Papers.

Can a Recommender System Induce Serendipitous Encounters?

This chapter presents a beginning effort to apply some ideas about serendipity to information retrieval and information filtering systems, especially in recommenders, to mitigate the over-specialization issue. The museum scenario is particularly interesting because items are arranged in a physical space and users interact with the environment. Thus disregarding context facets makes useless recommendations. Similar remarks are still valid in domains (different from cultural heritage fruition) in witch a physical or virtual space is involved and it represents a pragmatic justification to explain (supposed) serendipitous recommendations. As future work, we expect to carry out more extensive experimentation with more users and wider item collections. We plan also to gather user feedback and feeling by questionnaires focused on qualitative evaluation of the recommendations and the idea of getting suggestions that should surprise them. That is really important for the need to understand the effectiveness of the module in finding unknown items rather the ones that result best rated. Experimentation with users with different cultural levels and with different information seeking tasks are also important to find out which kind of user would like most serendipitous recommendations and to whom they are more useful. We expect also to implement the other suggestions given by Toms (Toms, 2000) and to develop further the heuristic proposed (maybe padding a parameter factor that multiplies the probabilities in order to balance better between categories) or also introduce new heuristics and make an experimental comparison.

lastfm Music Recommendation Dataset

lastfm Music Recommendation Dataset

The Homology of the Real Complement of a k-parabolic Subspace Arrangement

A Picture is Worth a Thousand Questions: Visualization Techniques for Social Science Discovery in Computational Spaces

On the design of autonomic, decentralized VPNs

Decentralized and P2P (peer-to-peer) VPNs (virtual private networks) have recently become quite popular for connecting users in small to medium collaborative environments, such as academia, businesses, and homes. In the realm of VPNs, there exist centralized, decentralized, and P2P solutions. Centralized systems require a single entity to provide and manage VPN server(s); decentralized approaches allow more than one entity to share the management responsibility for the VPN infrastructure, while existing P2P approaches rely on a centralized infrastructure but allow users to bypass it to form direct low-latency, high-throughput links between peers. In this paper, we describe a novel VPN architecture that can claim to be both decentralized and P2P, using methods that lower the entry barrier for VPN deployment compared to other VPN approaches. Our solution extends existing work on IP-over-P2P (IPOP) overlay networks to address challenges of configuration, management, bootstrapping, and security. We present the first implementation and analysis of a P2P system secured by DTLS (datagram transport layer security) along with decentralized techniques for revoking user access.

SocialDNS: A decentralized naming service for collaborative P2P VPNs

The ability to define domain names for resources in a collaborative virtual organization is usually reserved to network administrators through centralized domain name servers. We propose SocialDNS, a decentralized, naming service that gives individual collaborators the power to choose the domain names for their resources. Our approach is based on similar concepts of decentralized naming solutions available in local area networks. We enable short-names for resources by limiting the scope for uniqueness. We also employ a rank-based mechanism for dealing with name conflicts. We evaluate our design through graph level analysis to anticipate scope, bandwidth costs and latency. We also conducted experiments involving Amazon EC2 and PlanetLab to analyze the latencies in a real world environment.

ITAM-containing adaptors regulate NKG2D-mediated cytotoxicity in human NK cells (138.15)

NKG2D is expressed on cytotoxic lymphocytes and is an important immunosurveillance receptor that recognizes stress ligands such as MICA on tumor cells or virally infected cells. Unlike traditional immune activating receptors, NKG2D couples to the non-ITAM-containing transmembrane adaptor DAP10, which recruits PI3K and Grb2 to initiate target cell lysis. Current understanding predicts an ITAM-independent mechanism of NKG2D-mediated cytotoxicity. However, we show that in primary human NK cells, the ITAM-containing adaptors DAP12, FcϵR1γ, and CD3ζ selectively regulate the Grb2-Vav1 branch of NKG2D-mediated signal transduction and cytotoxicity. Furthermore, this ITAM-adaptor regulation of NKG2D is independent of ITAM activation. Thus our data reveal a more complex interplay between NKG2D and other activating receptors than was previously imagined, and we have uncovered a novel mechanism of ITAM-adaptor regulation of cell-mediated cytotoxicity. A more thorough understanding of immunoreceptor signaling networks will lead to more precise therapies when fighting diseases such as cancer, autoimmunity, and infection.

