2015 VLDB Endowment Awards:
- The 2015 VLDB Early Career Researcher Award by the VLDB Endowment:
Christopher Re, for Scalable Integration of Machine Learning and Database Systems Techniques
Christopher Re’s talk at VLDB-15 (research session 25, Thursday September 3 at 13:30):
Using Scalable Inference to Build DeepDive: A Declarative Dark Data System
Abstract for the invited talk: Many pressing questions require that one integrate information from numerous data sources, often expressed in natural languages or in graphics; these forms of media are fraught with imprecision and ambiguity. This talk describes DeepDive, a new type of system designed to cope with these problems. It treats extraction, integration, and prediction as a single inference problem instead of as a series of isolated data tasks. For some paleobiology and materials science tasks, DeepDive-based systems have surpassed human volunteers in data quantity and quality (recall and precision). DeepDive is declarative in that it does not allow users to write algorithms; instead, it asks them to write only features. Our key technical contributions are scalable inference techniques to support this declarative approach, including Hogwild!-style algorithms, and techniques to support feature engineering using classical data management ideas. DeepDive is open source on github and available from DeepDive.stanford.edu.
- VLDB 10-Year Best Paper Award by the VLDB Endowment:
C-Store: Looking back and looking forward, by Mike Stonebraker, Daniel J. Abadi, Adam Batkin, Xuedong Chen, Mitch Cherniack, Miguel Ferreira, Edmond Lau, Amerson Lin, Sam Madden, Elizabeth O’Neil, Pat O’Neil, Alex Rasin, Nga Tran, and Stan Zdonik
This award has been given for innovations in database architecture that impacted the database industry in a profound way.
Abstract for the invited talk: The idea to serialize relational tables on storage column-by-column instead of row-by-row has been around since the 1970s. However, the vast majority of data warehousing and analytical database system implementations stored tables row-by-row in 2005 when the C-Store paper was published. In this talk we will highlight the design decisions made in the C-Store architecture that resulted in a practical, highly performant implantation of a column-store that avoided many of the pitfalls and disadvantages of previous column-store systems, and served as a blueprint for a highly successful commercial column-store implementation (Vertica) — parts of which have been adopted by almost every major data warehousing and analytical database vendor in the decade since the C-Store paper was published. We will then look forward and discuss the future problems, technologies and architectures in the analytical database system space.
2015 VLDB Conference Awards:
- The VLDB PC Best-Paper Award:
Constructing an Interactive Natural Language Interface for Relational Databases, by Fei Li and H. V. Jagadish
The paper proposes an interactive natural language interface for relational databases, which enables novice users to construct complex queries. It improves the usability of an RDBMS, as it enables anyone to use the ask questions to a database system. This paper is likely to start a new line of research as well as products.
For a query expressed in natural language, the interface interacts with the user in several steps (as we do in real life to make our questions more precise) in determining the query semantics and subsequently generating the corresponding SQL. At each step, the system interactively presents to the user its own understanding of the query through alternatives, as opposed to just final answers. The authors rely on a query tree structure to represent the interpretation of an NLP query from the database’s perspective, which facilitates verification by users, and translation into SQL. The system (NLIDB) was implemented following the component-based approach, where each component can be independently constructed, optimized or substituted. The experiments involve real users and verify the feasibility of the approach and illustrate the strengths of the system/approach.
- The VLDB PC Best-Paper Runner Up:
Resource Bricolage for Parallel Database Systems, by Jiexing Li, Jeffrey Naughton, and Rimma V. Nehme
This is a core database systems paper that addresses a real problem, mainly how to deal with the heterogeneity in the machines composing a cluster-≠based database system. The paper proposes a mechanism (“Resource Bricolage”) to make efficient use of heterogeneous hardware when processing a workload in a parallel database system. It addresses a very relevant problem (clusters don’t grow homogeneously) and is the first paper on this subject. The approach is relatively simple and practical, using linear programming to optimize data distribution - and thus resource consumption - in a cluster. The techniques were implemented on top of Microsoft SQL server parallel data warehouse. Overall, this is an excellent and impactful paper. One can envision many extensions that can follow up from this research.