Introduction
The September 2020 issue of the Proceedings of the VLDB Endowment (PVLDB) marks the beginning of Volume 14. This "Front Matter" serves as an editorial roadmap, detailing significant structural changes to the journal and highlighting a major shift in the database research landscape: the integration of machine learning into core system components.
Structural Evolution of PVLDB
To better serve the evolving database community, the editors introduced several key tracks and procedural updates. These changes aim to bridge the gap between academic novelty and industrial practicality. The editors identify three main gaps in the traditional publishing model: the tension between practicality and novelty, the lack of a dedicated home for benchmarking, and the ongoing reproducibility crisis.
New Paper Categories
- Scalable Data Science (SDS): This track focuses on the design and evaluation of real-world, large-scale data engineering solutions. By accepting 8-page papers, the journal acknowledges that the engineering experience of building massive systems is a valuable scientific contribution.
- Benchmarks (EA&B): Recognizing that benchmarks are the engine of research progress, a dedicated home was created for rigorous benchmarking studies, previously grouped under general "Experiments and Analysis."
Editorial Leadership and Reviewing
The editorial team has also streamlined the review process. A new "minor revision" decision has been implemented to accelerate the publication of high-quality research that requires only small adjustments, reducing the burden on both authors and reviewers.
The "ML-for-DB" Revolution
A defining characteristic of this issue is the prevalence of machine learning. Five out of the six featured research papers utilize ML to address traditional database challenges. This shift suggests that learned models are increasingly replacing manual heuristics in core components. Notable examples from this issue include:
- Learned Indexes: A benchmark comparing traditional data structures like B-Trees against models that learn data distributions.
- Cardinality Estimation: "NeuroCard," a neural network approach to estimating join sizes across multiple tables.
- Entity Matching: Leveraging pre-trained language models (BERT-style architectures) to match product names and records.
Reproducibility and Artifacts
PVLDB has intensified its focus on the reproducibility crisis. Authors are now encouraged to provide supplemental code and data. Papers that provide stable, verified links to these resources receive an "Artifacts available" tag, a process overseen by dedicated reproducibility chairs. While not yet a full re-execution of experiments, this step ensures that the foundational data for research remains accessible to the community.
Conclusion
Volume 14, Issue 1 sets a forward-looking tone for the database community. By valuing scalable engineering, rigorous benchmarking, and the transformative potential of machine learning, PVLDB continues to adapt its editorial standards to the needs of both academia and industry.
Summary updated