| paper | d4-conf-ijcai-2015-0463 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | yes |
Theorem 1 is a named computational result, but its input already assigns probability mass to complete preference types and its output is the mass satisfying a ranking constraint. Reading that mass as cohort shares changes no input, computation, or result, so the proposed mirror restates the paper's existing distributional model.
fails bit b-already — the paper is already native to the continuum
The proponent flags this collision but does not resolve it: the paper already assigns mass to complete ranking types, and Theorem 1 computes satisfying mass, so cohort language adds no operation.
fatal: True
The query mirror covers Theorem 1's exact constrained-probability inference; it leaves Theorem 2's EM learning and the paper's other query results aside.
There is one credible, narrow mirror: Continuous Preference-Mass Query. Take a large cohort whose members’ types are complete rankings of a fixed set \(A\) of alternatives. Its society is a mass function \(\mu\) over the valid rankings \(\Omega(A)\). Represent that mass function by a rational PSDD, and represent a property of rankings—such as “\(c\) is in the top three and \(a\) is ranked above \(b\)”—by an SDD on the same vtree. The problem is to compute the exact cohort share
\[
q=\sum_{\sigma\in\Omega(A):\,\sigma\models\alpha}\mu(\sigma).
\]
The solution is the rational number \(q\). There is no intervention variable here: the computational task is to calculate the mass satisfying a ranking constraint.
The high-multiplicity scenario is a large recommender cohort ranking a fixed, modest catalogue, or a large electorate ranking a fixed candidate set. Users with the same complete ranking are one type; for fixed \(m=|A|\), there are at most \(m!\) types, while the cohort can grow to millions. A discrete cohort with \(N_\sigma\) members of each type gives \(\mu_\sigma=N_\sigma/N\). The PSDD is a compact representation of that same population mass function, useful when many ranking types have positive mass. This mirrors the paper’s constrained preference queries, rather than its use of probability over outcomes or of noise over finite profiles.
The anchor is Theorem 1, proved in this paper. It states that given a PSDD \(n\), inducing distribution \(\Pr_n\) and having size \(s_n\), and a Boolean formula \(\alpha\) represented by an SDD \(m\) of size \(s_m\), if the SDD and PSDD use the same vtree, then \(\Pr_n(\alpha)\) can be computed in time \(O(s_n s_m)\). In the mirror, \(\Pr_n(\alpha)\) is exactly the fraction of the cohort satisfying the property. The shared-vtree condition is part of the question’s input promise. The paper provides an arithmetic-operation bound; exact running time also depends on the binary encoding of the rational parameters.
I expect this mirror to be Class A under that representation: the paper directly gives an efficient exact algorithm. The claim is deliberately limited to constrained-query inference, not every result in the paper. The supplied paper has no numbered hardness result to anchor instead.
The weak point is that the paper already represents probability mass over rankings and computes mass on ranking constraints. That distribution can naturally be interpreted as a population’s composition, so the authors may recognize this as their existing problem under a population reading. If ChoCo’s novelty test treats that as the paper’s model already being continuous-population, this is a collision rather than a new continuization result. There is also a representation caveat: this mirror uses the paper’s succinct PSDD input; with only an explicit support list, the query can be answered by summing over that list, and Theorem 1’s compressed-input bound does not follow automatically. The case for mirrorability is stronger than the case for novelty.
The proposed mirror is a genuine computational result, so “there is no theorem to mirror” is not the case. The problem is that Theorem 1 already has the continuous object the mirror proposes. A PSDD defines a probability mass on complete rankings, and Theorem 1 computes the mass satisfying a ranking constraint. Set each type’s population share to \(\mu_\sigma=\Pr_n(\sigma)\); then the cohort share satisfying \(\alpha\) is exactly
\[
\sum_{\sigma\models\alpha}\mu_\sigma=\Pr_n(\alpha).
\]
Reading those masses as fractions of a cohort changes no input, computation, or output. This is a collision with the paper’s existing model, not merely an outside result that happens to answer a new question.
The high-multiplicity scenario is sensible: many people can share a ranking type. But the paper already represents distributions over precisely those types. The input-format distinction does not create a new mirror. With ChoCo’s explicit support list, the query can be answered by checking each of the \(\tau\) rankings and summing its mass. The faster bound in Theorem 1 applies to a succinct PSDD representation, which is the paper’s own representation of the preference distribution.
Theorem 2 does not provide an escape route: it learns the parameters of that same distribution over rankings from incomplete observations. Recasting the learned probabilities as population shares still leaves the same model. Adding a costed campaign or a collective winner-robustness objective might lead to worthwhile ChoCo research, but that adds a new computational problem; it is not a continuous mirror of a result this paper proves.
The negative case is strongest for the proposed query mirror, and weaker as a universal claim. One can append new social-choice tasks to the paper’s model, but that does not rescue this anchor: the mass-query computation is already formulated continuously in the 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.