| paper | d4-conf-ijcai-2019-0645 |
| 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 a computational guarantee or hardness result. Its convergence derivation is unnumbered and gives no complexity bound, so no anchor satisfies bit (a); the hypothetical task population does not supply the missing anchor.
fails bit a — no named computational result to mirror
No named computational result is covered; the convergence derivation and empirical accuracy comparisons remain unanchored.
This paper has no qualifying named computational result to anchor a continuous mirror. It states no numbered theorem, lemma, corollary, or proposition establishing that a problem is in \(\mathrm{P}\), NP-hard, FPT, or otherwise classifying its complexity. Section 4 analyzes convergence of the MNLDP recurrence, and the paper reports experimental accuracy results, but neither supplies the required named complexity result.
A possible mirror, if the anchor requirement were relaxed, would treat a large batch of recurring tasks as a distribution over task types, with each type carrying its features and initial noisy-label distribution; the output would be a propagated label distribution for each type. But that is only a proposed reformulation. Since the paper’s propagation depends on the feature-space neighborhood graph, compressing tasks into a small number of types may fail to preserve the problem when task features are mostly distinct.
So there is no honest per-anchor continuous problem or expected complexity classification to give under this brief.
There is no named computational result here to mirror. The paper proposes MNLDP, proves convergence of its propagation recurrence, and reports accuracy experiments. It does not classify the complexity of label integration or any other computational problem. Under the requested result-by-result standard, there is no anchor; the convergence analysis is not a complexity result.
The strongest unanchored candidate is to ask for the complexity of propagating labels through a continuous population of tasks. But the paper’s object is a finite feature graph: each instance has its own feature vector and noisy-label distribution, and its output is a label for that particular instance. The propagation depends on local neighborhoods built from those feature vectors. Collapsing repeated task types into masses can change those neighborhoods and the resulting outputs. To retain the original structure, a type would need to preserve enough of each task’s position in the graph that the proposed high-multiplicity compression loses its point.
One could instead define a measure over feature space and replace the \(k\)-nearest-neighbor graph with a continuous kernel or neighborhood operator. That may be a sensible continuum limit for graph-based learning, but it changes the central object into a field of labels over feature space. The paper supplies no optimization or complexity claim for that object, and the resulting question belongs to continuous learning and graph limits rather than a computational social-choice result this paper establishes.
So the negative case is strongest on scope and anchoring, not on impossibility: one can imagine useful continuous models of crowdsourced label propagation. What is missing is a paper result that makes a specific continuous computational-social-choice mirror worth pursuing. The proposed task-population mirror does not supply that missing anchor.
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.