Embedded Files
NOVAS Workshop
at VLDB 2026 in Boston, MA, USA
NOVAS stands for Novel Optimizations for Visionary AI Systems.
We want to bridge the gap between "data management'' and "generative AI'' research.
NOVAS stands for Novel Optimizations for Visionary AI Systems.
We want to bridge the gap between "data management'' and "generative AI'' research.
Where and when:
Monday 31st of August, Location tba
MIT EECS
MIT EECS
Abstract: tba
Cortex AI SQL
Harvard University
Harvard University
Abstract: tba
U Chicago
U Chicago
Abstract: tba
We are calling for work or early ideas which may be deemed innovative, controversial, or disruptive if considered from the perspective of more established research areas.
We are calling for work or early ideas which may be deemed innovative, controversial, or disruptive if considered from the perspective of more established research areas.
Recent advances in large language models (LLMs) have enabled a new generation of AI-powered systems and data management architectures, where reasoning, semantics, and learning are first-class components of data processing. The NOVAS workshop asks: How should system architectures, execution models, and optimization techniques evolve when LLM inference becomes a core system primitive? What are the right abstractions, optimization strategies, and benchmarks for serving LLM-powered workloads efficiently at scale? How do we trade off performance, cost, energy, and accuracy when data systems integrate reasoning, retrieval, and multi-agent execution?
We invite work and early ideas that address these questions through system design, optimization, or theoretical analysis, including contributions that may fall outside traditional database or ML categories but offer clear system-level insights.
Topics of particular interest for the workshop include, but are not limited to:
Declarative systems to compose AI agents and “multi-agent" systems for data processing.
Implementation and optimization for semantic operations, including semantic joins, semantic aggregations, semantic filters.
Multimodal embeddings and semantic question answering over multiple modalities.
DB-inspired techniques to optimize the forward pass in transformer-based architectures.
Scheduling and sharing of large batch workloads of hybrid relational-AI queries.
System-level methods for efficient LLM serving: performance, energy, and cost trade-offs
Strategies for generative AI-based data processing, e.g., RAG, chain-of-thought reasoning.
Benchmarks for data processing tasks using LLMs.
IMPORTANT DATES:
IMPORTANT DATES:
- Submission deadline: June 13, 2026
- Author notification: July 13, 2026
- Camera-ready version: July 31, 2026
- Workshop day: August 31, 2026
All deadlines are 11:59PM PST.
All deadlines are 11:59PM PST.
Submission website: tba
Organizing Committee
Organizing Committee
Gerardo Vitagliano
Gerardo Vitagliano
Postdoctoral Associate, MIT
Chunwei Liu
Chunwei Liu
Postdoctoral Associate, MIT
Liana Patel
Liana Patel
5th-year PhD Student, Stanford
Huan Sun
Huan Sun
Postdoctoral Fellow, Harvard
Andreas Kipf
Andreas Kipf
Professor, TU Nuremberg
Andreas Kipf
Andreas Kipf
Professor, EUROCOM
For any questions regarding the workshop please contact us at chairs@novasworkshop.org
For any questions regarding the workshop please contact us at chairs@novasworkshop.org
SUBMISSION GUIDELINES:
SUBMISSION GUIDELINES:
The workshop will accept regular and short papers. We welcome short papers that present exciting work in progress, dataset contributions, or visionary/outrageous ideas. All papers have to be submitted in a single anonymous format, and must be prepared in accordance with the PVLDB template available at https://vldb.org/pvldb/volumes/19/formatting. All submission to the workshop must adhere to the diversity and inclusion writing guidelines from PVLDB and ACM. The following are the page limits (excluding references):
- Regular papers: 6 pages
- Short papers: 4 pages
All submissions (in PDF format) should be sent to OpenReview.
All submissions (in PDF format) should be sent to OpenReview.
Submission link: tba
Submission link: tba
REVIEWING PROCESS:
REVIEWING PROCESS:
Submissions will be single blind: authors cannot see reviewer names, but reviewers can see author names. We use OpenReview to host papers and the reviewing process will be public. This means that reviewers' comments can be seen by all, during the submission and for accepted papers after decision, although the reviewers' identity will remain anonymous.
Conflicts of Interests (COIs) are handled using the same rules of VLDB 2026.
The use of LLMs is allowed as a general-purpose assist tool. Authors and reviewers should understand that they take full responsibility for the contents written under their name, including content generated by LLMs that could be construed as plagiarism or scientific misconduct (e.g., fabrication of facts). LLMs are not eligible for authorship.
Page updated
Google Sites
Report abuse