Database Seminar Series (2008-2009)

This talk argues that our community should seriously consider the problem of managing unstructured data (e.g., text, Web pages, emails, memos, and ads). While many other research communities have also been long laboring on this problem, I argue that we can make unique contributions, by adopting a structure and system focus: (a) make it very easy to extract structures from the raw data, and (b) build end-to-end systems, instead of working on just isolated research problems.

I then describe Cimple, a joint project between Wisconsin, Yahoo, and Microsoft that works toward the above goals. First I describe our current protype system that employs extraction, integration, and mass collaboration to manage unstructured data. Next, I discuss current efforts in applying the system to a range of applications, including structured Web portals, on-the-fly data integration, personal information management, and ad management. Finally, I discuss lessons learnt from the current Cimple work, and ways forward.

The Music Information Retrieval Evaluation eXchange (MIREX) is a community-based formal evaluation framework coordinated and managed by the International Music Information Retrieval Systems Evaluation Laboratory (IMIRSEL) at the University of Illinois at Urbana-Champaign (UIUC). IMIRSEL has been funded by both the National Science Foundation and the Andrew W. Mellon foundation to create the necessary infrastructure for the scientific evaluation of the many different techniques being employed by researchers interested in the domains of Music Information Retrieval (MIR) and Music Digital Libraries (MDL). This presentation is intended to provide an overview of MIREX including the unique challenges posed when trying evaluate specialized music information retrieval systems. This talk will place MIREX in the historical context of the International Conferences on Music Information Retrieval Systems (ISMIR) from which it developed. This lecture will also highlight the speaker's positive and negative experiences in running MIREX as a jumping off point for outlining the motivation for developing a more robust and sophisticated Networked Environment for Music Analysis (NEMA) architecture for the future sustainability of the MIREX program.

More information about MIREX is available can be found below:

Downie, J. Stephen (2008). The Music Information Retrieval Evaluation Exchange (2005-2007): A window into music information retrieval research. Acoustical Science and Technology 29 (4): 247-255. Available at: http://dx.doi.org/10.1250/ast.29.247

A recent article in the Philadelphia Inquirer:

Avril, Tom. 2008. Analyzing music the digital way: Computers have exquisite ears. Philadelphia Inquirer (Sept. 22, 2008). Available: http://www.philly.com/inquirer/entertainment/20080922_Computers_have_exquisite_ears.html

Data is the bloodline of modern Web applications. Be it for business analysis, feed aggregation, or to build a domain-specific social network, data must flow from sources and services to the users and back. And who knows more about how to make data flow than a database plumber? In the face of the increasing need for agile Web development, new languages (Ruby on Rails, Linq, Links, and XQuery to name a few) have emerged that integrate native data processing capabilities along with more traditional programming language features.

In this talk, I will argue that supporting efficient data processing in the context of a full-fledge programming language raises difficult, and largely unexplored, research questions. I will then present the DXQ project, which extends XQuery (the standard language for querying XML) with modules, imperative features, Web services support, and distribution. I will then illustrate how DXQ can support rapid development of Web applications, using three distinct examples: the DNS (Domain Name Server) protocol, the Narada overlay network protocol, and workflow-based processing of RSS feeds. If time permits, I will show how to adapt key existing database optimization techniques to such a language.

This is joint work with Mary Fernandez (AT&T Labs), Giorgio Ghelli (Universita di Pisa), Trevor Jim (AT&T Labs), Kristi Morton (University of Washington), Nicola Onose (UCSD), and Kristoffer Rose (IBM Watson).

Financial misstatements from Fortune 500 companies, dead people who still vote in elections, world-class gymnasts with uncertain birth dates: insiders often have the power and ability to tamper with electronic records to further their own goals. As electronic records supplant paper records, it becomes easy to carry out such manipulations without leaving a paper trail of illicit activities that can be used to track and prosecute offenders after the fact. The US Sarbanes-Oxley Act is perhaps the most (in)famous legislation to target these abuses, but it is just one of many regulations that mandate long-term tamper-free or tamper-evident retention of electronic records, all with the goal of maintaining societal trust in business and government at reasonable cost.

