The Observation Theory EncyclopediaFrom TSKAboutBy kindBy chapterBy Lean fileLedgerProvenance

fine-tuning instrument

DefinitionContinuing to train an encoder on a new objective or corpus, so that its quotient changes. Chapter 11 section 11.8 and chapter 12 section 12.4. Also fine-tun.
ExampleAn encoder tuned on translation pairs put seven of thirteen backbone areas in one class that the general encoder kept apart.
BookData Mining as Observation, draft 0.2, commit f3914f0; entry id fine-tuning, kind instrument.
Statusno ledger row names this entry. Corrections: none recorded.
Defining equationnone
Assumptions and scope
  • Continuing to train an encoder on a new objective or corpus, so that its quotient changes. A contrastive objective pulls declared pairs together, and the cross-corpus gate asks whether the tuned encoder still separates on a corpus it never saw.
  • An encoder fine-tuned on one relation is evaluated on that relation and on one it never saw, with a paired interval on each, and the within-corpus score of 0.75 to 0.955 collapsed to 0.47 to 0.55 across corpora until cross-corpus positives were added.
Prior artnone recorded
Evidencelean/DataMiningAsObservation/Contrastive.lean, lean/DataMiningAsObservation/CrossCorpusGate.lean
Reviewednot yet reviewed; generated 2026-09-10 from records at the commits on the provenance page.
three groups
Continuing to train an encoder so that its quotient changes.

Equation

none

Conditions

Conditions are curated in entries.toml rather than read from a record.

Ledger

none

First stated

Chapter 11 section 11.8 and chapter 12 section 12.4 of Data Mining as Observation, with the translation-trained encoder in turboquant-pro/docs/RESULTS_multilingual_strata.md:55-90.

Measurements

none

Failures and corrections

none

Invariance envelope

none declared

Machine checked

lean/DataMiningAsObservation/Contrastive.lean, theorems loss_nonneg, loss_eq_zero_iff, not_both, loss_mono_margin, at observation-data-mining f3914f0; what the check covers is stated in the book’s appendix C.

lean/DataMiningAsObservation/CrossCorpusGate.lean, theorems clears_bow, clears_untrained, not_validated_of_saturated, margin_example, validated_comp, at observation-data-mining f3914f0; what the check covers is stated in the book’s appendix C.

Used in

Data Mining as Observation chapters 11, 12.

Related

contrastive objective; encoder; cross-corpus gate; leakage; paraphrase class.

See also

Book equations stated beside the entry’s terms, not defining it: 12.3, 11.1.

Ledger rows that cite the entry’s records without naming it: GO-B-Llama, GO-B-Llama-rematch, GO-B-legal (035→036).

Sources-table rows that share a record with the entry without naming it: chapter 10 section 10.4, chapter 11 section 11.8, chapter 12 section 12.4, chapter 12 section 12.5.

Status

Generated 2026-09-10 by encyclopedia/generate.py; book at observation-data-mining f3914f0; the commit of every record is listed in the encyclopedia’s provenance.

← family-wise error ratefinite difference →