Neuromorphic Engineering (NE) has led to the development of biologically-inspired computer architectures whose long-term goal is to approach the performance of the human brain in terms of energy efficiency and cognitive capabilities. Although there are a number of neuromorphic platforms available for large-scale Spiking Neural Network (SNN) simulations, the problem of programming these brain-like machines to be competent in cognitive applications still remains unsolved. On the other hand, Deep Learning has emerged in Artificial Neural Network (ANN) research to dominate state-of-the-art solutions for cognitive tasks. Thus the main research problem emerges of understanding how to operate and train biologically-plausible SNNs to close the gap in cognitive capabilities between SNNs and ANNs. SNNs can be trained by first training an equivalent ANN and then transferring the tuned weights to the SNN. This method is called ‘off-line’ training, since it does not take place on an SNN directly, but rather on an ANN instead. However, previous work on such off-line training methods has struggled in terms of poor modelling accuracy of the spiking neurons and high computational complexity. In this thesis we propose a simple and novel activation function, Noisy Softplus (NSP), to closely model the response firing activity of biologically-plausible spiking neurons, and introduce a generalised off-line training method using the Parametric Activation Function (PAF) to map the abstract numerical values of the ANN to concrete physical units, such as current and firing rate in the SNN. Based on this generalised training method and its fine tuning, we achieve the state-of-the-art accuracy on the MNIST classification task using spiking neurons, 99.07%, on a deep spiking convolutional neural network (ConvNet). We then take a step forward to ‘on-line’ training methods, where Deep Learning modules are trained purely on SNNs in an event-driven manner. Existing work has failed to provide SNNs with recognition accuracy equivalent to ANNs due to the lack of mathematical analysis. Thus we propose a formalised Spike-based Rate Multiplication (SRM) method which transforms the product of firing rates to the number of coincident spikes of a pair of rate-coded spike trains. Moreover, these coincident spikes can be captured by the Spike-Time-Dependent Plasticity (STDP) rule to update the weights between the neurons in an on-line, event-based, and biologically-plausible manner. Furthermore, we put forward solutions to reduce correlations between spike trains; thereby addressing the result of performance drop in on-line SNN training. The promising results of spiking Autoencoders (AEs) and Restricted Boltzmann Machines (SRBMs) exhibit equivalent, sometimes even superior, classification and reconstruction capabilities compared to their non-spiking counterparts. To provide meaningful comparisons between these proposed SNN models and other existing methods within this rapidly advancing field of NE, we propose a large dataset of spike-based visual stimuli and a corresponding evaluation methodology to estimate the overall performance of SNN models and their hardware implementations.

Submitted by Luciana Ferreira (lucgeral@gmail.com) on 2014-10-09T12:30:33Z No. of bitstreams: 2 Dissertação - Savio Salvarino Teles de Oliveira - 2013.pdf: 6348358 bytes, checksum: 12e62cd925367772158d94e466de5827 (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) / Approved for entry into archive by Luciana Ferreira (lucgeral@gmail.com) on 2014-10-09T14:44:35Z (GMT) No. of bitstreams: 2 Dissertação - Savio Salvarino Teles de Oliveira - 2013.pdf: 6348358 bytes, checksum: 12e62cd925367772158d94e466de5827 (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) / Made available in DSpace on 2014-10-09T14:44:35Z (GMT). No. of bitstreams: 2 Dissertação - Savio Salvarino Teles de Oliveira - 2013.pdf: 6348358 bytes, checksum: 12e62cd925367772158d94e466de5827 (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) Previous issue date: 2013-06-28 / Fundação de Apoio à Pesquisa - FUNAPE / Geographic Information Systems (GIS) have received increasing attention in research institutes and industry in recent years. A Spatial Database Managament System (SDBMS) is one of the main components of a GIS and spatial join is one of the most important operations in SDBMS. Spatial join involves the relationship between two datasets, combining the geometries according some spatial predicate, such as intersection. Due to the increasing availability of spatial data, the growing number of GIS users, and the high cost of the processing of spatial operations, distributed SGBDEs (SGBDED) have been proposed as a good option to efficiently process spatial join on a cluster. This distributed processing brings some challenges, such as the data distribution and parallel and distributed processing of spatial join. This paper presents a platform for parallel and distributed processing of spatial joins in a cluster using data distribution techniques for dynamic datasets. Studies in the literature have explored data distribution techniques for static datasets, where any update requires data redistribution. This becomes unfeasible when using large datasets with frequent updates. Therefore, this paper proposes two new data distribution techniques for dynamic datasets: Proximity Area and Grid Proximity Area. These techniques have been evaluated to determine which scenarios each technique is more appropriate for. For this purpose, these techniques are evaluated in a real environment using datasets with different characteristics. Therefore, it is possible to evaluate the spatial join operation in real scenarios with each technique. / Os Sistemas de Informação Geográfica (SIG) têm recebido cada vez mais destaque nos institutos de pesquisa e na indústria nos últimos anos. Um Sistema de Gerência de Bancos de Dados Espaciais (SGBDE) é um dos principais componentes de um SIG e a junção espacial uma das operações mais importantes nos SGBDEs. Ela envolve o relacionamento entre duas bases de dados, combinando as geometrias de acordo com algum predicado espacial, como intersecção. Devido à crescente disponibilidade de dados espaciais, ao aumento no número de usuários dos SIGS e ao alto custo de processamento das operações espaciais, os SGBDE distribuídos (SGBDED) surgem com uma boa opção para processar a junção espacial de forma eficiente em um cluster de computadores. Esse processamento distribuído traz consigo alguns desafios, tais como a distribuição dos dados pelo cluster e o processamento paralelo e distribuído da junção espacial. O objetivo deste trabalho é apresentar uma plataforma de geoprocessamento paralelo e distribuído da junção espacial em um cluster de computadores, utilizando técnicas de distribuição de dados para bases de dados dinâmicas. Os trabalhos encontrados na literatura têm explorado técnicas de distribuição de dados indicadas para bases de dados estáticas, onde qualquer atualização da base de dados requer que todos os dados sejam novamente distribuídos pelo cluster. Isto se torna inviável com grandes bases de dados e que sofrem constantes atualizações. Por isso, este trabalho propõe duas novas técnicas de distribuição de dados com bases de dados dinâmicas: Proximity Area e Grid Proximity Area. Estas técnicas foram avaliadas para definir em quais cenários cada uma delas é mais apropriada. Para tal, estas técnicas foram avaliadas em um ambiente real com bases de dados com características diferentes, para que fosse possível experimentar a junção espacial distribuída em cenários diversos com cada técnica de distribuição de dados.