In this talk, we will discuss some of the technical challenges posed by the need for "term-immutable" retention of records. We will describe how industry has responded to these challenges, the security weaknesses in current product offerings, and the role that researchers and government can play in addressing these weaknesses. We will give an overview of research progress to date and describe the major open research problems in this area.

The Orchestra project focuses on how to support exchange of data (and updates) among collaborators with evolving databases, in a way that accommodates disagreement, different schemas, and different levels of authority and quality. Orchestra considers collaborators' databases to be logical *peers* into which data can be imported and then locally modified. It allows for a network of *schema mappings* that interrelate peers, annotated with *trust policies* specifying the conditions under which a peer is willing to import data. As a data item is mapped from site to site in the system, its *provenance* is recorded; a peer's trust policies use this provenance (and the values of the data) to assign a score to each incoming data item (based on perceived quality or relevance), and the peer then uses this score to reconcile conflicts and compute a consistent data instance, whose contents may be unique to the peer. The scores assigned to the individual sources can even be *learned* based on user feedback about query answers. The end result is a system that allows each database to selectively diverge from the others as appropriate, but to remain "in sync" in all other cases.

Joint work with Todd J. Green, Grigoris Karvounarakis, Nicholas Taylor, Val Tannen, Partha Pratim Talukdar, Marie Jacob, Muhammad Salman Mehmood, Koby Crammer, Fernando Pereira, and Sudipto Guha.

The Linking Open Data (LOD) community project is extending the Web by encouraging the creation of interlinks (RDF links between data items identified using dereferenceable URIs). This emerging web of linked data is closely intertwined with the existing web, since data items can be embedded into web documents, and RDF links can reference classic web pages. Abundant linked data justifies extending the capabilities of web browsers and search engines, and enables new usage scenarios, novel applications, and sophisticated mashups. This promising direction for publishing data on the web brings forward a number of challenges. While existing data management techniques can be leveraged to address the challenges, there are unique aspects to managing web scale interlinking.

In this talk, we describe two specific challenges; achieving and managing dense interlinking, and understanding the data and metadata that is used within and across datasets in the open data cloud. Approaches to solve these challenges are showcased in the context of LinkedMDB, the first open linked dataset for movies, and the winner of the Triplification Challenge. LinkedMDB has a large number of interlinks (over a quarter million) to several datasets in the LOD cloud, as well as RDF links to related webpages.

Increasingly, organizations capture, transform and analyze enormous data sets. The most prominent examples are Internet companies and e-science. The Map-Reduce scalable dataflow engine has become a popular platform for these applications. Its simple, explicit dataflow programming model is favored by
some over the more traditional high-level declarative approach: SQL. On the other hand, the extreme simplicity of Map-Reduce leads to much low-level hacking to deal with the many-step, branching dataflows that arise in practice. Moreover, users must repeatedly code standard operations such as Join by hand. These practices waste time, introduce bugs, harm readability, and impede optimizations.

We introduce a new dataflow programming language called Pig Latin, which aims at a sweet-spot between SQL and Map-Reduce. It offers SQL-style high-level data manipulation constructs, which can be assembled in an explicit dataflow and interleaved with custom Map- and Reduce-style functions or binaries. Pig Latin also offers nested data structures and other features that appeal to non-SQL programmers.

This talk introduces Pig Latin and its open-source implementation, Pig, which compiles Pig Latin programs into sequences of Hadoop Map-Reduce jobs. In addition to describing the language and philosophy behind Pig, we present some of our research on cross-program optimization and automatically-generated example data sets.

In this talk, we present our recent work on using declarative languages to specify, implement, and analyze secure distributed systems. In the first half of the talk, we describe Secure Network Datalog (SeNDlog), a declarative language that unifies declarative networking and logic-based access control languages. SeNDlog enables network routing, distributed systems, and their security policies to be specified and implemented within a common declarative framework. We describe extensions to distributed recursive query processing techniques to execute SeNDlog programs via authenticated communication among untrusted nodes. In the second half of the talk, we introduce the notion of network provenance naturally captured within our declarative framework, and demonstrate its applicability in the areas of network accountability, network forensic analysis and trust management. We further discuss our ongoing work at optimizing distributed query processors in order to process and maintain network provenance efficiently and securely.

To conclude the talk, we will briefly describe other declarative networking research at Penn, in the areas of adaptive wireless networking and formal network verification.