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

Xiaoyu Li, Zhao Song, Jiahao Zhang

Introduction

As artificial intelligence research continues its exponential growth, major conferences face an unprecedented challenge in managing reviewer workloads. To mitigate burnout, many flagship venues have implemented strict per-author submission limits. However, the current "industry standard" for enforcing these limits—desk-rejecting "excess" papers based on arbitrary submission IDs—is inefficient. This paper formalizes desk-rejection as a combinatorial optimization problem and introduces an algorithm that preserves up to 19.23% more papers for review while strictly adhering to all author-level constraints.

The Problem: Arbitrary Desk-Rejections

The "AI boom" has led to a surge in submissions, forcing conferences like CVPR, AAAI, and KDD to enforce hard limits on the number of papers a single individual can author, typically ranging from 7 to 25.

Table 1: Submission limits across major AI conferences over time.

Currently, violations are handled bluntly: if an author exceeds the limit, the conference desk-rejects "extra" papers, usually based on the order of submission (highest IDs first). Because papers often have multiple authors, this approach is suboptimal. Rejecting Paper A to satisfy Author X's limit might be unnecessary if rejecting Paper B instead would satisfy the limits for both Author X and Author Y. By treating desk-rejection as a clerical task rather than a global optimization problem, conferences inadvertently discard thousands of hours of research effort.

Methodology: Maximum Desk-Acceptance

The authors propose the Maximum Desk-Acceptance Submission Limit Problem, aiming to maximize the total number of papers accepted for review subject to the constraint that no author appears on more than $b$ papers.

  1. Integer Programming (IP) Formulation: The problem is first modeled as a discrete optimization task. Since solving this exactly is NP-hard and impractical for large-scale conferences, the authors employ a relaxation strategy.
  2. Linear Programming (LP) Relaxation: The "binary" requirement (accept/reject) is relaxed to allow fractional values, enabling the use of efficient modern LP solvers.
  3. MaxRounding Strategy: A specialized rounding algorithm converts fractional results back into hard decisions. This scheme ensures that final integer solutions never violate author limits while maximizing the total paper count.

Dataset and Empirical Context

The study utilizes 11 years of real-world ICLR data (2013–2025), crawled via the OpenReview API. ICLR is uniquely suited for this study as it provides public submission records detailed enough to reconstruct authorship networks.

Figure 1: The exponential growth of ICLR submissions from 2013 to 2025.

The dataset includes over 38,000 authors and 11,000 papers in the 2025 cycle alone. As shown in Figure 1, the scale of submissions has increased by orders of magnitude, making efficient management more critical than ever.

Key Insights and Results

The research highlights that desk-rejection is fundamentally a network problem. Because authors collaborate, the submission pool forms a complex graph where a single strategic rejection can resolve limit violations for multiple authors simultaneously.

Performance Gains

In the ICLR 2024 dataset, the proposed algorithm saved 19.23% more papers compared to standard policies when the submission limit was set to 22. The method consistently outperformed traditional "ForwardReject" (sequential acceptance) and "AllReject" (greedy rejection) across every year of data and every tested submission limit ($b=4$ to $b=25$).

Scalability and Fairness

The benefits of this optimized approach become more pronounced as conference size increases. For a massive venue like ICLR 2025, the entire optimization process takes less than 54 seconds on a standard server. This eliminates computational overhead as a barrier to adopting fairer policies. The approach shifts the focus from punishing authors to maximizing author welfare by keeping as much research as possible in the peer-review pipeline.

Conclusion

While the algorithm treats all papers as equal in quality—since they have not yet been reviewed—it significantly improves the efficiency of the submission process. Future work may involve integrating this optimization step directly into reviewer-assignment systems like TPMS to create a seamless, end-to-end conference management pipeline.

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