{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:7P6I7KSPTNWS6QGTEZKFVHANMX","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":"4551899e3bd2851564898ffb9d453f1d0b8e2c75228f165e25b7b279195de842","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-03-15T12:53:09Z","title_canon_sha256":"f8af1814d7a5dab1af988a2966084482dfe224fc54ad81de41a2a3ab0e8250c4"},"schema_version":"1.0","source":{"id":"2603.14360","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2603.14360","created_at":"2026-05-17T23:39:15Z"},{"alias_kind":"arxiv_version","alias_value":"2603.14360v2","created_at":"2026-05-17T23:39:15Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.14360","created_at":"2026-05-17T23:39:15Z"},{"alias_kind":"pith_short_12","alias_value":"7P6I7KSPTNWS","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"7P6I7KSPTNWS6QGT","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"7P6I7KSP","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:3d53363c3e1cdf9366db602e8fc02cae9251fbb9ea981200a17622bfcac04604","target":"graph","created_at":"2026-05-17T23:39:15Z","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":"M²RNN achieves perfect state tracking generalization at sequence lengths not seen during training. Hybrid M²RNN outperforms equivalent Gated DeltaNet hybrids by 0.4-0.5 perplexity points on a 7B MoE model while using 3× smaller state sizes for the recurrent layers."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the non-linear matrix-valued state transitions and state size expansion mechanism provide the claimed expressive power and efficiency gains without introducing training instability or hidden computational costs at scale."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"M²RNN achieves perfect state tracking at unseen lengths and outperforms Gated DeltaNet hybrids by 0.4-0.5 perplexity on 7B models with 3x smaller recurrent states."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Non-linear RNNs with matrix-valued states achieve perfect unseen-length state tracking and outperform equivalent attention hybrids by 0.4-0.5 perplexity points while using three times smaller recurrent states."}],"snapshot_sha256":"0ce4faf365470d0038d64ae0ba772f1335f485919b6313b1b9f7fcd21e69a076"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"Transformers are highly parallel but are limited to computations in the TC$^0$ complexity class, excluding tasks such as entity tracking and code execution that provably require greater expressive power. Motivated by this limitation, we revisit non-linear Recurrent Neural Networks (RNNs) for language modeling and introduce Matrix-to-Matrix RNN (M$^2$RNN): an architecture with matrix-valued hidden states and expressive non-linear state transitions. We demonstrate that the language modeling performance of non-linear RNNs is limited by their state size, and show how the state size expansion mecha","authors_text":"Ion Stoica, Joseph Gonzalez, Mayank Mishra, Shawn Tan, Tri Dao","cross_cats":["cs.AI"],"headline":"Non-linear RNNs with matrix-valued states achieve perfect unseen-length state tracking and outperform equivalent attention hybrids by 0.4-0.5 perplexity points while using three times smaller recurrent states.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-03-15T12:53:09Z","title":"M$^2$RNN: Non-Linear RNNs with Matrix-Valued States for Scalable Language Modeling"},"references":{"count":46,"internal_anchors":26,"resolved_work":46,"sample":[{"cited_arxiv_id":"","doi":"10.1145/3620665.3640366.https://doi.org/10.1145/3620665.3640366","is_internal_anchor":false,"ref_index":1,"title":"PyTorch 2: Faster machine learning through dynamic Python bytecode transformation and graph compilation","work_id":"abd261ad-4aca-4ea3-8090-344914daba35","year":null},{"cited_arxiv_id":"","doi":"10.1109/72.279181","is_internal_anchor":false,"ref_index":2,"title":"Simard, and Paolo Frasconi","work_id":"a6a7d71d-d4b9-4f4a-bd34-e010d6b09479","year":2019},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Language models are few-shot learners.Advances in neural information processing systems, 33:1877–1901","work_id":"84780adb-76cf-44ac-a8b7-e24d4fa5c592","year":1901},{"cited_arxiv_id":"1409.1259","doi":"","is_internal_anchor":true,"ref_index":4,"title":"On the properties of neural machine translation: Encoder-decoder approaches","work_id":"a6dc7801-0da7-423e-ba1d-00d246875f48","year":null},{"cited_arxiv_id":"1412.3555","doi":"","is_internal_anchor":true,"ref_index":5,"title":"Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling","work_id":"c7f2f5a9-ae4b-48db-aff0-24b9d0528995","year":null}],"snapshot_sha256":"801929d3413b61bd7956e72c02f99a2a95ae4cbdc001ebc78201f1e7f5d95ef8"},"source":{"id":"2603.14360","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-15T11:32:37.524446Z","id":"e83a3215-3f10-4d25-8adb-1f59661d8aa8","model_set":{"reader":"grok-4.3"},"one_line_summary":"M²RNN achieves perfect state tracking at unseen lengths and outperforms Gated DeltaNet hybrids by 0.4-0.5 perplexity on 7B models with 3x smaller recurrent states.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Non-linear RNNs with matrix-valued states achieve perfect unseen-length state tracking and outperform equivalent attention hybrids by 0.4-0.5 perplexity points while using three times smaller recurrent states.","strongest_claim":"M²RNN achieves perfect state tracking generalization at sequence lengths not seen during training. Hybrid M²RNN outperforms equivalent Gated DeltaNet hybrids by 0.4-0.5 perplexity points on a 7B MoE model while using 3× smaller state sizes for the recurrent layers.","weakest_assumption":"That the non-linear matrix-valued state transitions and state size expansion mechanism provide the claimed expressive power and efficiency gains without introducing training instability or hidden computational costs at scale."}},"verdict_id":"e83a3215-3f10-4d25-8adb-1f59661d8aa8"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:ff148016b014b85ee28d993902a1b41d0db0546609bc5f4622e1ad3bf759e9ae","target":"record","created_at":"2026-05-17T23:39:15Z","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":"4551899e3bd2851564898ffb9d453f1d0b8e2c75228f165e25b7b279195de842","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-03-15T12:53:09Z","title_canon_sha256":"f8af1814d7a5dab1af988a2966084482dfe224fc54ad81de41a2a3ab0e8250c4"},"schema_version":"1.0","source":{"id":"2603.14360","kind":"arxiv","version":2}},"canonical_sha256":"fbfc8faa4f9b6d2f40d326545a9c0d65e2103a79d75ec61db5d8d3cc40a2c23a","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"fbfc8faa4f9b6d2f40d326545a9c0d65e2103a79d75ec61db5d8d3cc40a2c23a","first_computed_at":"2026-05-17T23:39:15.774513Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:39:15.774513Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"9whNPXQ2l9RA9rSleqQ5JJ5H1bK5p8yvNX7ffCXuJO/uWdn0JIu0HLtz9qay93f1mOtW599z96aqnJcMh/CSBQ==","signature_status":"signed_v1","signed_at":"2026-05-17T23:39:15.775332Z","signed_message":"canonical_sha256_bytes"},"source_id":"2603.14360","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:ff148016b014b85ee28d993902a1b41d0db0546609bc5f4622e1ad3bf759e9ae","sha256:3d53363c3e1cdf9366db602e8fc02cae9251fbb9ea981200a17622bfcac04604"],"state_sha256":"922bcbd973ebd478dc4ed4af5af6590bd0e5b1bd684c5b8856360e95c39fa43f"}