{ "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": "2609.15871", "title": "LLM-Based Schema-Aware Split Learning for Privacy-Preserving Mental Distress Prediction Across Heterogeneous Surveys", "summary": "Replaces federated learning over raw survey data by serializing records into natural language and splitting a LoRA-fine-tuned LLaMA-3.2-3B between clients and a server, so raw records never leave each institution. Costs include server-side hosting of the LLM backbone, per-record LLM serialization and forward passes, and the orchestration complexity of split-training across institutions. It can fail through privacy leakage via intermediate activations, semantic loss when surveys serialize ambiguously, and dependence on a central server that reintroduces a trust and infrastructure assumption FL avoids.", "rationale": "The method targets multi-institution collaborative training under privacy constraints, a scenario outside the reader's single-GPU production scope, though the LoRA-based schema harmonization idea is worth monitoring if cross-organization data collaboration ever becomes relevant.", "technique": "Schema-aware split learning with LoRA-fine-tuned LLM encoder", "family": "treino_eficiente", "recommendation": "observar", "published": "2026-09-14", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2609.15871", "independent_implementations": 0, "total_implementations": 0, "stars": 0, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "Replaces federated learning over raw survey data by serializing records into natural language and splitting a LoRA-fine-tuned LLaMA-3.2-3B between clients and a server, so raw records never leave each institution. Costs include server-side hosting of the LLM backbone, per-record LLM serialization and forward passes, and the orchestration complexity of split-training across institutions. It can fail through privacy leakage via intermediate activations, semantic loss when surveys serialize ambiguously, and dependence on a central server that reintroduces a trust and infrastructure assumption FL avoids.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.15871", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Cuts per-client computation by three orders of magnitude versus federated learning; attains average ANLS of 0.708 with 2,000 training samples, surpassing FL in eight of nine settings using LLaMA-3.2-3B-Instruct.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.15871", "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_formula", "core_kind": "system", "equations_fetched_at": "2026-10-09", "equations_source_url": "", "equations_source_sha256": "", "exposures_reviewed_at": "", "exposures_source_sha256": "" }