| paper | d4-conf-ijcai-2009-0343 |
| 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 proposes an inference method and reports empirical comparisons, but it contains no numbered computational result under the stated gate. With bit (a) failing, neither the locations-as-mass analogy nor a newly aggregated resident-preference task can make a result of this paper mirrorable.
fails bit a — no named computational result to mirror
No result is covered: the proposed EM-EP inference method and empirical performance comparisons have no eligible named computational result to anchor.
The paper has no numbered Theorem, Lemma, Corollary, or Proposition asserting a computational result such as tractability or hardness. It proposes the SSPL Gaussian process method and reports empirical results, but gives no named complexity result to anchor. So there is no eligible anchor—and no continuous problem I can honestly present as *per-anchor*.
The strongest possible analogy is to treat housing locations as the population: a type is a location’s feature profile \(x\in\mathbb{R}^{32}\), mass is the fraction of candidate locations with that profile, and the task is to infer latent preferences from extreme examples and pairwise rankings. One could define a mass-weighted graph prior and weighted likelihood, then ask for the posterior predictions and hyperparameters. But that is a proposed extension, not a result of this paper; moreover, it continuizes the set of alternatives being ranked, rather than a society of agents. The paper’s locations are graph nodes, not repeated voters.
So the honest positive case is limited: the paper’s housing example offers a possible motivation for an aggregate preference-learning model, but it supplies no named computational result to mirror, and the natural version misses ChoCo’s population-continuization target.
On the required result-by-result standard, there is no named computational result to mirror. The paper proposes EP inference and EM-EP hyperparameter learning, then reports empirical comparisons; it proves no complexity or approximation guarantee. So the proponent is right that it has no eligible named anchor.
The housing analogy does not supply one. Treating locations as types makes the distribution describe candidate sites, while the paper learns a score for a particular site from one evaluator’s judgments. A continuous distribution over sites could support average suitability or predictions over a site domain, but that changes the object being modeled from a population of agents to the set of alternatives.
The strongest repair is to make residents or users the population and ask for an aggregate preference over fixed locations. That is plausible, but it changes the learning problem. The paper assumes one latent preference function shared by its observations. If users share that function, repeated evidence changes the amount of evidence for it; the population does not add the heterogeneous preferences that motivate social choice. If users differ, a model must represent their separate preference functions and define how to calibrate and aggregate them. The paper’s single-function GP likelihood does not do that.
That repaired problem could be worthwhile in its own right, but it is a new collective preference-learning problem, not a continuous mirror of a result in this paper. The negative case is strongest on that scope and anchoring gap; it cannot establish that no plausible population model could ever 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.