Patience-controlled FCFS beam decoding
Algorithms AIMech:PatienceControlledBeamDecoding DRAFT
Aliases: FCFS beam decoding with controlled patience
A generalization of first-come, first-served sequence beam decoding that retains beam width k for continuing hypotheses but uses a positive patience factor p to stop when the accumulated completed-hypothesis count reaches at least k·p.
Identity Notes & Concept Scope
This is the algorithmic stopping-policy variant introduced by Kasai and colleagues, not a standalone model, learned parameter or generic training patience rule. Setting p=1 recovers their FCFS baseline; changing p separates its finished-output threshold from its continuing-beam width. No normalized identity match was found across the 442 AIO and 141 prior native records; ignored-inclusive text hits in the main checkout were prior discovery leads and incidental substrings such as builds in records, source text and dependencies, not existing concept records.
Claims & Grounded Evidence
2 Claims
Mechanism
#search-procedure
Keep the FCFS candidate selection and separate persistent finished set, but replace the termination threshold k with k·p. Return the highest-scoring stored completion once that threshold is reached, subject to the method’s maximum-length limit. With integer completion counts, the displayed inequality corresponds mathematically to reaching ceil(k·p); this is an interpretation of the paper’s inequality, not a claim about rounding in any software library.
Limitation
#patience-scope
The paper separates stopping depth from beam breadth and reports task-dependent behavior; its limitations explicitly restrict summarization evidence to English news. More extensive search and better generated-text quality are not equivalent claims.
⚠️ Curated Limitations & Negative Results
- The finite maximum length can end decoding before the completion threshold is reached. Empty finished sets and exhausted candidate pools need an implementation-specific outcome.
- Increasing patience can expose more completions but does not guarantee better task quality, a globally optimal sequence, or equivalence to vanilla beam search.
- The parameter is a stopping threshold, not a score upper bound or a proof that no unfinished hypothesis can improve the incumbent.
- The source evaluates specific translation and news-summarization settings. English news summarization does not establish benefits for all languages, domains or tasks.
- No current library default, runtime gain, numerical quality improvement or trained-model execution is claimed.
Artificial Intelligence Ontology (AIO) Grounding
Mapped AIO Classes
| AIO Target Class | Mapping Relation | Mapping Rationale |
|---|---|---|
AIO:NaturalLanguageProcessing
Natural Language Processing
|
related_match | The primary paper defines these sequence-decoding policies for natural-language generation, including translation and summarization. The exact AIO CLASS describes the language-processing task. This is an algorithm-to-application-domain relationship, not equivalence or an is-a assertion; the decoder is not a trained language model and can also be used outside NLP. |
Evidence Sources & Primary Literature
2 Sources| Source ID | Publication / Document | Type / Access | Reference & Checksum |
|---|---|---|---|
method-source
|
Jungo Kasai, Keisuke Sakaguchi, Ronan Le Bras, Dragomir Radev, Yejin Choi, Noah A. Smith
|
primary paper
full_text
|
sha256:7d090ee35511…
|
patience-scope
|
Jungo Kasai, Keisuke Sakaguchi, Ronan Le Bras, Dragomir Radev, Yejin Choi, Noah A. Smith
|
primary paper
full_text
|
sha256:7d090ee35511…
|
Curation Provenance
Curator: Codex
Method: ai_assisted
Model: GPT-6
Created: 2026-10-11T06:26:52Z
Updated: 2026-10-11T06:26:52Z
Source File:
data/curated/PatienceControlledBeamDecoding.yaml