This thesis presents a further development of Neuroevolution of Augmenting topologies(NEAT)[21]. The author augments NEAT by parallelizing the fitness evaluation of the phenotypes enabling the method to be utilized on highly complex fitness evaluations by running it on a cluster. This augmented version of NEAT is then applied to the inherently complex problem of the Go board game, by using the Gnugo (See www.gnu.org/software/gnugo/.) software package as a fitness evaluator. The performance increase also enables the author to follow up on the predictions of Kenneth Stanley’s previous discussions that co-evolution will help evolve a more general Go player, rather than the predicted evolved behaviour of specializing in beating Gnugo.

Robotic mapping systems are traditionally separated into the metric and the topological paradigm. The metric approach provides geometrical accuracy, but is fragile because it is bounded in an absolute coordinate system and depends on the use of odometry for navigation. The topological paradigm provides a compact presentation and navigation free of accumulated error. In this thesis the topological and the metric paradigm is combined into a hybrid representation where a topological map joins a set of local maps. Each local map contains a pair of self-organizing maps, one that maps the metric space of the local map, and one that maps the perceptual space of the local map. Local navigation is performed over the SOM mapping the metric space and position correction is performed over the SOM mapping the perceptual space. A boundary-tracing behavior is used for navigation within the global map, while the local metric maps allow for more precise navigation and is navigated by performing path-integration.

Conventional artificial neural networks have traditionally faced inherent problems with efficient parallelization of neuron processing. Recent research has shown how artificial spiking neural networks can, with the introduction of biologically plausible synaptic conduction delays, be fully parallelized regardless of their network topology. This, in conjunction with the influx of fast, massively parallel desktop-level computing hardware leaves the field of efficient, large-scale spiking neural network simulations potentially open to even those with no access to supercomputers or large computing clusters. This thesis aims to show how such a parallelization is possible as well as present a network model that enables it. This model will then be used as a base for implementing a parallel artificial spiking neural network on both the CPU and the GPU and subsequently evaluating some of the challenges involved, the performance and scalability measured and the potential that is exhibited.

This master's thesis has investigated how methods from artificial intelligence (AI) can be used to perform and augment production optimization of sub-sea oil reservoirs. The methods involved in this work are genetic algorithms (GAs) and artificial neural networks (ANNs). Different optimization schemes were developed by the author to perform production optimization on oil reservoir simulator models. The optimization involves finding good input parameter values for certain properties of the model, relating to how the wells in the oil reservoir operate. The research involves straightforward optimization using GAs, model approximations using ANNs, and also more advanced schemes using these methods together with other available technology to perform and augment reservoir optimization. With this work, the author has attempted to make a genuine contribution to all the research areas this master's thesis has touched upon, ranging from computer science and AI to process and petroleum engineering. The methods and approaches developed through this research were compared to the performance of each other and also to other approaches and methods used on the same challenges. The comparison found some of the developed optimization schemes to be very successful, while others were found to be less appropriate for solving the problem at hand. Some of the less successful approaches still showed considerable promise for simpler problems, leading the author to conclude that the developed schemes are suited for solving optimization problems in the petroleum industry.

This work studies constraint mechanisms in frame-based knowledge representation systems with the aim of improving the knowledge modelling abilities of the TrollCreek system. TrollCreek is an implementation of Creek, an architecture for case based reasoning (CBR) that uses an explicit frame-based knowledge model to guide the CBR process. The objective of this project is to develop a constraint mechanism for TrollCreek. In doing this the earlier Lisp implementation of Creek and four other frame-based systems are examined with emphasize on their constraint mechanisms. Based on these systems a constraint mechanism for TrollCreek is discussed and specified. The part of the mechanism considered most central is implemented and evaluated.

In testing the viability of automatic knowledge-acquisition, using simple techniques and brute force on the Internet, a system was implemented in Java. Techniques from both multi-agent system and swarm intelligence paradigms were used to structure the system, improve searches, increase stability and increase modularity. The system presented relies on using existing search-engines to find texts on the World Wide Web, containing a user-specified key-word. Knowledge is identified in the texts using key-sentences, terms related to the key-word becomes new key-words in an incremental search. The result is expressed as sentences in a KR-language. The answers from a run were often interesting and surprising, and gave information beyond an encyclopedic scope, even if the answers often contained false information. The results of the implemented system verified the viability of both the designed framework and the theory behind it.

The work in this thesis is about natural language processing and understanding, within the context of artificial intelligence research. What was attempted to achieve here was to investigate how meaning is contained in language, particularly with respect to how that information is encoded and how it can be decoded, or extracted. The aspects deemed most relevant for this quest was automated processing of the syntactic structure of sentences, and their semantic components. Artificial neural networks was chosen as the tool to perform the research with, and as such part of the goal became research on connectionist methods. A side-goal of interest was to look into the possibility of using insight into neural networks to gain deeper understanding of how the human brain processes information, particularly language. This area was not explicitly focussed on during the research. The methodology selected for achieving the goals was to design and implement a framework for developing neural network models, and further to implement NLP and NLU systems within this framework. The systems selected to explore and implement were: a parser for handling the syntactic structure and a symbol grounding system for dealing with the semantic component. A third system was also implemented for investigation into an evolutionary-based communication model on the development of a shared vocabulary between autonomous agents. All implementations were based on recent research and results by others.

