File size: 4,753 Bytes
3d7e603 eed9594 3d7e603 eed9594 3d7e603 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 | {
"schema_version": 1,
"title": "Reproduction: Rational Transductors",
"emoji": "🎯",
"space_id": "vimarsh/repro-rational-transductors",
"paper": {
"arxiv_id": "2602.07599"
},
"tags": [
"icml2026-repro",
"paper-uEZpyELNuB"
],
"updated_at": "2026-07-19T10:06:40+00:00",
"root": {
"slug": "index",
"title": "Reproduction: Rational Transductors",
"file": "pages/index.md",
"children": [
{
"slug": "executive-summary",
"title": "Executive summary",
"file": "pages/executive-summary/page.md",
"children": []
},
{
"slug": "claim-1-rational-transductors-augment-a-transformer-stream-with-matrix-valued-recurrence-derived-from-weighted-finite-automata-wfa-injecting-this-signal-into-attention-via-a-deep-rational-injection-mechanism-section-2",
"title": "Claim 1: Rational Transductors augment a Transformer stream with matrix-valued recurrence derived from Weighted Finite Automata (WFA), injecting this signal into attention via a 'Deep Rational Injection' mechanism (Section 2).",
"file": "pages/claim-1-rational-transductors-augment-a-transformer-stream-with-matrix-valued-recurrence-derived-from-weighted-finite-automata-wfa-injecting-this-signal-into-attention-via-a-deep-rational-injection-mechanism-section-2/page.md",
"children": []
},
{
"slug": "claim-2-the-architecture-is-shown-to-capture-regular-languages-and-nc1-complete-problems-such-as-boolean-formula-evaluation-in-contrast-to-standard-self-attention-which-is-limited-to-ac0-under-hard-attention-or-tc0-under-soft-attention-sections-4-7",
"title": "Claim 2: The architecture is shown to capture Regular Languages and NC1-complete problems such as Boolean Formula Evaluation, in contrast to standard self-attention which is limited to AC0 under hard attention or TC0 under soft attention (Sections 4-7).",
"file": "pages/claim-2-the-architecture-is-shown-to-capture-regular-languages-and-nc1-complete-problems-such-as-boolean-formula-evaluation-in-contrast-to-standard-self-attention-which-is-limited-to-ac0-under-hard-attention-or-tc0-under-soft-attention-sections-4-7/page.md",
"children": []
},
{
"slug": "claim-3-random-rational-features-are-introduced-as-a-universal-basis-for-sequential-dependencies-while-differentiable-rational-features-are-introduced-to-close-the-remaining-representational-gap-of-the-rational-feature-approximation-section-3",
"title": "Claim 3: Random Rational Features are introduced as a universal basis for sequential dependencies, while Differentiable Rational Features are introduced to close the remaining representational gap of the rational-feature approximation (Section 3).",
"file": "pages/claim-3-random-rational-features-are-introduced-as-a-universal-basis-for-sequential-dependencies-while-differentiable-rational-features-are-introduced-to-close-the-remaining-representational-gap-of-the-rational-feature-approximation-section-3/page.md",
"children": []
},
{
"slug": "claim-4-the-proposed-architecture-preserves-a-parallel-training-time-complexity-of-o-l-log-t-in-sequence-length-t-retaining-the-parallelization-advantage-of-transformers-over-strictly-sequential-rnn-computation-section-2",
"title": "Claim 4: The proposed architecture preserves a parallel training time complexity of O(L + log T) in sequence length T, retaining the parallelization advantage of Transformers over strictly sequential RNN computation (Section 2).",
"file": "pages/claim-4-the-proposed-architecture-preserves-a-parallel-training-time-complexity-of-o-l-log-t-in-sequence-length-t-retaining-the-parallelization-advantage-of-transformers-over-strictly-sequential-rnn-computation-section-2/page.md",
"children": []
},
{
"slug": "claim-5-experiments-demonstrate-robust-length-generalization-on-algorithmic-tasks-requiring-state-tracking-the-regular-gap-closing-a-gap-where-standard-transformers-fail-appendix-b",
"title": "Claim 5: Experiments demonstrate robust length generalization on algorithmic tasks requiring state tracking (the 'Regular Gap'), closing a gap where standard Transformers fail (Appendix B).",
"file": "pages/claim-5-experiments-demonstrate-robust-length-generalization-on-algorithmic-tasks-requiring-state-tracking-the-regular-gap-closing-a-gap-where-standard-transformers-fail-appendix-b/page.md",
"children": []
},
{
"slug": "conclusion",
"title": "Conclusion",
"file": "pages/conclusion/page.md",
"children": []
}
]
},
"agent_view_tokens": 6188,
"revision": "1784455600701660489"
} |