Accept More, Reject Less: Reducing up to 19% Unnecessary Desk-Rejections over 11 Years of ICLR Data

Xiaoyu Li, Zhao Song, Jiahao ZhangSubmission to ICLR 2026 · Rejected

TL;DR

Major AI conferences have begun enforcing strict per-author submission caps, typically desk-rejecting excess papers using crude heuristics like submission ID ordering that cause unnecessary "collateral damage" to co-authors. This paper formalizes submission-limit compliance as an optimization problem and proposes a fast two-stage algorithm (linear programming relaxation followed by deterministic rounding) that strictly respects author quotas while rescuing up to 19.23% more papers for peer review.

What Problem Does This Paper Address?

The exponential growth of paper submissions has overwhelmed program committees and peer reviewers across AI conferences. To stem the tide, venues such as CVPR, ICCV, AAAI, and KDD have instituted strict author-level submission limits (typically capping authors at $b \in [7, 25]$ submissions).

Under current conference rules, if an author exceeds the limit, excess papers are automatically desk-rejected based on submission timestamp or paper ID. Because papers usually involve multiple collaborators, a single prolific author can trigger the desk-rejection of collaborative papers, severely penalizing co-authors (especially early-career researchers who only have one or two submissions). The authors ask: Can conferences enforce author-level limits while rejecting the absolute minimum number of papers?

Methods

The authors model the policy challenge as a discrete optimization task called the Maximum Desk-Acceptance Submission Limit Problem:

$$\max_{x \in \{0, 1\}^m} \mathbf{1}^\top x \quad \text{s.t.} \quad Ax \le b \cdot \mathbf{1}_n$$

where $A \in \{0, 1\}^{n \times m}$ is the author-by-paper authorship matrix, $x_j \in \{0, 1\}$ denotes whether paper $j$ is admitted to peer review, and $b$ is the per-author submission cap.

Because this integer linear program (ILP) is related to the NP-hard multidimensional knapsack problem, the authors introduce a scalable two-stage approach (OPTREJECT):

  1. LP Relaxation: Relax the binary constraints to $x \in [0, 1]^m$ and solve the resulting continuous linear program using standard off-the-shelf solvers (e.g., PuLP). Unconstrained authors and papers (those with no co-authors exceeding $b$) are pruned beforehand for speed.
  2. Greedy Max-Rounding (MAXROUNDING): Iteratively round the largest fractional variable $x_\ell$ up to $1$. If setting paper $\ell$ to accepted causes any author $i$ on that paper to exceed the limit $b$, the algorithm greedily drops (sets to $0$) a sufficient set of remaining fractional papers authored by $i$ to restore feasibility.

The rounding phase provably guarantees that the final binary decisions satisfy all author constraints in $O(nk_1 + mk_1k_2)$ time, where $k_1$ is the max papers per author and $k_2$ is the max authors per paper.

Datasets

Because conference desk-rejection records are typically private, the authors constructed a real-world evaluation dataset by scraping all public submission records from 11 years of ICLR (2013–2014 and 2017–2025) via the OpenReview API (2015 and 2016 lacked public or non-workshop records). The datasets scale from 67 papers and 161 authors in 2013 to 11,672 papers and 38,495 authors in 2025.

Key Insights and Ideas

  • Conference desk-rejection is a social welfare problem: Current policies (rejecting by paper ID or first-come-first-served) are arbitrary and discard valuable papers unnecessarily. Framing the process as maximizing author welfare yields substantial paper recovery without relaxing submission caps.
  • Fractional LP solutions are close to integral: Although general ILPs are hard, real-world authorship networks are sparse. The LP relaxation quickly isolates which papers can safely move forward, leaving only a small fractional boundary for the greedy rounding heuristic to resolve.
  • Benefits scale with submission volume: As conferences grow and co-authorship graphs become denser, naive heuristic pruning becomes increasingly wasteful, making combinatorial optimization significantly more impactful for large conferences than small ones.

Key Results

  • Up to 19.23% fewer desk rejections: Compared against current conference practices (ALLREJECT) and a sequential baseline (FORWARDREJECT), OPTREJECT consistently retained the most papers across all tested caps ($b \in [1, 32]$). For example, on ICLR 2024 at $b = 22$, the method reduced rejections by 19.23% (rejecting 21 papers instead of 26).
  • Consistently superior at scale: On recent large cohorts (ICLR 2024 and 2025), the optimization approach beat standard conference baselines across nearly every limit $b$, often saving dozens to hundreds of papers from unwarranted rejection.
  • Highly practical runtime: Solved entirely on a standard 2-vCPU machine (13 GB RAM, no GPUs), every single annual dataset solved within 53.64 seconds, proving the method can be integrated directly into conference management platforms with negligible overhead.

Limitations and Future Work

  • Equal Paper Weighting: The formulation treats all papers identically. It does not incorporate author preferences (e.g., an author might prefer to keep their primary-author paper over a minor collaboration) or paper quality estimates.
  • Synthetic Policy Setting: ICLR itself did not historically enforce submission limits during these years; the experiments simulate what would have happened had those limits been in place.
  • Future Directions: The authors suggest extending the method into an end-to-end conference management pipeline that jointly handles author-preference-aware paper preservation and downstream reviewer assignment.

Summary updated

Abstract

The explosive growth of AI research has driven paper submissions at flagship AI conferences to unprecedented levels, necessitating many venues in 2025 (e.g., CVPR, ICCV, KDD, AAAI, IJCAI, WSDM) to enforce strict per-author submission limits and to desk-reject any excess papers by simple ID order. While this policy helps reduce reviewer workload, it may unintentionally discard valuable papers and penalize authors’ efforts. In this paper, we ask an essential research question on whether it is possible to follow submission limits while minimizing needless rejections. We first formalize the current desk-rejection policies as an optimization problem, and then develop a practical algorithm based on linear programming relaxation and a rounding scheme. Under extensive evaluation on 11 years of real-world ICLR (International Conference on Learning Representations) data, our method preserves up to 19.23% more papers without violating any author limits. Moreover, our algorithm is highly efficient in practice, with all results on ICLR data computed within at most 53.64 seconds. Our work provides a simple and practical desk-rejection strategy that significantly reduces unnecessary rejections, demonstrating strong potential to improve current CS conference submission policies.
BibTeX
@misc{li2025accept,
  author = {Xiaoyu Li and Zhao Song and Jiahao Zhang},
  title = {{Accept More, Reject Less: Reducing up to 19\% Unnecessary Desk-Rejections over 11 Years of ICLR Data}},
  year = {2025},
  eprint = {2506.20141},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2506.20141},
}
Figure