Mapping Stream Programs into the Compressed Domain Due to the high data rates involved in audio, video, and signal processing applications, it is imperative to compress the data to decrease the amount of storage used. Unfortunately, this implies that any program operating on the data needs to be wrapped by a decompression and re-compression stage. Re-compression can incur significant computational overhead, while decompression swamps the application with the original volume of data. In this paper, we present a program transformation that greatly accelerates the processing of compressible data. Given a program that operates on uncompressed data, we output an equivalent program that operates directly on the compressed format. Our transformation applies to stream programs, a restricted but useful class of applications with regular communication and computation patterns. Our formulation is based on LZ77, a lossless compression algorithm that is utilized by ZIP and fully encapsulates common formats such as Apple Animation, Microsoft RLE, and Targa. We implemented a simple subset of our techniques in the StreamIt compiler, which emits executable plugins for two popular video editing tools: MEncoder and Blender. For common operations such as color adjustment and video compositing, mapping into the compressed domain offers a speedup roughly proportional to the overall compression ratio. For our benchmark suite of 12 videos in Apple Animation format, speedups range from 1.1x to 471x, with a median of 15x. 1.

Collaborative knowledge management systems such as the Wikipedia are becoming ever more popular – and these systems typically contain hypertext links between documents. The Wikipedia offers both manual and automated link creation. In fact several different systems providing links for Wikipedia documents now exit. Problematically the quality of automatically generated links has never been quantified. An evaluation method for Wikipedia link discovery approaches is essential. We introduce the Link-the-Wiki task launched at INEX in 2007. 90 documents were orphaned from the collection and participants were required to build systems that identified the missing links. The different automated link discovery techniques used by participants are outlined. Details of two successful techniques are given, one using the titles of pre-existing documents to identify anchors and destinations, the other using pre-existing links between documents to identify possible links in new documents. In this paper, we mainly focus on the analysis and assessment of Wikipedia link discovery and discuss possible future evaluation techniques. We examine one system in further detail and conduct a scalability experiment in which 1 % of all Wikipedia documents were used and the performance studied in detail – link discovery in this system is shown to be scalable. Finally, potential research directions for link discovery, assessment and evaluation are discussed.

From hundreds to thousands of computers, hubs to switched networks, and Ethernet to either ATM or 10Gbps Ethernet, administrators need more sophisticated network traffic monitoring and analysis tools in order to deal with the increase. These tools are needed, not only to fix network problems on time, but also to prevent network failure, to detect inside and outside threats, and make good decisions for network planning. This paper surveys all possible network traffic monitoring and analysis tools in non-profit and commercial areas. The tools are categorized in three categories based on data acquisition methods: network traffic flow from NetFlow-like network devices and SNMP, and local traffic flow by packet sniffer. The popular tools for each category and their main features and operating system compatibilities are discussed. The feature comparisons on each category are also made. Keywords:

To improve on its supply chain management (SCM) one of US's largest chain of supermarkets, Wal-Mart, on June 11, 2003, announced that from January 2005 its top 100 suppliers are required to put radio frequency (RFID) tags on their cases and pallets. This goal seems to be achieved as all of the a#ected suppliers have announced they will be ready. Other companies monitor the situation closely, and due to the apparent success they are expected to follow Wal-Mart's example soon. Basically RFID consists of two devices: A chip, called a transponder or tag, and a device which reads the contents of the chip, referred to as a reader. A tag/reader pair does not have to be in physical contact to communicate, as this is done through air using radio waves. This means that communication can be performed even if the reader cannot see the transponder i.e. no lineof -sight between them. To even further improve on SCM and the handling of inventory inside stores, placing a tag on individual items is presently discussed. The flipside is that this will bring RFID out to the individual consumer, where it can be used to invade his privacy. Anyone with a scanner (which does not have to be stationary!) will now be able to trace him and know what is in his bags. To prevent this "Big Brother"-like scenario, di#erent solutions have been suggested. Some of these are based on encryption, which is the objective of this report. At present the main problem regarding encryption in RFID systems is not the strength of the algorithms, but due to constraints whether it is possible (and feasible). The constraints are that tags need to be small, and that only a limited supply of power is available to a tag. Besides these limits tags are not allowed to cost much either! In this report several encryption algori...

