JY2T5G5D","canonical_record":{"source":{"id":"2605.11855","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-12T09:39:33Z","cross_cats_sorted":["cs.AI","cs.AR"],"title_canon_sha256":"e57ee3fe3e48df1b2c310a9378a729f935ff1efbb5bd2560a3c2656b2363be70","abstract_canon_sha256":"df8bc267d63eb68ba85c3b83def5c2d9be5652246fc94fde32e8cf9556e8d928"},"schema_version":"1.0"},"canonical_sha256":"4e353e9ba36a32a15683703846c5bfc5f6d32f01c5d74d5665387df61e95ffc6","source":{"kind":"arxiv","id":"2605.11855","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.11855","created_at":"2026-06-09T01:05:19Z"},{"alias_kind":"arxiv_version","alias_value":"2605.11855v2","created_at":"2026-06-09T01:05:19Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.11855","created_at":"2026-06-09T01:05:19Z"},{"alias_kind":"pith_short_12","alias_value":"JY2T5G5DNIZK","created_at":"2026-06-09T01:05:19Z"},{"alias_kind":"pith_short_16","alias_value":"JY2T5G5DNIZKCVUD","created_at":"2026-06-09T01:05:19Z"},{"alias_kind":"pith_short_8","alias_value":"JY2T5G5D","created_at":"2026-06-09T01:05:19Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:JY2T5G5DNIZKCVUDOA4ENRN7YX","target":"record","payload":{"canonical_record":{"source":{"id":"2605.11855","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-12T09:39:33Z","cross_cats_sorted":["cs.AI","cs.AR"],"title_canon_sha256":"e57ee3fe3e48df1b2c310a9378a729f935ff1efbb5bd2560a3c2656b2363be70","abstract_canon_sha256":"df8bc267d63eb68ba85c3b83def5c2d9be5652246fc94fde32e8cf9556e8d928"},"schema_version":"1.0"},"canonical_sha256":"4e353e9ba36a32a15683703846c5bfc5f6d32f01c5d74d5665387df61e95ffc6","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-09T01:05:19.089705Z","signature_b64":"HzgjolWIGw66i+Fnu59d6omN3UvBNDR766Sggfoaqqlgf3TSNvZF9EVhE5VjHp5lEsE1+kKyhmFerMXxy6v/Dg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4e353e9ba36a32a15683703846c5bfc5f6d32f01c5d74d5665387df61e95ffc6","last_reissued_at":"2026-06-09T01:05:19.089260Z","signature_status":"signed_v1","first_computed_at":"2026-06-09T01:05:19.089260Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.11855","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-09T01:05:19Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"2ct1s61TltsKkg/l2e1egiqq0iYviH0O+BYaum1sJFYQLLJgcsQeN6BXclxCgEy7bSKYPO/FKcctyDCm+HBFBw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-18T02:17:15.224080Z"},"content_sha256":"0ffdff45acbef9c354a6474a0556cdd9c0c36812af8e2dc950dc287042abfcaf","schema_version":"1.0","event_id":"sha256:0ffdff45acbef9c354a6474a0556cdd9c0c36812af8e2dc950dc287042abfcaf"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:JY2T5G5DNIZKCVUDOA4ENRN7YX","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Improving the Performance and Learning Stability of Parallelizable RNNs Designed for Ultra-Low Power Applications","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"The cumulative update formulation in CMRU and αCMRU restores gradient flow while preserving bistable memory for ultra-low power RNNs.","cross_cats":["cs.AI","cs.AR"],"primary_cat":"cs.LG","authors_text":"Arthur Fyon, Damien Ernst, Guillaume Drion, Julien Brandoit","submitted_at":"2026-05-12T09:39:33Z","abstract_excerpt":"Sequence learning is dominated by Transformers and parallelizable recurrent neural networks (RNNs) such as state-space models, yet learning long-term dependencies remains challenging, and state-of-the-art designs trade power consumption for performance. The Bistable Memory Recurrent Unit (BMRU) was introduced to enable hardware-software co-design of ultra-low power RNNs: quantized states with hysteresis provide persistent memory while mapping directly to analog primitives. However, BMRU performance lags behind parallelizable RNNs on complex sequential tasks. In this paper, we identify gradient"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Experiments show that the cumulative formulation dramatically improves convergence stability and reduces initialization sensitivity. The CMRU and αCMRU match or outperform Linear Recurrent Units (LRUs) and minimal Gated Recurrent Units (minGRUs) across diverse benchmarks at small model sizes, with particular advantages on tasks requiring discrete long-range retention, while the CMRU retains quantized states, persistent memory, and noise-resilient dynamics essential for analog implementation.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the cumulative update formulation preserves the quantized states with hysteresis, persistent memory, and noise-resilient dynamics required for direct analog hardware mapping while restoring gradient flow.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Cumulative state updates in CMRU restore gradient flow through time in quantized bistable RNNs, yielding more stable convergence and competitive or superior performance versus LRUs and minGRUs on long-range sequence tasks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"The cumulative update formulation in CMRU and αCMRU restores gradient flow while preserving bistable memory for ultra-low power RNNs.