IECO constrained elliptical-mixture classification
Algorithms AIMech:IECOConstrainedMixtureClassification DRAFT
A constrained classification procedure that models each known class by a mixture of elliptical category distributions, marginalizes category likelihoods, and assigns independent observations to distinct class slots by minimizing negative log mixture costs together with specified missing-slot prior costs.
Identity Notes & Concept Scope
Current identity assessment at base 7d38dafe42243c93664bc246a5ff77beca593cd3 compared the IDs, names, aliases and scoped synonyms of all 1234 records (4896 field occurrences; 3739 unique normalized values), 19 AIO contexts and 17 native neighbors against the accepted unchanged historical records, plus the current ignored/hidden-inclusive four-root search and full frontier labels with matched/plausible owner bodies. This record is the supplied-density IECO constrained virtual-slot assignment procedure, with categories marginalized before MAP of Phi. It is distinct from POP/BAS/OUT parameter fitting, independent per-object Gaussian QDA/LDA, Naive Bayes feature independence, SeLa learned pseudo-label transport, and iterative conditional estimation. The 1995 singleton-category fixed-count setting is contextual lineage, not an exact alias; the 1996 maximum-over-quality rule differs from the 1997 category sum. No universal MAP claim for aggregated biological labels is made. No exact pinned AIO identity was established in the inspected scope; the selected AIO:Classification related_match connects an algorithm to a task, not an equivalence or subclass claim. Proposed main:913 is shared source responsibility for two separate records, retaining all 22 bibliography occurrences and explicit residual questions. AutoClass retains main:610. Frozen issue402 and issues404/406 plus the prior twelve search questions remain unchanged; no source exhaustion is claimed.
Claims & Grounded Evidence
5 Claims
Mechanism
#category-marginalized-class-density
For each class slot j, provide category densities f_jk with nonnegative category priors p_jk summing to one. Under the stated object/category independence assumptions, the class likelihood for observation x_i is sum_k p_jk*f_jk(x_i). IECO specializes these laws to elliptically symmetric regular and outlier distributions; its radial-family choices are model inputs. The observation-to-class cost is the negative natural logarithm of that mixture sum.
Mechanism
#injective-assignment-and-missing-slots
For m observations and n class slots with m<=n, require an injective assignment of observations to slots. With a missing-subset prior proportional to the product of alpha_j over missing slots and an independent uniform permutation prior, append n−m dummy rows of costs −log(alpha_j). Minimize the total n-by-n assignment cost. Repeated slots can encode known class multiplicities; the generic linear-assignment solver is a component, not the classifier identity.
Scope
#missing-weight-interpretation
General alpha_j values are nonnegative subset weights with a normalizing constant at fixed m. In the independent-coin construction they are failure odds g_j/(1−g_j), not generally the marginal probabilities that classes are absent. Only the normalized m=n−1 specialization identifies alpha_j with the probability that slot j is missing.
Scope
#class-versus-category-decision
The decision maximizes the posterior of Φ, the injective assignment to indexed virtual class slots, after marginalizing categories. It does not universally maximize the posterior of a biological labeling obtained by merging equivalent slots or category-dependent identities; that target requires aggregating posterior mass over its compatible realizations. Choosing a maximum-posterior category after the slot assignment also does not in general give the joint slot-and-category MAP decision. The 1996 abstract’s minimum-over-quality scoring therefore cannot be substituted for the 1997 mixture-sum costs.
Scope
#iec-lineage-special-case
With one category per class and fixed class multiplicities, the1997 construction includes the earlier independent elliptical classifier setting. The1995 cancellation of class-dependent determinant factors relies on fixed multiplicities; it does not justify dropping those factors when comparing different missing-class or count hypotheses. The generic1997 mixed-class construction supports broader densities, while this proposed IECO record retains the authors’ elliptical specialization.
⚠️ Curated Limitations & Negative Results
- The MAP interpretation concerns Φ, the injective indexed-slot variable of Section 3, conditional on valid normalized component laws, category priors, object independence, the specified missing-subset product prior, an independent uniform permutation prior and feasible slot constraints. Equivalence to MAP over coarser biological labels requires a separate aggregation argument; exchangeable fixed-multiplicity cases have constant representation counts. This general boundary does not establish failure of the paper’s specific 45-chromosome construction. It does not guarantee correct biological labels or calibrated future error.
- Positive mixture likelihoods and positive weights give finite costs. Zero likelihoods or zero missing weights require infinite-cost constraints; an all-infeasible assignment has no finite optimum and is not silently made valid.
- The source’s generic section3 permits multiple category types, while IECO denotes the elliptical regular/outlier specialization. A chromosome-specific construction, a general assignment solver and every finite-mixture classifier are not exact aliases.
- The author-hosted1997 PDF p.11 Eq13 displays a density symbol equated to an expression with the form of a negative log cost, and its radial argument differs from the explicit norm form in the1996 abstract. The proposal uses the valid supplied-density construction of section3, does not certify Eq13 literally and does not silently repair it. No defect in p.9 Eq9 was established.
- The1995 Pareto radial family has a support cutoff; an unrestricted log-distance score omitting that cutoff is not asserted equivalent to its literal likelihood. This predecessor boundary is retained rather than imported into IECO as an unqualified formula.
- The1996 two-page conference abstract is context only and its maximum-over-quality decision differs from the1997 sum-over-category rule. The earlier cutREG notation is not used to patch the1997 fitting text.
- Complete numerical reproduction needs component fitting, cutoff calibration, category priors, covariance treatment, missing-slot conventions and a solver implementation. No author code, historical benchmark, universal complexity or numerical performance advantage is claimed.
- The three source artifacts are exact author-hosted versions. Journal publication metadata identifies the works but does not establish equivalence to publisher PDF bytes. Source-version comparison remains a retained question.
Artificial Intelligence Ontology (AIO) Grounding
Mapped AIO Classes
| AIO Target Class | Mapping Relation | Mapping Rationale |
|---|---|---|
AIO:Classification
Classification
|
related_match | The procedure assigns observations to externally specified classes using fitted class-conditional distributions and global class-slot constraints. The inspected target is a classification task under SupervisedLearning, so the association is related rather than an exact or subclass mapping. Unobserved regular/outlier categories do not make the external class labels latent clusters, and two qualities do not imply binary class assignment. |
Evidence Sources & Primary Literature
3 Sources| Source ID | Publication / Document | Type / Access | Reference & Checksum |
|---|---|---|---|
ritter1997
|
Gunter Ritter, María Teresa Gallegos
|
primary paper
full_text
|
ID: doi:10.1016/S0167-8655(97)00049-4
sha256:3e52f264d980…
|
ritter1995-context
|
Gunter Ritter, María Teresa Gallegos, Karl Gaggermeier
|
primary paper
full_text
|
sha256:aefbb4588e63…
|
gallegos1996-context
|
María Teresa Gallegos, Gunter Ritter
|
primary paper
abstract_only
|
sha256:2f7496ce08ab…
|
Curation Provenance
Curator: Codex
Method: ai_assisted
Model: GPT-6
Created: 2026-10-16T02:03:17Z
Updated: 2026-10-16T02:03:17Z