{"paper":{"title":"Latent Chain-of-Thought Improves Structured-Data Transformers","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Latent chain-of-thought via recurrent feedback tokens improves transformer performance on time-series forecasting and tabular prediction.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Carson Dudley, Samet Oymak","submitted_at":"2026-05-11T21:41:21Z","abstract_excerpt":"Chain-of-thought and more broadly test-time compute are known to augment the expressive capabilities of language models and have led to major innovations in reasoning. Motivated by this success, this paper explores latent chain-of-thought as well as the impact of depth and looping for time-series and tabular data. We propose a recurrent scheme in which a structured-data transformer, after an initial forward pass, compresses its query-position hidden states into feedback tokens that are appended to the input and processed again, allowing multiple rounds of latent computation before prediction. "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Across 36 datasets in time-series forecasting and tabular prediction, latent chain-of-thought improves over the baseline on 8/9 time-series datasets (+10.99% average gain) and 22/27 tabular datasets (+5.31% average gain). Across both settings, the CoT models perform the best on average.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the observed gains are attributable to the latent chain-of-thought mechanism rather than incidental effects of the recurrent architecture or token compression, and that the gains generalize beyond the 36 chosen datasets and the specific transformer backbone used.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Latent chain-of-thought via recurrent feedback tokens improves average performance of structured-data transformers on time-series forecasting and tabular prediction.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Latent chain-of-thought via recurrent feedback tokens improves transformer performance on time-series forecasting and tabular prediction.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"809c745b81c3c31b80020bbcc52821cb3720a43e7dd175c6434d061bfa65f733"},"source":{"id":"2605.11262","kind":"arxiv","version":2},"verdict":{"id":"33656d52-67b7-43ea-ab99-bc13e1562b6d","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-13T01:44:46.292821Z","strongest_claim":"Across 36 datasets in time-series forecasting and tabular prediction, latent chain-of-thought improves over the baseline on 8/9 time-series datasets (+10.99% average gain) and 22/27 tabular datasets (+5.31% average gain). Across both settings, the CoT models perform the best on average.","one_line_summary":"Latent chain-of-thought via recurrent feedback tokens improves average performance of structured-data transformers on time-series forecasting and tabular prediction.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the observed gains are attributable to the latent chain-of-thought mechanism rather than incidental effects of the recurrent architecture or token compression, and that the gains generalize beyond the 36 chosen datasets and the specific transformer backbone used.","pith_extraction_headline":"Latent chain-of-thought via recurrent feedback tokens improves transformer performance on time-series forecasting and tabular prediction."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.11262/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-19T12:39:32.204627Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T10:01:17.285704Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T08:37:05.507499Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"8aff626376bba00443f92e7cf6ce5eb43e0aaa6b1fb6e9a2cfd6dae3aacfecd0"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"88be73c0dcace7f2a750dc157e773d0d2151fb00d8f55b2ff7e35f3fe33fa191"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}