In the traditional model of computer science research, publication is often a high-stakes, "one-shot" game. Authors target a single annual deadline, and reviewers provide a binary accept-or-reject decision. This "conference-first" culture, as noted by Hendrik Blockeel and colleagues in A Revised Publication Model for ECML PKDD, often leads to a "Resubmission Paradox." As Yichi Zhang explores in A System-Level Analysis of Conference Peer Review, papers are frequently rejected from one venue only to be accepted at another months later with minimal changes, wasting reviewer effort and delaying scientific progress.
To solve this, a growing movement across fields like data management, machine learning, and visualization is shifting toward continuous, iterative peer review. By merging the rapid dissemination of conferences with the rigorous revision cycles of journals, researchers are transforming the "tombstone" PDF into a living document.
Breaking the Binary: The Hybrid Publication Model
The core innovation is the decoupling of the review timeline from the event date. Instead of a single annual cutoff, venues like the Proceedings of the VLDB Endowment (PVLDB) utilize monthly submission deadlines.
As Jian Pei and Sihem Amer-Yahia explain in their Front Matter, this hybrid framework allows for a formal "revise and resubmit" phase. Rather than rejecting a paper for a fixable technical flaw, reviewers can request specific improvements. This mechanism ensures that the final published work is technically mature and has addressed expert critiques—a level of rigor typically reserved for journals, but delivered at the speed of a conference. Magdalena Balazinska observes in Front Matter (2020) that this iterative process leads to a significant majority of accepted papers undergoing substantial, meaningful improvements before they ever reach the stage.
Technical Shepherding and High-Bandwidth Feedback
Intuition suggests that a paper is better if the authors and reviewers can actually talk to each other. Traditional reviews are often a "black box," but newer models emphasize shepherding.
In this process, a committee member is assigned to guide authors through specific refinements. Insik Shin and the RTSS 2023 chairs describe this as a way to ensure promising research isn't lost due to addressable presentation flaws. Similarly, Geoffrey J. Gordon and David B. Dunson experimented at AISTATS 2011 with allowing authors to submit fully revised manuscripts during the rebuttal phase. This "high-bandwidth" loop allows reviewers to evaluate technical corrections directly within the context of the paper, rather than guessing if the authors will fix the issues.
Reducing Redundancy via Carry-Over Reviews
One of the greatest inefficiencies in science is the "start-from-scratch" nature of resubmissions. When a paper is rejected, the next set of reviewers usually has no idea what the previous set found.
Modern frameworks like the EMNLP 2020 process allow authors to carry over previous reviews and provide a formal response to them when resubmitting. This transforms the review process from a series of isolated events into a continuous conversation. As Björn Bringmann notes regarding the ECML PKDD journal track, this carry-over mechanism allows papers requiring extensive revisions to move between annual cycles without restarting the clock, prioritizing the depth of research over strict adherence to a calendar.
Toward Versioned and Modular Science
If we view research as an evolving process rather than a static product, the "final" paper is just one snapshot in time. Shagun Sodhani and colleagues, summarizing the NeurIPS 2019 Retrospectives Workshop, propose treating research like software version control. In this "versioned research" model, updates and corrections are issued as citable "commits," allowing the scientific record to be corrected and refined continuously.
This aligns with the "Open Knowledge" framework proposed by Junsuk Rho and others in Fostering cultural change in research. By sharing research in discrete, time-stamped modules and seeking iterative community validation, the focus shifts from a competitive race to be "first" to a collaborative effort to be "most robust."
As Frank van Harmelen argues in his vision for an Innovation Accelerator, replacing linear pipelines with cyclic systems of discovery and fusion is essential for science to keep pace with the modern era. By embracing continuous revision, the research community is building a more resilient, transparent, and efficient engine for progress.