Extended Guided Local Search
Algorithms AIMech:ExtendedGuidedLocalSearch DRAFT
Aliases: EGLS
The Tsang–Mills–Ford extension of Guided Local Search that combines cost-scaled penalty weighting, aspiration for a new best original-objective solution despite an augmented-objective penalty, and probabilistic random neighborhood moves.
Identity Notes & Concept Scope
EGLS here denotes the combined coefficient, aspiration and randomness extension in CSM371. GLS+A and GLS+R are separate ablations, not exact aliases of the combined algorithm. The method modifies the GLS search framework; it is not FLS activation-bit bookkeeping, simulated annealing, or a generic name for any later GLS extension. 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
Retain GLS feature counters, augmented objective h=g+lambda*sum(p_i*I_i), and penalty increments for present features tied for maximal cost/(1+p_i) utility. Scale lambda=a*C, where C estimates a characteristic cost contribution; the TSP illustration uses the first local-minimum tour cost divided by the number of cities. Add the report’s literal aspiration condition: accept an examined neighbor with g(s_next)<g(best_so_far) even when h(s_next)>h(s_current). This bypasses an augmented-cost rejection; it does not erase feature counters. Add randomization so that, with configured probability P_r per move, a randomly selected neighbor is accepted regardless of its objective or penalty quality. The combined EGLS uses all three changes. Its reported recommendations a=1 and P_r=.2 are experimental defaults, not universal constants or a proof of parameter independence.
⚠️ Curated Limitations & Negative Results
- The report title does not establish a parameter-free algorithm. Feature definitions, neighborhood and local solver, cost estimate C, coefficient a, randomization probability P_r, initialization and stopping remain choices; the proposed defaults were evaluated only in the reported problem settings.
- The literal aspiration inequality is strict in both original improvement and augmented deterioration. It does not specify equality handling when h(s_next)=h(s_current); the underlying local solver and concrete implementation must declare that boundary and the order of random versus ordinary/aspiration moves.
- Scaling by a typical cost contribution is problem-dependent. The TSP first-tour average is not a universal normalization for arbitrary objectives, especially zero, negative or mixed-scale terms.
- Random moves can worsen either objective and need not improve exploration on a particular step. Aspiration protects access to an examined new original best, not global optimality or monotonic improvement of the entire trajectory. No comparative performance claim is curated.
Artificial Intelligence Ontology (AIO) Grounding
Explicit AIO Snapshot Assessment: No Match
Rationale: No suitable target was found for this particular general optimization procedure in the inspected snapshot and semantic neighborhood. Shared learning, feedback or evolutionary vocabulary does not establish a model or task mapping.
Search Scope: All 442 pinned AIO labels/scoped synonyms and 307 preceding maintained IDs/names/aliases; ignored/hidden text in both full checkouts; eight exact AIO CLASS nodes with scoped synonyms, comments and outgoing edges; five complete native neighboring records.
| Examined AIO Candidate | Rejection Reason |
|---|---|
AIO:ArtificialNeuralNetwork
|
The general GLS/genetic optimization procedure is not a network of artificial neurons. The neural history of GLS does not make every later optimizer a neural architecture. |
AIO:Network
|
A population, feature collection or move graph is not by itself a defining network architecture; this very broad node does not identify the search procedure. |
AIO:HopfieldNetwork
|
The method is not the recurrent content-addressable memory architecture specified by this target. |
AIO:FeedbackNetwork
|
Adaptive search feedback is not this neural architecture with input, hidden and output layer relationships. |
AIO:SymmetricallyConnectedNetwork
|
No symmetric neural connectivity with equal bidirectional weights defines this optimizer. |
AIO:TrainingStrategy
|
The method solves general constrained optimization and is not restricted to preprocessing or training a machine-learning model. |
AIO:MachineLearningTask
|
The target concerns building methods that learn from data. Searching assignments under a supplied objective does not necessarily instantiate that learning task. |
AIO:EvolutionaryLLM
|
The target is a large language model. Its related evolutionary-algorithms synonym does not identify all genetic optimizers. |
Evidence Sources & Primary Literature
1 Source| Source ID | Publication / Document | Type / Access | Reference & Checksum |
|---|---|---|---|
method-source
|
Edward Tsang, Patrick Mills, John Ford
|
author technical report
full_text
|
ID: CSM371
sha256:4b74e69de002…
|
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/ExtendedGuidedLocalSearch.yaml