| paper | d4-conf-ijcai-2016-0538 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | unclear |
The paper proves submodularity for hypergraph IC and non-submodularity for HEMI, but neither claim is a numbered result stating an algorithm, guarantee, or complexity classification. The opponent’s population model may suggest a research question, but it cannot supply the required qualifying anchor, so bit (a) fails.
fails bit a — no named computational result to mirror
No result qualifies as an anchor under the required numbered-result criterion; the proposed population versions of HEMI and expected-spread optimization therefore do not establish a surviving mirror.
This paper has no qualifying named computational result to anchor a continuous mirror. It contains no numbered Theorem, Lemma, Corollary, or Proposition asserting a complexity or algorithmic classification. Its main claims are that hypergraph IC influence is submodular (§5) and that HEMI is non-submodular (§7); both are unnumbered, and neither classifies a computational problem as tractable or hard.
HEMI’s campaign scenario suggests a plausible high-multiplicity setting, but proposing a continuous seed-allocation problem for it would not meet your required anchor rule. There are therefore no per-anchor questions to state honestly.
The “no computational result” objection is too narrow. The paper has no numbered complexity theorem, and it does not classify HEMI as tractable or hard. But its named HEMI problem is an explicit optimization problem, and its proofs establish algorithmic facts about that problem: hypergraph IC’s expected-spread objective is submodular, while HEMI’s majority objective is not. Those facts determine whether the standard greedy approximation argument applies. Their numbering is editorial; their computational role is substantive.
A direct mirror of the HEMI result is available. Consider a large campaign population with repeated user types and repeated affiliation templates. A user’s type records everything the model uses: affiliation pattern, propagation behavior, and seeding cost. Let \(\mu_t\) be the mass of users of type \(t\), and let \(x_t\in[0,\mu_t]\) be the seeded mass, subject to a budget such as \(\sum_t c_t x_t\le b\). Groups can be represented by affiliation templates with their own masses and type compositions. The objective is the expected mass of groups in which influenced members exceed half the group’s membership mass.
That asks a computational question about the population itself: how hard is it to choose seed masses that maximize expected majority influence, and what approximation guarantees are possible? The paper’s non-submodularity result makes this especially well motivated: does the loss of diminishing returns survive in the mass model, and if so, what algorithmic structure replaces the greedy guarantee? Propagation rates may need a natural population scaling to keep the continuum limit nontrivial, but that is a model co-modification, not a reason the question ceases to be HEMI’s problem.
This regime has a plausible use case: a large platform campaign across recurring community types, where many users share the same relevant propagation and affiliation characteristics. It does not fit every social network, especially one where each user’s neighborhood is unique. But the claim at issue is that no scenario yields a worthwhile mirror. A credible repeated-community regime is enough to challenge that universal claim.
The paper’s submodularity proof supplies a second mirror: characterize the expected-spread optimization problem when seeds are chosen as population mass, and determine whether the reachability argument yields an efficient approximation algorithm in terms of the type support and model description. That answer might be straightforward, or it might reveal limits of type aggregation; either way, it concerns computational optimization over a continuous society.
The fair limitation is that these are mirrors of the paper’s proved algorithmic structure and defined optimization objectives, not continuizations of a numbered complexity classification. That weakens claims about what the paper itself has already established, but it does not support the stronger conclusion that there is no worthwhile continuous question to ask.
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.