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"38c8565bdac0a2fbacfd4e55a4b1d74fb1272592a16bb32da7f77a02b0940571"},"source":{"id":"2605.11855","kind":"arxiv","version":2},"verdict":{"id":"3309673f-1cd4-4736-806a-1a9c54dc7e54","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-13T06:28:40.442295Z","strongest_claim":"Experiments show that the cumulative formulation dramatically improves convergence stability and reduces initialization sensitivity. The CMRU and αCMRU match or outperform Linear Recurrent Units (LRUs) and minimal Gated Recurrent Units (minGRUs) across diverse benchmarks at small model sizes, with particular advantages on tasks requiring discrete long-range retention, while the CMRU retains quantized states, persistent memory, and noise-resilient dynamics essential for analog implementation.","one_line_summary":"Cumulative state updates in CMRU restore gradient flow through time in quantized bistable RNNs, yielding more stable convergence and competitive or superior performance versus LRUs and minGRUs on long-range sequence tasks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the cumulative update formulation preserves the quantized states with hysteresis, persistent memory, and noise-resilient dynamics required for direct analog hardware mapping while restoring gradient flow.","pith_extraction_headline":"The cumulative update formulation in CMRU and αCMRU restores gradient flow while preserving bistable memory for ultra-low power RNNs."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.11855/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-26T16:42:10.424676Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-20T18:31:27.329327Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-20T12:46:31.131884Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-20T03:22:00.418294Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"e4a66f9c05e363780f6ef9e07f36ea7742c42b10c8aae654d7ff80317aa09721"},"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"},"verdict_id":"3309673f-1cd4-4736-806a-1a9c54dc7e54"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-09T01:05:19Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"7RsP4UyfrCl6VLTPMzhuf4PoXDzpHQbqM3/ctFiQjfQIEYLwZrULKjsEQkIzpto00JKSXNT0re9Abygn2Qp3BQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-18T02:17:15.224617Z"},"content_sha256":"91ef8bec2015ac1eabb362c60aeeab62b1aee43e67fb301f7292d04dcb0780cf","schema_version":"1.0","event_id":"sha256:91ef8bec2015ac1eabb362c60aeeab62b1aee43e67fb301f7292d04dcb0780cf"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/JY2T5G5DNIZKCVUDOA4ENRN7YX/bundle.json","state_url":"https://pith.science/pith/JY2T5G5DNIZKCVUDOA4ENRN7YX/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/JY2T5G5DNIZKCVUDOA4ENRN7YX/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":"2026-07-18T02:17:15Z","links":{"resolver":"https://pith.science/pith/JY2T5G5DNIZKCVUDOA4ENRN7YX","bundle":"https://pith.science/pith/JY2T5G5DNIZKCVUDOA4ENRN7YX/bundle.json","state":"https://pith.science/pith/JY2T5G5DNIZKCVUDOA4ENRN7YX/state.json","well_known_bundle":"https://pith.science/.well-known/pith/JY2T5G5DNIZKCVUDOA4ENRN7YX/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:JY2T5G5DNIZKCVUDOA4ENRN7YX","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":"df8bc267d63eb68ba85c3b83def5c2d9be5652246fc94fde32e8cf9556e8d928","cross_cats_sorted":["cs.AI","cs.AR"],"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-12T09:39:33Z","title_canon_sha256":"e57ee3fe3e48df1b2c310a9378a729f935ff1efbb5bd2560a3c2656b2363be70"},"schema_version":"1.0","source":{"id":"2605.11855","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.11855","created_at":"2026-06-09T01:05:19Z"},{"alias_kind":"arxiv_version","alias_value":"2605.11855v2","created_at":"2026-06-09T01:05:19Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.11855","created_at":"2026-06-09T01:05:19Z"},{"alias_kind":"pith_short_12","alias_value":"JY2T5G5DNIZK","created_at":"2026-06-09T01:05:19Z"},{"alias_kind":"pith_short_16","alias_value":"JY2T5G5DNIZKCVUD","created_at":"2026-06-09T01:05:19Z"},{"alias_kind":"pith_short_8","alias_value":"JY2T5G5D","created_at":"2026-06-09T01:05:19Z"}],"graph_snapshots":[{"event_id":"sha256:91ef8bec2015ac1eabb362c60aeeab62b1aee43e67fb301f7292d04dcb0780cf","target":"graph","created_at":"2026-06-09T01:05:19Z","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":"Experiments show that the cumulative formulation dramatically improves convergence stability and reduces initialization sensitivity. The CMRU and αCMRU match or outperform Linear Recurrent Units (LRUs) and minimal Gated Recurrent Units (minGRUs) across diverse benchmarks at small model sizes, with particular advantages on tasks requiring discrete long-range retention, while the CMRU retains quantized states, persistent memory, and noise-resilient dynamics essential for analog implementation."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the cumulative update formulation preserves the quantized states with hysteresis, persistent memory, and noise-resilient dynamics required for direct analog hardware mapping while restoring gradient flow."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Cumulative state updates in CMRU restore gradient flow through time in quantized bistable RNNs, yielding more stable convergence and competitive or superior performance versus LRUs and minGRUs on long-range sequence tasks."