One of the main challenges in today’s world of vastly distributed sources of information is to re-combine information sources to provide uniform access. In this work, we propose a distributed data management system that should provide the “glue ” for combining data sources. This system advocates services as a means to access data. New services are defined on demand and the creation of services is supported by a behaviorist approach that incorporates new service ideas provided by the user. Services can be based on data and/or based on the output of existing services. To increase the usability of services in our system we utilize two ontologies to denote relevant metadata. Service ontology structures existing services and helps in discovering services. Parameter ontology structures the parameters used in services and supports the creation of new services. Our proposal of a services-based data management system exhibits similarities to the newsgroup approach in that both “systems ” examples of semantic search engines based on user interaction. By exploring these similarities and by looking at some statistics of newsgroup user/posting behavior, we validate our services-based approach. 1.

I hereby declare that I am the sole author of this thesis. This is a true copy of the thesis, including any required final revisions, as accepted by my examiners. I understand that my thesis may be made electronically available to the public. ii Research in patient education suggests that tailored educational materials can improve patient’s understanding of a treatment plan and help to achieve patient engagement and compliance. The goal of the HealthDoc Project has been the creation of automated Natu-ral Language Generation systems for producing educational materials that are tailored to an individual patient’s medical condition and personal situation. The project has so far focused on developing computational linguistic tools needed to author tailorable content from which customized versions could be generated. Also the HealthDoc model of docu-ment generation assumes the existence of previously authored textual material. Therefore, a new approach is needed to construct these materials and ensure that the relevant medical knowledge will be captured and delivered to the patient by providing a means to assist the

Wikipedia is a “free ” online encyclopedia. It contains millions of entries in many languages and is growing at a fast pace. Due to its volume, search engines play an important role in giving access to the information in Wikipedia. The “free ” availability of the collection makes it an attractive corpus for information retrieval experiments. In this paper we describe the evaluation of a search engine that provides focused search access to Wikipedia, i.e., a search engine which gives direct access to individual sections of Wikipedia pages. The main contributions of this paper are twofold. First, we introduce Wikipedia as a test corpus for information retrieval experiments in general and for semi-structured retrieval in particular. Second, we demonstrate that focused XML retrieval methods can be applied to a wider range of problems than searching scientific journals in XML format, including accessing reference works. 1.

Adequate representation of natural language semantics requires access to vast amounts of common sense and domain-specific world knowledge. Prior work in the field was either based on purely statistical techniques that did not make use of background knowledge or on huge manual efforts, such as the CYC projects. Here we propose a novel method, called Explicit Semantic Analysis (ESA), for finegrained semantic interpretation of unrestricted natural language texts. Our method represents meaning in a high-dimensional space of concepts derived from Wikipedia, the largest encyclopedia in existence. We use machine learning techniques that allow us to explicitly represent the meaning of any text in terms of Wikipedia-based concepts. We evaluate the effectiveness of our method on automatically computing the degree of semantic relatedness between fragments of natural language text. Compared with the previous state of the art, using ESA results in substantial improvements in correlation of computed relatedness scores with human judgments: from r = 0.56 to 0.75 for individual words and from r = 0.60 to 0.72 for texts. Consequently, we anticipate ESA to give rise to the next generation of natural language processing tools. Importantly, due to the use of natural concepts, the ESA model is easy to explain to human users. 1

This thesis focuses on the fields of real-time strategy games, evolutionary computation, distributed machine learning and multi-agent systems. In general, the problem is to automatically learn the best strategy to play a real time strategy game, more precisely-a two-player combat of marines and tanks. The idea was inspired by ORTS RTS Game AI Competition held annually at University of Alberta. The given problem is very complex and multicriterial, thus final solutions presented here are the result of a constant development and countless improvements. In the paper we try to underline the iterative nature of this process and propose a methodology that could be used for different problems in the real-time games field. We show how to model the strategy as a multi-agent system and how to fine-tune the evolutionary process of searching best players. We also explore the subject of distributed learning, focusing on using a computation cluster for evaluating solutions. The methods of evaluation are also elaborated in the context of co-evolution, we compare two different methods that use competitive fitness- single elimination tournament and hall of fame. In order to

I hereby declare that I am the sole author of this thesis. This is a true copy of the thesis, including any required final revisions, as accepted by my examiners. I understand that my thesis may be made electronically available to the public. ii The primary objective of this study was to quantify and characterise lipid deposition on soft (hydrogel) contact lenses, particularly those containing siloxane components. Studies involving a variety of in vitro doping and in vivo worn contact lenses were undertaken, in which lipid deposition was analyzed by either TLC or HPLC. Specific experiments were completed to optimize a method to extract the lipid from the lens materials, to compare the total lipid deposition on nine different hydrogel lenses and to analyze the effect that lipid deposition had on wettability. A method for extracting lipid from contact lenses using 2:1 chloroform: methanol was developed. This study also showed that siloxane-containing contact lens materials differ in the degree to which they deposit lipid, which is dependent upon their chemical composition. Small differences in lipid deposition that