| paper | d4-conf-ijcai-2016-0284 |
| 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 has no numbered Theorem, Lemma, Corollary, or Proposition asserting an algorithmic or complexity result; its active-learning methods are proposals, and its experimental findings state no computational guarantee. No result therefore meets the required computational-anchor gate, regardless of whether crowdsourced evaluators could sensibly be modeled as types.
fails bit a — no named computational result to mirror
None: the progressive comparison procedure, two active-learning criteria, and experimental findings have no qualifying numbered computational result to mirror.
I can’t make an honest anchor-based case for a continuous mirror here. The paper contains no numbered Theorem, Lemma, Corollary, or Proposition asserting a complexity result. Its named contributions are methods—progressive comparison and two active-learning criteria—and empirical findings in §5, not qualifying anchors.
A crowd of similar evaluators could be modeled as a continuous population, but no named result in this paper studies or depends on that model. Its Gaussian distributions describe uncertainty about object scores; they do not make the evaluator population continuous. Turning the experiments into an optimization problem over fractional evaluator mass would introduce a new problem rather than mirror a result of the paper. So there is no qualifying anchor for which to state a prover-ready continuous problem or predict a place in the trichotomy.
The proponent raised no anchors, and the paper offers no result that supplies one. Its contributions are a data-collection procedure, two active-learning criteria, and empirical findings. The findings in §5 depend on the tested datasets; none states a computational complexity result for a problem over a society distribution. There is therefore no named result here that yields a prover-ready continuous question.
The strongest possible extension would model crowdsourced evaluators as types, with masses \(\mu_t\) and type-specific preferences, costs, expertise, and evaluation history. One could then study how to allocate queries across those types to improve a ranking under a budget. That is coherent, and the paper’s crowdsourcing setting makes it plausible. But the paper chooses the next object for an evaluator based on that evaluator’s previous object and the model’s score uncertainty. Its Gaussian distributions describe uncertainty about object scores, not a population distribution over evaluators. Adding a mass-allocation problem would introduce a new objective and computational task, rather than continuize one of the paper’s results. Preserving evaluator history in each type would make the extension more faithful, but still would not supply a population-level result to mirror.
The negative case is strongest at the anchor gate, not as proof that no conceivable scenario could ever be worthwhile. A population-level active-learning problem for crowdsourcing could be valuable research; it would need its own formulation and results. On the paper’s evidence, though, it is a new problem rather than a continuous mirror.
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.