FocusedILS
Algorithms AIMech:FocusedILS DRAFT
A ParamILS algorithm-configuration method that adaptively extends paired evaluation prefixes until one configuration dominates the other, and allocates bonus evaluations to configurations that win the comparison.
Identity Notes & Concept Scope
FocusedILS is the adaptive comparison instantiation of ParamILS. Its domination relation compares common instance/seed prefixes and requires the dominating configuration to have at least as many cached runs. It is not a fixed-sample BasicILS comparison, a Friedman significance test, or ROAR’s uniform proposal method with a different intensification routine. No normalized identity match was found across all 442 pinned AIO labels/scoped synonyms and 265 preceding native IDs/names/aliases. Full-checkout fixed-string searches included ignored and hidden files in both worktree and main checkout, returning 20 lines in the worktree and 25 in main. These were prior candidate leads, source references, neighboring-method mentions and five unrelated ROAR namespace entries in ignored dependencies, not existing records for these identities. Seven exact AIO CLASS nodes with synonyms and outgoing edges and six neighboring native mechanisms were inspected. Absence is bounded to this snapshot and semantic neighborhood.
Claims & Grounded Evidence
1 Claim
Mechanism
#method
Within the ParamILS search framework, compare configurations using cached target-algorithm runs on a common ordered instance/seed sequence. A configuration dominates another when it has at least as many runs and its estimated cost over the other’s full prefix is no greater. Begin a comparison by extending the shorter prefix by one run, or both when their lengths are equal; continue extending until either configuration dominates. Procedure3 increments its global bonus counter once per comparison and again when the initial run counts tie. When the first configuration dominates, add the accumulated number of bonus evaluations to it and reset the counter; otherwise return the unsuccessful comparison without that reset. Keep an incumbent while the search continues.
⚠️ Curated Limitations & Negative Results
- The 2009 formulation has a finite configuration domain; numerical parameters require discretization. Objective, run cutoff and training-instance distribution affect the selected configuration. Independent held-out evaluation is needed to assess generalization; no universal solver speedup is asserted.
- The printed first-improvement loop accepts the supplied non-strict comparison. Equal-cost neighboring configurations can therefore cycle under a literal reading; this record does not assert unconditional termination of that loop or executable equivalence to a software release.
- Adaptive capping is a separable acceleration, not the defining identity of this search/comparison method. Section4 describes trajectory-preserving and more aggressive alternatives. Procedure4 prints a max operation inconsistent with shortening the run cutoff described in the prose, and uses inconsistent cutoff symbols; the printed routine is not treated as verified executable code.
- The paper’s asymptotic statements require a finite space, positive-probability random restarts, consistent cost estimates and indefinitely continuing iterations/evaluations. They are not a finite-budget optimality certificate. Repeated sampling from a finite training set does not by itself establish consistency for a different population of instances.
Artificial Intelligence Ontology (AIO) Grounding
Explicit AIO Snapshot Assessment: No Match
Rationale: No inspected AIO target captures this general algorithm-configuration procedure or a defining component. The local algorithm remains supported by its primary source; this is a bounded mapping assessment, not absence from all ontologies.
Search Scope: All 442 pinned AIO CLASS identities and scoped synonyms, all 265 preceding maintained identities, ignored/hidden checkout text, and semantic inspection of seven AIO targets plus neighboring configuration/resource-allocation methods.
| Examined AIO Candidate | Rejection Reason |
|---|---|
AIO:TrainingStrategy
|
The pinned definition concerns preprocessing/training of machine-learning models. This method configures target algorithms generally; an optional use for model tuning does not establish a training-strategy identity or necessary relation. |
AIO:MachineLearningTask
|
The target is a broad process for building methods that learn from data. Configuring an arbitrary solver over instances need not build such a learning method; shared use of observations does not justify equivalence or subclassing. |
AIO:MetaLearning
|
The pinned task learns from metadata about machine-learning experiments. These methods require no collection of ML experiments or learned cross-task prior; solver configuration alone is not that identity. |
AIO:ActiveLearning
|
The pinned task queries labels for data points. Choosing configuration evaluations or instance runs is not label acquisition in that defined sense. |
AIO:ReinforcementLearning
|
Exploration and empirical feedback do not by themselves establish the pinned learning-task identity; no reward-learning agent or policy-learning component is required by these configurators. |
AIO:RandomForest
|
These methods contain no defining decision-tree ensemble surrogate; random sampling, randomized local search and rank tests are not random forests. |
AIO:EvaluationBias
|
This is a bias concept, not a configuration algorithm. Potential overfitting or distribution mismatch is a limitation rather than an identity mapping. |
Evidence Sources & Primary Literature
1 Source| Source ID | Publication / Document | Type / Access | Reference & Checksum |
|---|---|---|---|
method-source
|
Frank Hutter, Holger H. Hoos, Kevin Leyton-Brown, Thomas Stützle
|
primary paper
full_text
|
ID: doi:10.1613/jair.2861
sha256:3a1223d28e13…
|
Curation Provenance
Curator: Codex
Method: ai_assisted
Model: GPT-6
Created: 2026-10-11T12:04:49Z
Updated: 2026-10-11T12:04:49Z
Source File:
data/curated/FocusedILS.yaml