Which Coauthor Should I Nominate in My 99 ICLR Submissions? A Mathematical Analysis of the ICLR 2026 Reciprocal Reviewer Nomination Policy

Zhao Song, Song Yue, Jiahao ZhangSubmission to ICLR 2026 · Withdrawn

TL;DR

To handle soaring submission volumes, conferences like ICLR 2026 introduced reciprocal reviewer policies where submissions must nominate an author to review—and any paper nominating an irresponsible reviewer is desk-rejected. This paper analyzes this mechanism from the authors' perspective, formulating the task of choosing nominees across many submissions to minimize desk-rejection risk. The authors prove that while basic risk minimization is solved by a simple greedy choice, adding limits to prevent catastrophic single-author failures maps cleanly onto minimum-cost network flows and linear programming.

What Problem Does This Paper Address?

Top machine learning conferences face unsustainable reviewing workloads. To cope, ICLR 2026 enacted a reciprocal reviewer policy: each submission must designate an author to serve as a reviewer. If a nominated author is flagged as "irresponsible" (e.g., failing to submit reviews), every paper that nominated them is summarily desk-rejected.

For researchers and labs submitting multiple papers, this creates a high-stakes portfolio problem:

  • Nominating the single "most reliable" coauthor on every submission minimizes the raw expected number of desk rejections.
  • However, this creates a catastrophic single point of failure: if that one person drops the ball, dozens of papers could be desk-rejected simultaneously.

The authors study how authors or research groups can optimize their reciprocal reviewer nominations to balance expected rejections against correlated worst-case risks.

Methods

The paper models authors $j \in [m]$ with known irresponsibility probabilities $p_j \in [0, 1]$ and an authorship incidence matrix $a \in \{0, 1\}^{n \times m}$. They formalize three optimization variants:

  1. Unconstrained Risk Minimization (Basic Problem): Minimizes $\sum_{i,j} x_{i,j} p_j$ subject to each paper nominating exactly one coauthor.
  2. Hard Author Nomination Limits: Restricts any author from being nominated more than $b$ times ($\sum_i a_{i,j} x_{i,j} \le b$). Because naive LP relaxation can yield fractional answers, the authors construct an exact mapping from this integer program to a minimum-cost circulation (flow) problem on a directed graph with integral capacities and costs equal to $p_j$.
  3. Soft Author Nomination Limits: Hard limits are often infeasible (e.g., when a prolific single author writes more than $b$ papers). The authors relax the bound into a non-smooth $\ell_1$-penalty: $\lambda \sum_{j} \max\{0, \sum_i a_{i,j} x_{i,j} - b\}$. Using the epigraph reformulation trick, they convert this into an equivalent linear program (LP) with auxiliary variables $y_j \ge \sum_i a_{i,j}x_{i,j} - b$ and provide an efficient rounding scheme.
[Source] --(cap=b, cost=0)--> [Author j] --(cap=1, cost=p_j)--> [Paper i] --(cap=1, cost=0)--> [Sink]

Datasets

Not discussed in the paper. The paper is purely theoretical and algorithmic. It references empirical conference sizes from prior literature (noting that the number of vertices and edges in ICLR scales on the order of $10^4$) to contextualize the algorithmic complexity.

Key Insights and Ideas

  • The "Reliability Trap" in Greedy Optimization: The basic unconstrained problem decomposes independently across papers and can be solved greedily in $O(\text{nnz}(a))$ time. However, this strategy heavily concentrates nominations on the single lowest-$p_j$ author, magnifying catastrophic tail risk.
  • LP Relaxation Can Produce Non-Integral Optima: For the hard-limit problem, standard LP relaxation does not guarantee integer solutions; the authors provide an explicit counterexample demonstrating fractional global minima.
  • Exact Min-Cost Flow Duality: By framing hard nomination limits as routing $n$ units of flow from authors to papers, the integer problem satisfies network total unimodularity. This guarantees that optimal solutions are integral without requiring heuristic branch-and-bound or rounding.
  • Sparsification to Observed Authorships: By setting $x_{i,j} = 0$ wherever $a_{i,j} = 0$, the LP formulation scales with the number of author-paper incidences $\text{nnz}(a)$ rather than $n \times m$, making interior-point methods practical even for large conferences.

Key Results

  • Basic Problem: Solvable to global optimality in $O(\text{nnz}(a))$ time via a paper-by-paper minimum search.
  • Hard Limit Formulation: Exactly reducible to minimum-cost circulation. Leveraging modern flow algorithms (e.g., Brand et al., 2021), it can be solved in $\widetilde{O}(N + M^{1.5})$ time, guaranteeing exact integral solutions.
  • Soft Limit Formulation: Reformulated into a smooth LP solvable in $\widetilde{O}(K^{2.37} \log(K/\delta))$ time for $K = O(\text{nnz}(a))$ non-zero variables. The proposed rounding algorithm (Algorithm 1) produces a valid integer assignment respecting all paper nomination requirements in $O(\text{nnz}(a))$ time.

Limitations and Future Work

  • Known Irresponsibility Probabilities: The model assumes $p_j$ is known, fixed, and statistically independent. In practice, estimating a colleague's probability of defaulting on reviews is subjective, noisy, and socially awkward.
  • No Conference-Side Load or Expertise Matching: The framework optimizes solely from the authors' perspective. It does not account for reviewer topic expertise, conflicts of interest with assigned submissions, or global reviewer quotas enforced by conference chairs.
  • Decentralized Game Theory: The paper models a single coordinated entity (e.g., an author or large lab). In multi-institution collaborations with conflicting author incentives, reviewer nomination becomes a strategic game that remains unmodeled.

Summary updated

Abstract

The rapid growth of AI conference submissions has created an overwhelming reviewing burden. To alleviate this, recent venues such as ICLR 2026 introduced a reviewer nomination policy: each submission must nominate one of its authors as a reviewer, and any paper nominating an irresponsible reviewer is desk-rejected. We study this new policy from the perspective of author welfare. Assuming each author carries a probability of being irresponsible, we ask: how can authors (or automated systems) nominate reviewers to minimize the risk of desk rejections? We formalize and analyze three variants of the desk-rejection risk minimization problem. The basic problem, which minimizes expected desk rejections, is solved optimally by a simple greedy algorithm. We then introduce hard and soft nomination limit variants that constrain how many papers may nominate the same author, preventing widespread failures if one author is irresponsible. These formulations connect to classical optimization frameworks, including minimum-cost flow and linear programming, allowing us to design efficient, principled nomination strategies. Our results provide the first theoretical study for reviewer nomination policies, offering both conceptual insights and practical directions for authors to wisely choose which co-author should serve as the nominated reciprocal reviewer.
BibTeX
@misc{song2025which,
  author = {Zhao Song and Song Yue and Jiahao Zhang},
  title = {{Which Coauthor Should I Nominate in My 99 ICLR Submissions? A Mathematical Analysis of the ICLR 2026 Reciprocal Reviewer Nomination Policy}},
  year = {2025},
  eprint = {2512.17950},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2512.17950},
}
Figure