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"The cumulative update formulation in CMRU and αCMRU restores gradient flow while preserving bistable memory for ultra-low power RNNs."}],"snapshot_sha256":"38c8565bdac0a2fbacfd4e55a4b1d74fb1272592a16bb32da7f77a02b0940571"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-26T16:42:10.424676Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-20T18:31:27.329327Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-20T12:46:31.131884Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"claim_evidence","ran_at":"2026-05-20T03:22:00.418294Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2605.11855/integrity.json","findings":[],"snapshot_sha256":"e4a66f9c05e363780f6ef9e07f36ea7742c42b10c8aae654d7ff80317aa09721","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Sequence learning is dominated by Transformers and parallelizable recurrent neural networks (RNNs) such as state-space models, yet learning long-term dependencies remains challenging, and state-of-the-art designs trade power consumption for performance. The Bistable Memory Recurrent Unit (BMRU) was introduced to enable hardware-software co-design of ultra-low power RNNs: quantized states with hysteresis provide persistent memory while mapping directly to analog primitives. However, BMRU performance lags behind parallelizable RNNs on complex sequential tasks. In this paper, we identify gradient","authors_text":"Arthur Fyon, Damien Ernst, Guillaume Drion, Julien Brandoit","cross_cats":["cs.AI","cs.AR"],"headline":"The cumulative update formulation in CMRU and αCMRU restores gradient flow while preserving bistable memory for ultra-low power RNNs.","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-12T09:39:33Z","title":"Improving the Performance and Learning Stability of Parallelizable RNNs Designed for Ultra-Low Power Applications"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.11855","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-13T06:28:40.442295Z","id":"3309673f-1cd4-4736-806a-1a9c54dc7e54","model_set":{"reader":"grok-4.3"},"one_line_summary":"Cumulative state updates in CMRU restore gradient flow through time in quantized bistable RNNs, yielding more stable convergence and competitive or superior performance versus LRUs and minGRUs on long-range sequence tasks.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"The cumulative update formulation in CMRU and αCMRU restores gradient flow while preserving bistable memory for ultra-low power RNNs.","strongest_claim":"Experiments show that the cumulative formulation dramatically improves convergence stability and reduces initialization sensitivity. The CMRU and αCMRU match or outperform Linear Recurrent Units (LRUs) and minimal Gated Recurrent Units (minGRUs) across diverse benchmarks at small model sizes, with particular advantages on tasks requiring discrete long-range retention, while the CMRU retains quantized states, persistent memory, and noise-resilient dynamics essential for analog implementation.","weakest_assumption":"That the cumulative update formulation preserves the quantized states with hysteresis, persistent memory, and noise-resilient dynamics required for direct analog hardware mapping while restoring gradient flow."}},"verdict_id":"3309673f-1cd4-4736-806a-1a9c54dc7e54"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:0ffdff45acbef9c354a6474a0556cdd9c0c36812af8e2dc950dc287042abfcaf","target":"record","created_at":"2026-06-09T01:05:19Z","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":"df8bc267d63eb68ba85c3b83def5c2d9be5652246fc94fde32e8cf9556e8d928","cross_cats_sorted":["cs.AI","cs.AR"],"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-12T09:39:33Z","title_canon_sha256":"e57ee3fe3e48df1b2c310a9378a729f935ff1efbb5bd2560a3c2656b2363be70"},"schema_version":"1.0","source":{"id":"2605.11855","kind":"arxiv","version":2}},"canonical_sha256":"4e353e9ba36a32a15683703846c5bfc5f6d32f01c5d74d5665387df61e95ffc6","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"4e353e9ba36a32a15683703846c5bfc5f6d32f01c5d74d5665387df61e95ffc6","first_computed_at":"2026-06-09T01:05:19.089260Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-06-09T01:05:19.089260Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"HzgjolWIGw66i+Fnu59d6omN3UvBNDR766Sggfoaqqlgf3TSNvZF9EVhE5VjHp5lEsE1+kKyhmFerMXxy6v/Dg==","signature_status":"signed_v1","signed_at":"2026-06-09T01:05:19.089705Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.11855","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:0ffdff45acbef9c354a6474a0556cdd9c0c36812af8e2dc950dc287042abfcaf","sha256:91ef8bec2015ac1eabb362c60aeeab62b1aee43e67fb301f7292d04dcb0780cf"],"state_sha256":"df1f5f35bfbac3979271e683e01008518b6604b2c00ffd0692748d29114c0666"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"OOlLFldrAdo2+ECnSQeN1YvVio+z3jVhQ537amrIF68oScgXlEicelDWHH3+wXwGUMpIOOvEz4xlEa/+Du1yBw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-07-18T02:17:15.227159Z","bundle_sha256":"5eef23615e6e240676256d5de40ef46d4b0206b27f5644a28c97028dc72c3899"}}