Die vorliegende Arbeit beschäftigt sich mit Mechanismen, die in Einzelzellmodellen zu einer frequenzabhängigen Informationsübertragung führen können. Um dies zu untersuchen, werden Methoden aus der theoretischen Physik (Statistische Physik) und der Informationstheorie angewandt. Die Informationsfilterung in mehreren stochastischen Neuronmodellen, in denen unterschiedliche Mechanismen zur Informationsfilterung führen können, werden numerisch und, falls möglich, analytisch untersucht. Die Bandbreite der betrachteten Modelle erstreckt sich von reduzierten strombasierten ’Integrate-and-Fire’ (IF) Modellen bis zu biophysikalisch realistischeren leitfähigkeitsbasierten Modellen. Anhand numerischer Untersuchungen wird aufgezeigt, dass viele Varianten der IF-Neuronenmodelle vorzugsweise Information über langsame Anteile eines zeitabhängigen Eingangssignals übertragen. Der einfachste Vertreter der oben genannten Klasse der IF-Neuronmodelle wird dahingehend erweitert, dass ein Konzept von neuronalem ’Gedächtnis’, vermittelst positiver Korrelationen zwischen benachbarten Intervallen aufeinander- folgender Spikes, integriert wird. Dieses Model erlaubt eine analytische störungstheoretische Untersuchung der Auswirkungen positiver Korrelationen auf die Informationsfilterung. Um zu untersuchen, wie sich sogenannte ’unterschwelligen Resonanzen’ auf die Signalübertragung auswirken, werden Neuronenmodelle mit verschiedenen Nichtlinearitäten anhand numerischer Computersimulationen analysiert. Abschließend wird die Signalübertragung in einem neuronalen Kaskadensystem, bestehend aus linearen und nichtlinearen Elementen, betrachtet. Neuronale Nichtlinearitäten bewirken eine gegenläufige Abhängigkeit (engl. "trade-off") zwischen qualitativer, d.h. frequenzselektiver, und quantitativer Informations-übertragung, welche in allen von mir untersuchten Modellen diskutiert wird. Diese Arbeit hebt die Gewichtigkeit von Nichtlinearitäten in der neuronalen Informationsfilterung hervor. / Neurons transmit information about time-dependent input signals via highly non-linear responses, so-called action potentials or spikes. This type of information transmission can be frequency-dependent and allows for preferences for certain stimulus components. A single neuron can transmit either slow components (low pass filter), fast components (high pass filter), or intermediate components (band pass filter) of a time-dependent input signal. Using methods developed in theoretical physics (statistical physics) within the framework of information theory, in this thesis, cell-intrinsic mechanisms are being investigated that can lead to frequency selectivity on the level of information transmission. Various stochastic single neuron models are examined numerically and, if tractable analytically. Ranging from simple spiking models to complex conductance-based models with and without nonlinearities, these models include integrator as well as resonator dynamics. First, spectral information filtering characteristics of different types of stochastic current-based integrator neuron models are being studied. Subsequently, the simple deterministic PIF model is being extended with a stochastic spiking rule, leading to positive correlations between successive interspike intervals (ISIs). Thereafter, models are being examined which show subthreshold resonances (so-called resonator models) and their effects on the spectral information filtering characteristics are being investigated. Finally, the spectral information filtering properties of stochastic linearnonlinear cascade neuron models are being researched by employing different static nonlinearities (SNLs). The trade-off between frequency-dependent signal transmission and the total amount of transmitted information will be demonstrated in all models and constitutes a direct consequence of the nonlinear formulation of the models.

Made available in DSpace on 2016-08-17T14:53:01Z (GMT). No. of bitstreams: 1 Roberval Gomes Mariano.pdf: 3806410 bytes, checksum: 98c37c22e17816b87c3a646527ac2c4e (MD5) Previous issue date: 2008-12-05 / The huge amount of data available on the Web and its dynamic nature create a demand for information filtering applications such as recommender systems. The lack of semantic structure of data available on the Web constitutes a barrier for increasing the effectiveness of such applications family. This work discusses the analysis, design, implementation and evaluation of Semantic Web based hybrid filtering agents. Such agents were integrated in ONTOSERS, an application family for the development of recommender systems based on the Semantic Web technology. The implemented agents were evaluated and their results were compared with the results of collaborative and content-based filtering agents. The hybrid filtering techniques presented better results than the other approaches in the conducted experiments. The tested hybrid filtering approaches were the weighted and switched ones. The explicit feedback was used to validate the recommendations, presenting a better correlation with the hybrid filtering techniques. The developed agents were also evaluated through the reuse of the ONTOSERS systems family, a multi-agent recommender system in the Brazilian tributary domain. / A grande quantidade de dados disponíveis na Web e a sua natureza dinâmica criam uma demanda por aplicações de filtragem de informação, tais como os sistemas de recomendação. A falta de estruturação semântica dos dados disponíveis na Web é uma barreira para a melhoria da efetividade desta família de aplicações. Este trabalho apresenta a análise, projeto, implementação e avaliação de agentes de filtragem híbrida baseados na tecnologia da Web Semântica. Estes agentes foram integrados na ONTOSERS, uma família de aplicações para o desenvolvimento de sistemas de recomendações baseados na tecnologia da Web Semântica. Os agentes implementados foram testados e tiveram seus resultados comparados com os resultados de agentes utilizando filtragem colaborativa e baseada em conteúdo. As técnicas de filtragem híbrida apresentaram resultados melhores do que os obtidos com as outras técnicas nos experimentos realizados. As técnicas de filtragem híbrida testadas foram a ponderada e a alternada. O feedback explícito foi utilizado para validar as recomendações, apresentando uma melhor correlação com as técnicas de filtragem híbrida. Os agentes desenvolvidos foram ainda avaliados através do reuso da família de sistemas ONTOSERS na construção do INFOTRIB, um sistema multiagente de recomendações no domínio tributário brasileiro.