The emergence of complex data objects that must to be indexed and accessed in databases has created a need for access methods that are both dynamic and efficient. Lately, metric tree structures have become a popular way of handling this because of the advantages they have compared to traditional methods based on spatial indexing. The most common way to handle indexing is to build tree structures and then prune out branches of the trees during search, and for a dynamic indexing structure it is important that these trees stay balanced in order to keep the worst case search time as low as possible. Normally, this is done based on complex criteria and reshuffling operations. Another way to handle balancing is General Balanced Trees (GBT), proposed by Arne Andersson (Journal of Algorithms 30, 1999), which uses simple, global criteria for rebalancing binary search trees by using total and partial rebuilding. This thesis explores if it is possible to apply this to metric tree structures, and especially two static metric tree structures called the Vantage Point Tree and the Multiple Vantage Point Tree. It discusses how to best make these into dynamic tree structures and how to apply balancing by using GBT paradigms on them. The results of the performance of the new tree structures are analyzed, and the results are compared against already existing structures. The results shows that this works for balancing the trees, and that the structures perform reasonably well compared to already existing structures.

Hovedfokus i denne hovedfagsoppgaven er å gi en oversiktlig og lettfattelig innføring i data og kommunikasjonsikkerhet. Det er lagt vekt på å holde oppgaven på et såpass lavt teknisk og matematisk nivå at den vil være mulig å forstå for personer uten store datakunnskaper. Oppgaven er organisert på den måten at den starter med en grunnleggende innføring i kryptografi. Dette er nødvendig for å forstå resten av oppgaven. Deretter forklares og vises de mest utbredte sikkerhetsmekanismene innen data og kommunikasjonsikkerhet. Etter det tar jeg for meg en del spesifikke applikasjoner som brukes mye, og hva som kan gjøres for å sikre kommunikasjonen ved å ta i bruk disse. Til slutt i den teoretiske delen går jeg gjennom andre kommunikasjonsmetoder enn ethernet, som datamaskiner som oftest bruker, og hvilke sikkerhetsmekanismer som ligger innebygd i disse samt hva vi som brukere kan gjøre for å øke sikkerheten.For å gjøre sikkerhet mer tilgjengelig for vanlige brukere har jeg i den praktiske delen av oppgaven valgt å gi detaljerte beskrivelser på oppsett av tre vanlige sikkerhetsmekanismer. De tre jeg har valgt å forklare er IPSec, SSL og digitale signaturer. Grunnen til at valget falt på disse er at de på en måte representerer hvert sitt aktuelle felt innen sikkerhet. IPSec benyttes når en vil ha total sikkerhet i kommunikasjonen mellom maskiner, SSL benyttes når en bare vil ha sikkerhet på informasjonsutveksling mellom utvalgte applikasjoner og digitale signaturer tar for seg integriteten til informasjon som sendes over usikre nettverk. Figurene og beskrivelsene vil gjøre de som leser denne oppgaven i stand til å sette dette opp selv på sine egne datamaskiner.

Målet med denne oppgaven er å se om en automatisk kan organisere kunnskapsobjekter tilrettelagt for et tredimensjonalt navigerbart konseptkart. Forutsetningen for å gjøre dette mulig er at kunnskapen er fremstilt i enheter, kalt læringsobjekter. Et læringsobjekt består av en eller flere kunnskapsobjekter knyttet sammen med ett eller flere læringsmål, og ulike læringsaktiviteter. Et kunnskapsobjekt er bygd opp av en eller flere ressurser. Disse ressursene kan være tekst, lyd, bilde, video osv. En ser for seg at kunnskapsobjektene befinner seg i samlinger, repositories. Kunnskapsobjektene skal kunne utvikles av både faglig veileder og lærende, men hovedsakelig tenker en at de skal produseres av profesjonelle aktører. Her er det snakk om en produksjonslinje. Den profesjonelle aktøren står for produksjon og presentasjon av objektene, mens faglig veileder sørger for innhold og pedagogisk tilrettelegging. Til hvert kunnskapsobjekt er det knyttet en tekstlig beskrivelse som danner grunnlag for vektorisering. Det er denne vektoriseringen som gjør det mulig å indeksere kunnskapsobjektene, samt sammenligne dem. Sammenligningen av de ulike objektene danner grunnlag for klyngeanalyse. Klyngeanalysen vil organisere like objekter i grupper slik at hver gruppe deler felles egenskaper, dvs. har en stor grad av likhet. Med denne grupperingen kan man representere objektene visuelt, for eksempel i en tredimensjonal graf. Som en del av denne oppgaven er det utviklet to prototyper for klyngeanalyse. I den ene benyttes det K-means-algoritme, mens i den andre benyttes en agglomerativ hierarkisk algoritme. Dataen fra disse analysene er tenkt brukt i en tredimensjonal representasjon. Prototypen for K-means har en todimensjonal framstilling slik at man får et visst innblikk i hva en visuell framstilling av objektene innebærer. En vil blant annet se av testene at K-means med brukervalgte initielle klyngerepresentanter gir det beste resultatet for klyngeanalysen, og at de tekstlige beskrivelsene av kunnskapsobjektene bør være av størrelse som en vanlig A4-side.