{ "schema_version": 1, "builder_plan": { "scenario": "You already need task-specific training and want to reduce its memory or compute cost.", "change": "Change only the training technique on a small controlled run after checking prerequisites.", "baseline": "Your current training recipe with the same model, data split, and evaluation.", "steps": [ "Locate the paper's implementation and check its license, model support, data needs, and hardware requirements. Stop here if you cannot obtain the required artifacts.", "Save a small set of representative product tasks and expected outcomes. Keep evaluation tasks out of any training or tuning, and choose quality and cost limits before running.", "Run the baseline and the proposed change on identical inputs. Keep other settings fixed and repeat paired runs to expose variation.", "Save per-task outputs, failures, timings, and all usage costs. Review regressions before expanding the test." ], "measure": "Held-out quality, peak memory, training time, and full compute cost to the target quality.", "decision": "Keep the change only if it meets your preselected quality and latency limits and improves the metric you care about after setup and operating costs. Otherwise retain the baseline. This is an exploration plan, not a validated recommendation." }, "builder_review": null, "arxiv_id": "2609.28988", "title": "Personalized Korean Lipreading as Visual Speech Recognition: Transfer, Census and Adaptation on OLKAVS", "summary": "The method replaces full per-user fine-tuning of a video-only Conformer VSR model with a low-rank adapter trained on minutes of the target user's frontal video, on top of English-initialized weights transferred to Korean. It costs a small adapter (4.6% of parameters) per user plus per-user data collection and training, and requires the multi-camera OLKAVS-style corpus and protocol to reproduce. It can fail on speakers with atypical articulation (per-speaker CER spans 1.0 to 52.2%), degrades about six CER points on cameras above the mouth plane, and its gains are validated only on one Korean corpus.", "rationale": "Unless the reader builds lipreading or visual speech products, this is a domain-specific system paper whose reusable lesson, LoRA-based per-user adaptation, is already standard practice.", "technique": "LoRA personalization of visual speech recognition", "family": "treino_eficiente", "recommendation": "nao_aplica", "published": "2026-09-24", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2609.28988", "independent_implementations": 0, "total_implementations": 1, "stars": 0, "citations": null, "claims": [ { "claim_id": "claim-01", "statement": "The method replaces full per-user fine-tuning of a video-only Conformer VSR model with a low-rank adapter trained on minutes of the target user's frontal video, on top of English-initialized weights transferred to Korean. It costs a small adapter (4.6% of parameters) per user plus per-user data collection and training, and requires the multi-camera OLKAVS-style corpus and protocol to reproduce. It can fail on speakers with atypical articulation (per-speaker CER spans 1.0 to 52.2%), degrades about six CER points on cameras above the mouth plane, and its gains are validated only on one Korean corpus.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.28988", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Low-rank adapter (4.6% of parameters, 4-29 minutes of user video) lowers CER of twelve high-error speakers by 2.13-3.58 points, retaining 85% of full fine-tuning gain at 12% of its cost; base model reaches 9.95-12.19% CER versus the published 26.64 baseline.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.28988", "source_page": null, "source_excerpt": "" } ], "exposure_map": [ { "dimension": "quality", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "compute", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "latency", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "operations", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "compatibility", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "security", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "data_and_training", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "reproducibility", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" } ], "risks": [], "minimum_test": [], "open_questions": [], "independent_tests": [], "lab_reproductions": [], "report_available": false, "validation_tier": "unknown", "evidence_tier": "unknown", "infrastructure_basis": "unknown", "equations": [], "equations_status": "not_fetched", "core_kind": "", "equations_fetched_at": "", "equations_source_url": "", "equations_source_sha256": "", "exposures_reviewed_at": "", "exposures_source_sha256": "" }