Automated Software Development in Heterogeneous GPU/CPU Environments for Seismic Modeling

Publishing Bibliographic Data on the Semantic Web using BibBase

Releasing Private Contingency Tables

Statistical agencies routinely publish aggregate data in the form of contingency tables. In this paper, we consider the problem of releasing private contingency tables so that the privacy of individual respondents in the table is preserved. We first uncover funda- mental problems with existing cell suppression algorithms that are used for this purpose. We then present a rigorous definition of privacy and a generic algorithmic framework for cell suppression given this definition. Using this framework we build a complete cell suppression solution for the special case of boolean private attributes. We study both theoretically and experimentally the utility of our approach. Along the way, we demonstrate a connection to the query auditing problem in statistical databases and make a foundational contribution to this problem as well. In particular, we analyze an unexamined assumption from the literature regarding the prior knowledge of attackers.

Constraints on Decaying Dark Matter

We explore a dark-matter model in which there are two dark-matter species nearly degenerate in mass, with epsilon = Delta M/M << 1. The heavier particle undergoes two-body decay with a half-life tau, to the lighter dark-matter particle and a noninteracting massless particle. Unlike previous work on decaying dark matter, we explore the regime tau > 100 Myr and non-relativistic kick speeds vk / c = epsilon. Using a set of N-body simulations of isolated dark-matter halos, we show how halos change as a function of tau and vk. We find that tau < 40 Gyr is ruled out for vk > 20 km s^{-1} (epsilon > 10^{-4}) when we compare the simulations to observations of dwarf-galaxy- to cluster-mass dark matter halos. We highlight which set of observations should provide better future constraints for decays and other types of dark-matter physics.

An Agency Theoretic Formulation of Organizational Control Theory

Sticky Cost Behavior: Theory and Evidence

Power Game between the CEO and Labor: Evidence from Outsourcing

Competing for Entrepreneurial Ideas: Matching and Contracting in the Venture Capital Market

Strategic Managerial Compensation Arising From Product Market Competition

Case-Based Subgoaling in Real-Time Heuristic Search for Video Game Pathfinding.

Combined Task and Motion Planning for Mobile Manipulation

We present a hierarchical planning system and its application to robotic manipulation.  The novel features of the system are: 1) it finds high-quality kinematic solutions to task-level problems; 2) it takes advantage of subtask-specific irrelevance information, reusing optimal solutions to state-abstracted subproblems across the search space.  We briefly describe how the system handles uncertainty during plan execution, and present results on discrete problems as well as pick-and-place tasks for a mobile robot.

Using Closed Captions as Supervision for Video Activity Recognition

Recognizing activities in real-world videos is a difficult problem exacerbated by background clutter, changes in camera angle & zoom, and rapid camera movements. Large corpora of labeled videos can be used to train automated activity recognition systems, but this requires expensive human labor and time. This paper explores how closed captions that naturally accompany many videos can act as weak supervision that allows automatically collecting "labeled" data for activity recognition. We show that such an approach can improve activity retrieval in soccer videos. Our system requires no manual labeling of video clips and needs minimal human supervision. We also present a novel caption classifier that uses additional linguistic information to determine whether a specific comment refers to an ongoing activity. We demonstrate that combining linguistic analysis and automatically trained activity recognizers can significantly improve the precision of video retrieval.

Metrized Small World Approach for Nearest Neighbor Search

Predictive risk modelling for forecasting high-cost patients: a real-world application using Medicaid data

MEET DB2

Automatic rule refinement for information extraction

Proteomic analysis of peach fruit mesocarp softening and chilling injury using difference gel electrophoresis (DIGE)

Using After-Action Review Based on Automated Performance Assessment to Enhance Training Effectiveness

Training simulators have become increasingly popular tools for instructing humans on performance in complex environments. However, the question of how to provide individualized and scenario-specific assessment and feedback to students remains largely an open question. In this work, we follow-up on previous evaluations of the Automated Expert Modeling and Automated Student Evaluation (AEMASE) system, which automatically assesses student performance based on observed examples of good and bad performance in a given domain. The current study provides a rigorous empirical evaluation of the enhanced training effectiveness achievable with this technology. In particular, we found that students given feedback via the AEMASE-based debrief tool performed significantly better than students given only instructor feedback on two out of three domain-specific performance metrics.

