In this presentation, Dr. Henry Tirri, EVP & CTO of Nokia and UC Berkeley will discuss insights into the social and scientific implications of the technology trends being driven by the rise of large scale multi-device cloud-based computing. The cloud has already fundamentally transformed everyday experiences visible to consumers through their access to streaming media, social networking and location services via multiple different computing devices, from phones and tablets to connected accessories. As bits continue to eat atoms, more elements of the physical world will turn first into code and then into code that lives in the cloud. This is driven by the enormous advantages in sharing, indexing and elasticity of computing code has with respect to physical objects. We have seen this happening to photos and videos, music and books, but this is only the beginning. Consumers will increasingly turn to connected experiences that will in turn produce a wealth of data at a at rate unprecedented in the history of mankind. Due to the pervasiveness of this change, the next phase of the cloud era will see increased partnerships between public and private sectors around long-term technology trends in areas ranging from urban planning to health care. It can be anticipated that connected hardware will continue to diversify, with an increased emphasis on "multi-sensing". Wearables, sensor clusters in vehicles and smart devices, and independent sensors will all become first class citizens of the cloud: both feeding data into it and drawing it back out again. We will look at one exciting example of this trend -- indexing of the real world made possible by global scale location services. We argue that ultimately computing will totally de-centralize and live throughout a heterogeneous cloud-based architecture, but energy will continue to be the "One Ring to Rule Them All," that will define where in the cloud the execution happens.

Personal identification and using that information is in the heart of many ubiquitous systems. We present two complementary techniques, namely personal identification without directly observing the subject, and using that information for understanding the social relations between the subjects. We show that with certain presumptions it is possible to predict one's identity with reasonable certainty only by observing one's Bluetooth neighborhood without the need to directly observe the subject. We also show how this information can be used for exploring the social relations between the subjects

We analyze classification error on unseen cases, i.e. cases that are different from those in the training set. Unlike standard generalization error, this off-training-set error may differ significantly from the empirical error with high probability even with large sample sizes. We derive a datadependent bound on the difference between off-training-set and standard generalization error. Our result is based on a new bound on the missing mass, which for small samples is stronger than existing bounds based on Good-Turing estimators. As we demonstrate on UCI data-sets, our bound gives nontrivial generalization guarantees in many practical cases. In light of these results, we show that certain claims made in the No Free Lunch literature are overly pessimistic.

In information retrieval relevance ranking of the results is one of the most important single tasks there are. There are many diffierent ranking algorithms based on the content of the documents or on some external properties e.g. link structure of html documents.We present a temporally adaptive content-based relevance ranking algorithm that explicitly takes into account the temporal behavior of the underlying statistical properties of the documents in the form of a statistical topic model. more we state that our algorithm can be used on top of any ranking algorithm.

No abstract available.

We profile a system for search and analysis of large-scale email archives. The system builds around four facets: content-based search engine, statistical topic model, automatically inferred social networks, and time-series analysis. The facets correspond to the types of information available in email data. The presented system allows chaining or combining the facets flexibly. Results of one facet may be used as input to another yielding remarkable combinatorial power. In information retrieval point of view, the system provides support for exploration, approximate textual searches and data visualization. We present some experimental results based on a large real-world email corpus.

It is often a case that technologically-oriented research on e-learning stresses the tools and individual features used in e-learning platforms rather than the pedagogical model and the underlying course structures. This study compares the outcomes from a similar course in a similar setting using very different learning platforms designed by the same research group. The tool used in the first course, EDUCO, offers awareness of other learners by real-time social navigation features. The tool used in the second course, EDUCOSM, relies on easy-to-make joint asynchronous annotations on documents. The pedagogical model for the courses was the same: student-centered learning in self-organizing and self-evolving groups using peer support to tackle open-ended large problems. The results suggest that the learning outcomes can vary, especially in a case where the environment is open and transparent in a sense that it enables learners to easily rely on and help each other in peer-to-peer fashion.

In this paper we analyse data from a self-rated questionnaire together with statistics of the learners' real activity in a collaborative learning environment called EDUCOSM. Besides being of theoretical interest, the manifestation of self-perceptions in actual behavior is important for the development of better support facilities. For example, correct interpretation of both questionnaire and log data is likely to be necessary for partially automated group formation. This paper focuses on motivation and social abilities. The results suggest that group formation carried out without any supporting information from the system does not always lead to good results.

In this survey-style paper we demonstrate the usefulness of the probabilistic modelling framework in solving not only the actual positioning problem, but also many related problems involving issues like calibration, active learning, error estimation and tracking with history. We also point out some interesting links between positioning research done in the area of robotics and in the area of wireless radio networks.

