Home / Curated / Architectures / GENET

GENET

Architectures AIMech:GENET DRAFT

A connectionist architecture for finite constraint satisfaction that represents variable assignments by active label nodes and alternates input-maximizing assignment updates with increased inhibitory weights on violated constraints when the network stabilizes without a solution.

Identity Notes & Concept Scope This is the GENET constraint-satisfaction architecture described by Davenport, Tsang, Wang and Zhu, including the general-constraint extension. It is not a biological genetic network, genetic algorithm, trained classifier or synonym of GLS. The binary network has inhibitory nogood connections, while the general-constraint extension includes constraint nodes and asymmetric fixed/adaptive connections. A Hopfield identity or globally symmetric network mapping would overstate that general architecture. No normalized identity match was found across all 442 pinned AIO labels/scoped synonyms and 307 preceding native IDs/names/aliases. Full-checkout fixed-string searches included ignored and hidden files in worktree and main, returning 4 and 4 lines respectively. All eight text hits were prior discovery leads in the two checkouts, not existing candidate records. Eight exact AIO CLASS nodes with scoped synonyms/comments and outgoing edges and five native neighboring mechanisms were inspected. Absence is bounded to this snapshot and semantic neighborhood.

Claims & Grounded Evidence

1 Claim

Mechanism #method

Create one cluster for each finite-domain variable and one label node per possible value, with exactly one active label per cluster. In the binary formulation, incompatible label pairs have inhibitory connections initially weighted −1; a label input sums the weights from currently active connected labels. A cluster activates a label of maximal input, breaking ties randomly. These updates need not be strict improvements. The described parallel implementation is explicitly not exactly synchronous, and the reported simulator selects variables sequentially. When an update cycle leaves the assignment unchanged, terminate if all active-label inputs are zero; otherwise reinforce violated connections, using W_new=W−V_i*V_j in the binary case. The extension represents higher-order constraints by constraint nodes with fixed label-to-constraint inputs and adaptive inhibitory outputs to labels. For a violated general constraint, all its outgoing connection weights are decremented according to Equation5; this must not be silently replaced by the binary active-endpoint rule. The paper defines distinct output functions for forbidden tuples and atmost constraints.

supports Source: GENET: A Connectionist Architecture for Solving Constraint Satisfaction Problems by Iterative Improvement · Locator: PDF pp.1–4 (printed325–328), architecture, convergence/learning algorithms, footnote2, Equations1–9, and Generalised Constraint Representation; limitations and implementation discussion p.6 (printed330)

⚠️ Curated Limitations & Negative Results

The source explicitly says GENET is not complete: learning may create other local minima. It supplies no guarantee of finding a satisfying assignment or proving unsatisfiability within a finite budget.
Exactly synchronous updates can oscillate; the source footnote excludes that interpretation of the parallel pseudocode. Ties can change assignments without strict objective improvement, distinguishing this mechanism from the strict hill-climbing record.
The 1994 experiments use a sequential C++ simulation. Proposed speedups from future VLSI implementations are expectations, not measured hardware results curated here. The GENET2 conflicted-variable variant is not silently treated as the default variable-selection rule.
General-constraint output functions and representations require care. At an atmost bound, activation in the same variable cluster must account for all participating values; independently inhibiting each inactive value can introduce a false penalty. No numerical benchmark result is curated.

Artificial Intelligence Ontology (AIO) Grounding

Mapped AIO Classes

AIO Target Class	Mapping Relation	Mapping Rationale
`AIO:ArtificialNeuralNetwork` Artificial Neural Network	broad_match	The pinned target is a network of connected artificial-neuron units. GENET is a specific connectionist network of label/constraint nodes with activation and inhibitory weight-update rules, so the AIO target is broader than this local architecture. No content-addressable Hopfield identity is asserted, and general constraint connections are asymmetric; the broader ANN mapping does not imply universal symmetric connectivity.

Evidence Sources & Primary Literature

1 Source

Source ID	Publication / Document	Type / Access	Reference & Checksum
`method-source`	GENET: A Connectionist Architecture for Solving Constraint Satisfaction Problems by Iterative Improvement Andrew Davenport, Edward Tsang, Chang J. Wang, Kangmin Zhu	primary paper full_text	sha256:5dfc2658b9e8…

Curation Provenance

Curator: Codex

Method: ai_assisted

Model: GPT-6

Created: 2026-10-11T14:42:39Z

Updated: 2026-10-11T14:42:39Z

Source File: data/curated/GENET.yaml