Information Filtering and Recommender Systems have been used and has been implemented in various ways from various entities since the dawn of the Internet, and state-of-the-art approaches rely on Machine Learning and Deep Learning in order to create accurate and personalized recommendations for users in a given context. These models require big amounts of data with a variety of features such as time, location and user data in order to find correlations and patterns that other classical models such as matrix factorization and collaborative filtering cannot. This thesis researches, implements and compares a variety of models with the primary focus of Machine Learning and Deep Learning for the task of music recommendation and do so successfully by representing the task of recommendation as a multi-class extreme classification task with 100 000 distinct labels. By comparing fourteen different experiments, all implemented models successfully learn features such as time, location, user features and previous listening history in order to create context-aware personalized music predictions, and solves the cold start problem by using user demographic information, where the best model being capable of capturing the intended label in its top 100 list of recommended items for more than 1/3 of the unseen data in an offine evaluation, when evaluating on randomly selected examples from the unseen following week. / Informationsfiltrering och rekommendationssystem har använts och implementeratspå flera olika sätt från olika enheter sedan gryningen avInternet, och moderna tillvägagångssätt beror påMaskininlärrning samtDjupinlärningför att kunna skapa precisa och personliga rekommendationerför användare i en given kontext. Dessa modeller kräver data i storamängder med en varians av kännetecken såsom tid, plats och användardataför att kunna hitta korrelationer samt mönster som klassiska modellersåsom matris faktorisering samt samverkande filtrering inte kan. Dettaexamensarbete forskar, implementerar och jämför en mängd av modellermed fokus påMaskininlärning samt Djupinlärning för musikrekommendationoch gör det med succé genom att representera rekommendationsproblemetsom ett extremt multi-klass klassifikationsproblem med 100000 unika klasser att välja utav. Genom att jämföra fjorton olika experiment,så lär alla modeller sig kännetäcken såsomtid, plats, användarkänneteckenoch lyssningshistorik för att kunna skapa kontextberoendepersonaliserade musikprediktioner, och löser kallstartsproblemet genomanvändning av användares demografiska kännetäcken, där den bästa modellenklarar av att fånga målklassen i sin rekommendationslista medlängd 100 för mer än 1/3 av det osedda datat under en offline evaluering,när slumpmässigt valda exempel från den osedda kommande veckanevalueras.

Diese Arbeit befasst sich mit Internetzensur innnerhalb der EU, und hier insbesondere mit der technischen Umsetzung, das heißt mit den angewandten Sperrmethoden und Filterinfrastrukturen, in verschiedenen EU-Ländern. Neben einer Darstellung einiger Methoden und Infrastrukturen wird deren Nutzung zur Informationskontrolle und die Sperrung des Zugangs zu Websites und anderen im Internet verfügbaren Netzdiensten untersucht. Die Arbeit ist in drei Teile gegliedert. Zunächst werden Fälle von Internetzensur in verschiedenen EU-Ländern untersucht, insbesondere in Griechenland, Zypern und Spanien. Anschließend wird eine neue Testmethodik zur Ermittlung der Zensur mittels einiger Anwendungen, welche in mobilen Stores erhältlich sind, vorgestellt. Darüber hinaus werden alle 27 EU-Länder anhand historischer Netzwerkmessungen, die von freiwilligen Nutzern von OONI aus der ganzen Welt gesammelt wurden, öffentlich zugänglichen Blocklisten der EU-Mitgliedstaaten und Berichten von Netzwerkregulierungsbehörden im jeweiligen Land analysiert. / This is a thesis on Internet censorship in the European Union (EU), specifically regarding the technical implementation of blocking methodologies and filtering infrastructure in various EU countries. The analysis examines the use of this infrastructure for information controls and the blocking of access to websites and other network services available on the Internet. The thesis follows a three-part structure. Firstly, it examines the cases of Internet censorship in various EU countries, specifically Greece, Cyprus, and Spain. Subsequently, this paper presents a new testing methodology for determining censorship of mobile store applications. Additionally, it analyzes all 27 EU countries using historical network measurements collected by Open Observatory of Network Interference (OONI) volunteers from around the world, publicly available blocklists used by EU member states, and reports issued by network regulators in each country.