Fred Jelinek

Model Independent Bounds on Kinetic Mixing

Nearly Supersymmetric Dark Atoms

A brief report on working smarter with Agile software development

Agile software development methods have proven to be a substantial catalyst for improvements in software quality, speed of delivery, and fitness for use, and are a driver behind the IBM Smarter Planet™ initiative, which involves intelligent approaches that address the needs of a world that is becoming increasingly interconnected and instrumented. Software vendors who use a traditional or waterfall approach (i.e., sequential software development process) for requirement analysis and development often find that they cannot react with sufficient speed to meet rapidly changing market demand. As a result, stakeholders must wait for the vendor to complete a lengthy development cycle, which, when complete, may miss the market window for the software or fail to adequately address other needs of these stakeholders. Agile development methods, as practiced across geographically diverse and broad development teams at IBM, directly address this issue. One purpose of this paper is simply to provide a brief high-level overview and record of the adoption of these practices and methods across a large company, i.e., the IBM Corporation.

Session details: Query analysis

A multi-resolution approach to learning with overlapping communities

Scalable crisis relief: Crowdsourced SMS translation and categorization with Mission 4636

The peer-to-peer trace archive

Social sensing: obesity, unhealthy eating and exercise in face-to-face networks

Reactive extensions (Rx): curing your asynchronous programming blues

Asynchronous, event-driven programming is way too hard in today's world of development tools and frameworks. The huge amount of manual and error-prone plumbing leads to incomprehensible and hard to maintain code. As we reach out to services in the cloud, the desire for asynchronous computation is ever increasing, requiring a fresh look on the problems imposed by reactive programming. Centered around the concept of observable data sources, Rx provides a framework that takes care of the hard parts of reactive programming. Instead of focusing on the hard parts, you now can start dreaming about the endless possibilities of composing queries over asynchronous data sources, piggybacking on convenient LINQ syntax. In this session, we'll cover the design philosophy of Rx, rooted on the deep duality between the interactive IEnumerable interface and the new reactive IObservable interface in .NET 4. From this core understanding, we'll start looking at various combinators and operators defined over observable collections, as provided by Rx, driving concepts home by a bunch of samples. Finally, if time permits, we'll look at the Reactive Extensions for JavaScript which allows us to take the concepts we already know from Rx and apply them to JavaScript and have deep integration with libraries such as jQuery. Democratizing asynchronous programming starts today. Don't miss out on it!

Scaling Scala at Twitter

Camlp4 and Template Haskell

Implementing web sites with Scala and Lift

Comparing XML processing performance in middleware and database: a case study

Generating document summaries from user annotations

Supervised identification and linking of concept mentions to a domain-specific ontology

Web search solved?

Building the right cloud solutions

Given that cloud computing is still one of, if not the most hyped buzzword in technology these days, we think it is important to bring the community of developers together to discuss and share the initial experiences in building the cloud solutions that makes the most sense in todays reality. According to NIST[2], the four deployment models of cloud computing are community clouds, private clouds, public clouds and hybrid clouds. The workshop aims to discuss and share experiences and state of the practice in deploying these cloud models.

Language virtualization for heterogeneous parallel computing

As heterogeneous parallel systems become dominant, application developers are being forced to turn to an incompatiblemix of low level programming models (e.g. OpenMP, MPI, CUDA, OpenCL). However, these models do little to shield developers from the difficult problems of parallelization, data decomposition and machine-specific details. Most programmersare having a difficult time using these programming models effectively. To provide a programming modelthat addresses the productivity and performance requirements for the average programmer, we explore a domainspecificapproach to heterogeneous parallel programming.

Type classes as objects and implicits

Type classes were originally developed in Haskell as a disciplined alternative to ad-hoc polymorphism. Type classes have been shown to provide a type-safe solution to important challenges in software engineering and programming languages such as, for example, retroactive extension of programs. They are also recognized as a good mechanism for concept-based generic programming and, more recently, have evolved into a mechanism for type-level computation.

Lightweight modular staging

Software engineering demands generality and abstraction, performance demands specialization and concretization. Generative programming can provide both, but the effort required to develop high-quality program generators likely offsets their benefits, even if a multi-stage programming language is used.We present lightweight modular staging, a library-based multi-stage programming approach that breaks with the tradition of syntactic quasi-quotation and instead uses only types to distinguish between binding times. Through extensive use of component technology, lightweight modular staging makes an optimizing compiler framework available at the library level, allowing programmers to tightly integrate domain-specific abstractions and optimizations into the generation process.We argue that lightweight modular staging enables a form of language virtualization, i.e. allows to go from a pure-library embedded language to one that is practically equivalent to a stand-alone implementation with only modest effort.

Fast sorting on flash memory sensor nodes.

Workshop report

Query intent prediction and recommendation