Site-based or topic-specific search engines work with mixed success because of the general difficulty of the information retrieval task, and the lack of good link information to allow authorities to be identified. We are advocating an open source approach to the problem due to its scope and need for software components. We have adopted a topic-based search engine because it represents the next generation of capability. This paper outlines our scalable system for site-based or topic-specific search, and demonstrates the developing system on a small 250,000 document collection of EU and UN web pages.

Informal chat-room conversations have intrinsically different properties from regular static document collections. Noise, concise expressions and dynamic, changing and interleaving nature of discussions make chat data ill-suited for analysis with an off-the-shelf text mining method. On the other hand, interactive human communication has some implicit features which may be used to enhance the results. In our research we infer social network structures from the chat data by using a few basic heuristics. We then present some preliminary results showing that the inferred social graph may be used to enhance topic identification of a chat room when combined with a state-of-the-art topic and classification models. For validation purposes we then compare the performance effects of using this social information in a topic classification task.

This study presents a shared document-based annotation tool, EDUCOSM. Usefulness of the system is empirically evaluated in a real-life collaborative learning context. Relationships between learner's self-rated use of learning strategies, cognitive outcomes, and completion of various tasks in the system are investigated. An empirical study (n=31) was conducted in order to investigate various dependencies between variables from precourse self-rated questionnaire, system log file data collected during the course and postcourse e-mail survey.

In spite of advances in educational technology, most Web-based computer science courses rely on costly pre-made learning materials. By shifting the emphasis to peer-to-peer learning and other student-centred learning principles, more meaningful learning process can be accomplished, without preparing stand-alone Web-courses. A course applying these principles is presented along with a tool built for the purpose. The tool offers the possibility to build a joint information pool and publish new work while constructing knowledge by collaborative annotation of the information or published work. According to our experince, transparency in the learning process is also well-accepted and viewed beneficial by the students.

Student-centred learning can be used in Web-courses to increase student activity, motivation and commitment. EDUCO is a system for student-centred learning, both for the learners and the teachers. Students can use EDUCO within a standard Web-browser to navigate towards useful information and Web-resources gathered into the system. The key issue is that every participant can see everyone else in the system and their navigational steps, so that the feeling of student companions taking part in the same tasks is increased. The implications of this type of social navigation are discussed along with the description of the system itself.

Some location estimation methods, such as the GPS satellite navigation system, require nonstandard features either in the mobile terminal or the network. Solutions based on generic technologies not intended for location estimation purposes, such as the cell-ID method in GSM/GPRS cellular networks, are usually problematic due to their inadequate location estimation accuracy. In order to enable accurate location estimation when only inaccurate measurements are available, we present an approach to location estimation that is different from the prevailing geometric one. We call our approach the statistical modeling approach. As an example application of the proposed statistical modeling framework, we present a location estimation method based on a statistical signal power model. We also present encouraging empirical results from simulated experiments supported by real-world field tests.

B-Course is a free Web based Bayesian data mining service. This service allows the users to analyze their own data for multivariate probabilistic dependencies represented as Bayesian network models. In addition to this, B-Course also offers facilities for inferring certain types of causal dependencies from the data. The software is especially suitable for educational purposes as the tutorial style user friendly interface intertwines steps in the data analysis with support material that gives an informal introduction to the Bayesian approach adopted. Nevertheless, although the analysis methods, modeling assumptions and restrictions are totally transparent to the user, this transparency is not achieved at the expense of analysis power: with the restrictions stated in the support material, B-Course is a powerful analysis tool exploiting several theoretically elaborate results developed recently in the fields of Bayesian and causal modeling.

B-Course is a free Web-based online data analysis tool, which allows users to analyze their data for multivariate probabilistic dependencies. These dependencies are represented as Bayesian network models. In addition to this, B-Course also offers facilities for inferring certain type of causal dependencies from the data. The software uses a novel "tutorial style" user-friendly interface which intertwines the steps in the data analysis with support material that gives an informal introduction to the Bayesian approach adopted. Although the analysis methods, modeling assumptions and restrictions are totally transparent to the user, this transparency is not achieved at the expense of analysis power: with the restrictions stated in the support material, B-Course is a powerful analysis tool exploiting several theoretically elaborate results developed recently in the fields of Bayesian and causal modeling. B-Course can be used with most Web-browsers (even Lynx), and the facilities include features such as automatic missing data handling and discretization, a flexible graphical interface for probabilistic inference on the constructed Bayesian network models (for Java enabled browsers), automatic pretty-printed layout for the networks, exportation of the models, and analysis of the importance of the derived dependencies. We discuss both the theoretical design principles underlying the B-Course tool, and the pragmatic methods adopted in the implementation of the software.

