Parameter-Expanded Expectation Maximization

Identity Notes & Concept Scope PX-EM changes the complete-data representation without enlarging the original observed-data model family. Auxiliary parameters are not independently identified from the observed data. It is deterministic likelihood fitting, distinct from parameter-expanded data-augmentation sampling, generic model overparameterization, ECM constrained updates and transformation-based training-data augmentation. No normalized identity match was found across all 442 pinned AIO labels and scoped synonyms and 463 preceding native IDs, names and aliases. Fixed-string searches of worktree and main included ignored and hidden files, yielding 52 and 53 matched lines respectively. All 41 distinct contextual passages across 105 line instances and nine checkout-qualified paths were read; these concern prior discovery references and factor-analysis evidence, plus an unrelated ignored URI-template parameter-expansion string. They are not existing identities for these three algorithms. All 223 semantic label/synonym rows were read. Fifteen exact annotated AIO CLASS nodes and outgoing edges were inspected: fourteen verified unchanged after prior reading and DataAugmentation newly read. Twenty-five contextual YAMLs comprise fifteen AIO and ten native records; twenty are unchanged verified prior readings, two are hash-verified final records fully read in the preceding cohort, and three were fully read here. General EM, tempered EM, Gaussian factor models, PPCA mixtures, Gibbs/Metropolis sampling, specialized deep Bayesian data augmentation and alternating Newton fitting do not identify these particular conditional/expanded EM procedures. Absence is bounded to the pinned snapshot, maintained corpus and inspected neighborhood.