Supplemented Expectation Maximization
Algorithms AIMech:SupplementedExpectationMaximization DRAFT
Aliases: Supplemented EM, SEM Algorithm (Supplemented EM)
A numerical supplement to expectation maximization that combines expected complete-data information with the local Jacobian of the EM update map to estimate the observed-data asymptotic parameter covariance.
Identity Notes & Concept Scope
Supplemented EM estimates uncertainty around a converged EM fit; it is distinct from stochastic EM, structural equation modeling, EM acceleration and the fitting algorithm itself. Bare SEM is not asserted as an alias. The full 1995 SECM primary follow-up gives the SEM equations and explicit implementation as a special case, supporting this record despite the inaccessible original 1991 full PDF. All 442 pinned AIO labels/scoped synonyms and 466 preceding native IDs, names and aliases were compared. The sole normalized search-term collision, bare SAGE, identifies the existing sparse additive text model; the new optimization method uses qualified aliases. Ignored-inclusive fixed-string searches of both full checkouts returned 436 and 459 line instances. Context windows for all 143 distinct normalized matched texts were inspected; duplicate ledger/review mentions, unrelated environment strings and the stochastic-EM lead do not establish these four method identities. All 225 semantic label/synonym rows and 17 annotated AIO nodes with outgoing edges were inspected; sixteen nodes were byte-verified unchanged after prior reading and UncertaintyBias was newly read. Thirty-one contextual YAMLs comprise 17 AIO and 14 native records, with 27 verified prior readings and four full new readings; the existing sparse text SAGE was also reread. Absence is bounded to this pin, maintained corpus and inspected neighborhood.
Claims & Grounded Evidence
3 Claims
Mechanism
#information-from-em-map
At a regular interior fit theta*, let Ic be the conditional expected negative complete-data Hessian and D the EM map derivative using input-coordinate rows and output-coordinate columns. The missing-information identity gives Io=(I-D)Ic, so Vobs=Ic^(-1)(I-D)^(-1). Equivalently add Ic^(-1)D(I-D)^(-1) to the complete-data covariance. The factors and derivative convention matter; a single scalar convergence rate cannot replace the full Jacobian in a multiparameter problem.
Mechanism
#coordinate-map-probes
Obtain a tightly converged fit and expected complete information. Perturb one input coordinate away from theta*, holding the other input coordinates at theta*, and perform one full ordinary EM update. Divide each output change from theta* by the input perturbation to approximate one row of D, using stabilized ratios as perturbations approach the fit. Section 3 implements SEM by making CMSTEPS a joint Q maximization and setting the separate CM derivative to zero.
Limitation
#asymptotic-and-numerical-boundaries
Inverse observed curvature is an asymptotic uncertainty approximation requiring a regular interior fit and nonsingular information. Ratio calculations become unreliable with inaccurate fitting, tiny denominators or missing-information fractions near one. Components reaching their fitted values exactly can require special treatment instead of zero-denominator ratios. Inspecting asymmetry is a useful diagnostic, but symmetric output does not certify the correct covariance or statistical coverage.
⚠️ Curated Limitations & Negative Results
- The inspected full-text mechanism source is the 1995 primary follow-up, not the original 1991 paper. Original-paper access remains a documented evidence gap; no unseen original proof is claimed.
- The method additionally requires expected complete-data information. The simple sufficient-statistic implementation assumes an appropriate exponential-family complete model; general expected-curvature calculation may require integration.
- Boundary optima, nonidentifiability, singular information and incomplete convergence can invalidate the inverse-curvature calculation or its statistical interpretation. It does not provide a full posterior distribution or guaranteed finite-sample interval coverage.
- The source’s strong informal language about symmetry assuring correctness is not adopted as a theorem. A wrong common information scaling can preserve symmetry while giving wrong variances. Finite Gaussian matrix checks are not a calibration study or replication of the source examples.
Artificial Intelligence Ontology (AIO) Grounding
Explicit AIO Snapshot Assessment: No Match
Rationale: This method computes asymptotic parameter covariance by supplementing an EM-type fit with expected complete information and update-map derivatives. The inspected pin supplies no fitting-uncertainty or observed-information method counterpart. A generic task, fitted model, preprocessing operation or uncertainty-related bias would conflate the inferential calculation with a different kind of concept.
Search Scope: All 442 pinned CLASS labels and scoped synonyms,466 preceding native identities and ignored-inclusive full-checkout searches;225 semantic label/synonym rows and 17 annotated AIO nodes with outgoing edges were inspected. Assessment is bounded to the recorded AIO snapshot.
| Examined AIO Candidate | Rejection Reason |
|---|---|
AIO:MachineLearningTask
|
The target is the learning process/task. Supplemented covariance computation is a general inferential calculation after fitting, not itself a learned-task identity; that broad association alone is not used to force a link. |
AIO:Model
|
The calculation does not define a distribution/model family or a trained model instance. |
AIO:TrainingStrategy
|
This pinned node is a preprocessing strategy, not an inverse-information or numerical covariance procedure. |
AIO:RegressionAnalysis
|
Regression is only one possible application; the procedure applies to other incomplete-data likelihood models. |
AIO:DataAugmentation
|
The pinned term enriches training examples by transformations. Conditional latent-data information in EM is a different construction. |
AIO:UncertaintyBias
|
This bias favors groups better represented in training because of lower prediction uncertainty. It does not denote parameter covariance estimation. |
AIO:EvaluationBias
|
This is a bias in evaluation; a numerical covariance calculation is not such a bias or a predictive benchmark. |
Evidence Sources & Primary Literature
1 Source| Source ID | Publication / Document | Type / Access | Reference & Checksum |
|---|---|---|---|
method-source
|
David A. van Dyk, Xiao-Li Meng, Donald B. Rubin
|
primary paper
full_text
|
ID: https://www.ma.imperial.ac.uk/~dvandyk/Research/95-sinica-secm.pdf
sha256:31ec9fba0523…
|
Curation Provenance
Curator: Codex
Method: ai_assisted
Model: GPT-6
Created: 2026-10-12T03:40:12Z
Updated: 2026-10-12T03:45:56Z