Unprecisiated (raw) natural language does not lend itself to computation. Natural languages are intrinsically imprecise. A major source of imprecision is unsharpness of class boundaries. Unsharpness of class boundaries is coextensive with fuzziness. In this perspective, most human concepts are fuzzy. Everyday examples are near, fast, sweet, similar, etc. To deal with fuzzy concepts what is needed is fuzzy logic. Computing with Words is based on fuzzy logic. A CW engine has two principal components: a Precisiation module and a Computation module. The function of the Precisiation module is that of constructing a computational model of the information, I, which is to be computed with. In CW, a computational model of a proposition, p, drawn from a natural language, is represented as a generalized constraint. Precisiated I is the input to the Computation module. Computation is carried out through propagation and counterpropagation of generalized constraints. What comes into play at this stage is the machinery of Granular Computing, GrC. In this machinery, a principal tool is the Extension Principle of fuzzy logic. Application of this principle reduces the problem of computation with natural language to that of solving a mathematical program.

Slides

Large distributed systems, aka Cloud Computing, are a Zoo: machines being added and removed, configuration changes, usage spikes, and load balancing; distributed applications must also handle failures such as network partitions and crashes. We developed ZooKeeper to help the distributed application developer build applications in this chaotic environment. It provides a simple abstraction for coordination that developers can use extensively to handle changes and failures. In our experience, applications that use ZooKeeper are not only more robust but also easier to develop than those using ad-hoc solutions. This talk will motivate the design of ZooKeeper, show how it is used, and share insights gained from its use in production.

Slides

Traditional information systems return answers after a user submits a complete query. Users often feel "left in the dark" when they have limited knowledge about the underlying data, and have to use a try-and-see approach for finding information. A recent trend of supporting autocomplete in these systems is a first step towards solving this problem. In this talk we will discuss a new information-access paradigm, called "interactive search" or "search as you type," which allows users to search and explore underlying data as they type in query keywords. We focus on how to answer such search queries on large data sets efficiently (within milliseconds per query) in order to achieve an interactive speed. For illustration purposes, we will demonstrate several systems powered by these techniques.

Slides

The wealth of freely available, structured information on the Web is constantly growing. This is especially true for public data from and about governments and administrations. Data-providing projects, such as DBpedia and Freebase from the linked open data community, as well as structured data from domain-specific sites, such as senate.gov or UniProt.org, make it possible to integrate data from multiple sources and thus create new data sets with added value. The talk highlights the extreme heterogeneities of such web data and points to methods to overcome them: To make use of this data and create added value for a given domain of interest, a multitude of tasks must be completed: Source selection to identify appropriate and high quality sources, data extraction to create structured data, scrubbing to standardize and clean data, entity matching to associate different occurrences of the same entity, and finally data transformation and data fusion to combine all data about an entity in a single, consistent representation.

Slides