| paper | d4-conf-ijcai-2017-0128 |
| 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's latency and task-count claims are unnumbered analysis, not a numbered theorem, lemma, corollary, or proposition asserting a computational result. The proposed worker-population model adds type-specific capacities and policy optimization that the paper does not formulate, so no qualifying continuous mirror is anchored in a result of this paper.
fails bit a — no named computational result to mirror
No numbered computational result is covered; the proposed worker-type policy is a new optimization problem, while the paper's latency and task-count analysis, budget strategy, and collision-risk method remain unanchored.
Strictly, I have no qualifying anchor: the supplied paper contains no numbered theorem, lemma, corollary, or proposition asserting a complexity result. It does make computational claims. In §3, it says that for fixed \(k\) and task size \(s\), Crowd-Top-\(k\) has logarithmic latency and uses a number of ranking tasks linear in the item set size, but it says the proof is omitted. The randomized variant and budgeting strategy are also described, with experiments reported later. Those are meaningful contributions, but they do not meet your specified anchor rule.
The strongest positive case I can make is therefore conditional: the most plausible mirror is about the crowd as a population of workers. A realistic regime is a large platform with many workers in a small number \(\tau\) of behavior classes, such as reliable workers, workers who make occasional mistakes, random spammers, and adversarial spammers. Let type \(w\) specify a worker’s response distribution, task capacity, and cost; let \(\mu_w\) be that type’s fraction of the workforce. With \(N\) workers, each type has about \(N\mu_w\) members. That is a sensible high-multiplicity regime, and the paper’s own error and spam models give it some support.
A concrete candidate is Continuous Crowd Top-\(k\) Recovery. An instance has a finite item set \(I\), a target \(k\), a maximum ranking-task size \(s\), a finite set of worker types \(W\), a rational workforce distribution \(\mu\), type capacities \(h_w\), per-task costs \(c_w(S)\), and known response kernels \(P_w(\rho\mid S,\beta)\). Here \(\beta\) is the unknown baseline ranking, \(S\subseteq I\) is a task, and \(\rho\) is the ranking returned by a type-\(w\) worker. A solution is an adaptive policy that issues tasks to worker types, observes their responses, and returns a \(k\)-item set \(\hat T\). The problem is to minimize worst-case expected cost over \(\beta\), subject to
\[
\Pr_{\Pi,\beta}\!\left[\hat T=\operatorname{Top}_k(\beta)\right]\ge 1-\delta
\quad\text{for every }\beta,
\]
and type-level capacity constraints determined by \(H\mu_w h_w\), where \(H\) is the campaign’s workforce scale. Task assignments are counted as normalized mass, representing many task assignments spread across a large worker population.
The best connection is to the paper’s treatment of worker inaccuracy, its adversarial-spammer budget allocation in §3.2, and its risk-controlled randomized variant in §3.3. The authors could plausibly recognize this as their underlying top-\(k\)-from-crowd problem with a high-multiplicity workforce made explicit. I would initially expect the complexity to be Class B: the hidden ordering and the choice of comparisons still involve combinatorial structure in the items, while smoothing the workforce counts may not remove that structure. That is a hypothesis, not a hardness claim. Natural follow-ups include whether adaptive policies beat fixed batch plans, whether the top-\(k\) guarantee can be relaxed to bounded quality loss, and when a fractional workforce policy can be rounded with controlled loss.
My weakest point is the gap between that candidate and the paper’s formal contribution. The paper treats the crowd as a source of noisy answers; it does not model worker-type masses or optimize over a worker population. Its stated complexity claim is about the number of items and ranking tasks, not the number of workers or worker types. So this is a plausible extension of the authors’ problem, but—without a qualifying numbered anchor—I cannot honestly present it as a mirror of a named result from this paper.
The proponent’s concession is decisive under the requested anchor rule: the paper gives no named theorem or other numbered result asserting the complexity of a computational problem. Its claim that Crowd-Top-\(k\) has logarithmic latency and uses linearly many ranking tasks is relevant algorithm analysis, but the proof is omitted, and the bounds concern crowdsourcing resources rather than computational running time. The budgeting strategy is described as an optimization problem and evaluated experimentally; the randomized variant bounds collision risk. Neither is a named complexity result. There is therefore no qualifying anchor here.
The proposed worker-population model is the strongest possible rescue, and it is not defeated just by saying that workers have individual identities. A large platform can plausibly have many workers in a few behavior classes, and the paper’s spam and error models give that scenario some grounding. But the paper treats the crowd as a source of noisy observations about an unknown baseline ranking \(\beta\). Its output is an estimate of the top \(k\) items under \(\beta\), not a collective choice made from worker preferences. A distribution of worker types would describe the observation process; it would not turn the item set or the ranking problem into a continuous society.
The better version of the proposed mirror would give worker types distinct response distributions, prices, and task capacities, then optimize a task policy for a required probability of recovering the true top \(k\). That is a coherent possible research question. But its central ingredients—known response kernels, type-specific costs and capacities, a workforce scale, and optimization over adaptive policies—are added to the paper. The paper does not model a finite worker pool whose capacity is consumed, or compare alternative policies against type-level workforce constraints. Calling this a mirror of its treatment of worker inaccuracy, budgeting, or randomized risk would stretch “mirror” to include a new resource-allocation problem that borrows the same application.
The connection is weakest for the two named sections the proponent invokes. In §3.2, the budget is allocated across recursive calls, not across worker types. In §3.3, collision risk comes from randomly partitioning items; changing the distribution of worker types does not change that event. Worker errors affect the quality of the observed rankings, but the paper handles them through aggregation and redundancy. To make worker composition the object of computation, one must add a new constraint or objective; the paper’s results do not supply it.
The proponent’s Class B expectation is not a reason against a mirror: what the continuous answer turns out to be cannot settle whether the question is worth asking. The stronger point is that this paper offers no named complexity result to anchor such a question to, and the workforce formulation changes what is being optimized. A continuous worker-pool problem might be worthwhile in its own right, especially where type-specific capacities bind. The paper alone does not establish that case. The negative argument is strongest at the anchor gate; it cannot prove that no future, separately motivated workforce model could be useful.
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.