| paper | d4-conf-aaai-2018-1200 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | no |
The paper contains no numbered theorem, lemma, corollary, or proposition asserting an algorithmic or complexity result, so there is no qualifying result to anchor a mirror. Its equations describe methods and its runtime evidence is empirical; neither supplies the required computational result.
fails bit a — no named computational result to mirror
There is no named computational result to mirror; the consensus formulation, sampling strategies, and empirical findings therefore remain uncovered.
This paper has no numbered Theorem, Lemma, Corollary, or Proposition establishing a computational complexity or algorithmic result, so it offers no qualifying anchor for a continuous mirror under your criteria. It does present optimization formulations in equations (1)–(11) and empirical results, but neither is a named result of the required kind. I can’t honestly state a continuous problem per anchor without inventing one.
The proponent supplied no actual anchor to defeat. The paper has no named theorem, lemma, or corollary establishing a computational complexity result. Equations (1)–(4) define an iterative consensus method; equations (5)–(8) give sampling criteria; the rest is empirical evaluation. Runtime plots are not complexity guarantees. On the programme’s stated standard, there is no result here whose complexity landscape can be mirrored.
The strongest proposed mirror would treat labelers as voters and replace the finite committee with a population of annotators grouped by expertise. That is a plausible crowdsourcing question, but preserving this paper’s model requires a type to include each annotator’s predictions across instances: those predictions form the edges in its consensus graph. Grouping only by label reliability loses that information. Grouping by complete deterministic response profiles may leave nearly every annotator in a different type. A model with stochastic response rates for expertise types could support a continuum, but it changes the paper’s reliability updates and individual judgments into an anonymous crowd model. The paper’s authors might recognize that as a useful extension; it still would not mirror a computational result they proved.
Continuizing the data instances is weaker. The classifier and influence-based sampler use each instance’s feature vector, so a type must retain that vector. Grouping instances only by labeler predictions loses information needed for training and query selection; retaining the features can leave the support essentially as large as the original dataset. Also, the paper’s sampling step selects individual instances. In a nonatomic population, a single instance has zero mass and its effect on expected agreement vanishes. Selecting a positive mass of instances could define a sensible batch problem, but it changes the operation the paper studies.
So the negative case is strongest on the absence of any eligible computational result to anchor a mirror, with the population-axis mismatch reinforcing it. The universal claim remains less secure: a large crowd with repeated expertise types is a credible new setting, and the authors could plausibly care about it. Its weakness is that it extends the application rather than continuizing a named computational result of this paper.
The adversarial triple: the proponent anchors on up to three named results; the opponent sees that case and must defeat every anchor; the judge decides which case convinced it. These are the pipeline’s own outputs, generated by tools/triple_run.py — no human edited them. The paper’s own text is not reproduced here beyond the quoted statement above.