We propose a data reduction method based on a probabilistic similarity framework where tw o vectors are considered similar if they lead to similar predictions.We sho w ho w this type of a probabilistic similarity metric can be de ned both in a supervised and unsupervised manner.As a concrete application of the suggested multidimensional scaling scheme, we describe how the method can be used for producing visual images of high-dimensional data, and give several examples of visualizations obtained by using the suggested scheme with probabilistic Bayesian net w ork models.

In this paper we present a probabilistic formalization of the instance-based learning approach. In our Bayesian framework, moving from the construction of an explicit hypothesis to a data-driven instancebased learning approach, is equivalent to averaging over all the (possibly infinitely many) individual models. The general Bayesian instance-based learning framework described in this paper can be applied with any set of assumptions defining a parametric model family, and to any discrete prediction task where the number of simultaneously predicted attributes is small, which includes for example all classification tasks prevalent in the machine learning literature. To illustrate the use of the suggested general framework in practice, we show how the approach can be implemented in the special case with the strong independence assumptions underlying the so called Naive Bayes classifier. The resulting Bayesian instance-based classifier is validated empirically with public domain data sets and the results are compared to the performance of the traditional Naive Bayes classifier. The results suggest that the Bayesian instancebased learning approach yields better results than the traditional Naive Bayes classifier, especially in cases where the amount of the training data is small. These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

ALCHEMIST is a general purpose transformation generating environment, which supports specification, generation and execution of data transformations. ALCHEMIST allows an abstract specification of the transformation through a window-based interface and supports the generation and compilation of transformation program code from these specifications. Unlike compiler-compilers, ALCHEMIST is intended to automate building transformations between two complex representation formats and is thus especially suitable for constructing transformations between database tools, CASE tools, graphical editors or text formatters. In this paper we describe the design principles and the structure of ALCHEMIST, and demonstrate its use. We also discuss our experiences with several example transformations and present a real-life case study of using ALCHEMIST for interfacing two software development environments.

A scheme is presented for translating high-level descriptions of conceptual hierarchies into a neural network representation. The intuitive semantics of a conceptual hierarchy is provided by a Bayesian net, and the neural network implementation provably approximates the behavior of this net under a stochastic simulation rule

The computer-vision problem of determining object orientation from the consensus of orientations of individual symbols or marks is examined. The problem arises in automatic inspection where orientation can be detected from printed text but there is no knowledge of the content of the text. This is a high-dimensional classification problem, and there is a requirement for highly accurate detection and rapid processing. The typical multilayer threshold networks are seen as unsuitable, and the optimal Bayesian detector is derived and found to have the highly parallel structure of a feedforward network. The learning vector quantization neural network method of T. Kohonen (1988) is also applied. Experimental results, comparisons, and a complete implementation are described.>

A transaction model for relational databases is presented. The model is more realistic than the traditional entity based model, since it takes into consideration the special information available in the relational model and in the transactions of a given transaction set. Therefore the model can serve as the basis of an implementation, unlike in the present situation, where implementations must rely on ad hoc techniques or physical level concepts. A concurrency control method is developed based on the transaction model. The method is reminiscent of precision locking, but again it is tailored for the relational model. An efficient implementation of the method is outlined.

Several online schedulers of transactions are considered. The scheduling algorithms are presented in a modular and uniform way. Deadlock and livelock prevention (using restarts, if necessary) is integrated into the algorithms. A metrics is defined for comparing the behavior of two schedulers on the same set of transactions. The measure that is used makes only a very simple assumption about the transactions. This assumption is easily motivated, and thus the measure is a useful approximation of measures appearing in reality. The behavior of the schedulers is compared on the basis of user delays caused by one scheduler on logs accepted as such by another scheduler. No order of magnitude differences are found, but the examples used for proving lower bounds for the delays are illustrative of the characteristics of the schedulers.

We examine the problem of determining whether a given set of locked transactions, accessing a distributed database, is free from deadlock. A deadlock graph is used to derive a new characterization for deadlock-free two-transaction systems in a distributed environment. The characterization provides a direct and efficient polynomial test for deadlock-freedom in two-transaction systems. The method is not dependent on the number of sites in a distributed database, and hence improves previously known results, which are exponential in the number of sites.