Signal detection from Adverse Event Reports (AERs) is important for identifying and analysing drug safety concern after a drug has been released into the market. A safety signal is defined as a possible causal relation between an adverse event and a drug. There are a number of safety signal detection algorithms available for detecting drug safety concern. They compare the ratio of observed count to expected count to find instances of disproportionate reportings of an event for a drug or combination of events for a drug. In this thesis, we present an algorithm to mine the AERs to identify drugs which show sudden and large changes in patterns of reporting of adverse events. Unlike other algorithms, the proposed algorithm creates time series for each drug and use it to identify start of a potential safety problem. A novel vectorized timeseries utilizing multiple attributes has been proposed here. First a time series with a small time period was created; then to remove local variations of the number of reports in a time period, a time-window based averaging was done. This method helped to keep a relatively long time-series, but eliminated local variations. The steps in the algorithm include partitioning the counts on attribute values, creating a vector out of the partitioned counts for each time period, use of a sliding time window, normalizing the vectors and computing vector differences to find the changes in reporting over time. Weights have been assigned to attributes to highlight changes in the more significant attributes. The algorithm was tested with Adverse Event Reporting System (AERS) datasets from Food and Drug Administation (FDA). From AERS datasets the proposed algorithm identified five drugs that may have safety concern. After searching literature and the Internet it was found that the five drugs the algorithm identified, two were recalled, one was suspended, one had to undergo label change and the other one has a lawsuit pending against it.

Η παρούσα διπλωματική εργασία πραγματοποιήθηκε κατά το διάστημα 2012-2013 στο Εργαστήριο Συστημάτων Υπολογιστών (CSL) του Πανεπιστημίου Πατρών. Στόχος της εργασίας είναι η επίλυση ενός συνόλου προβλημάτων χρονοπρογραμματισμού εξετάσεων (ETP, Carter Dataset), με χρήση πληροφορημένου γενετικού αλγορίθμου. Στην εργασία αυτή θα παρουσιαστούν, τα βασικά μοντέλα λειτουργίας των γενετικών αλγορίθμων, του ETP καθώς και παρουσίαση βασικών εννοιών των παράλληλων συστημάτων. Τέλος παρουσιάζεται ο σειριακός κώδικας που υλοποιήθηκε σε ANSI-C και στην συνέχεια γίνεται σύγκριση με τον παράλληλο κώδικα που υλοποιήθηκε με MPI-C και παρουσιάζονται τα αποτελέσματα της σύγκρισης μεταξύ των δύο. / The Aim of this thesis which was completed during the 2012/2013 academic year at the Computer Systems Laboratory (CSL) at the University of Patras is to solve a set of Examination Timetabling Problems (Carter Dataset,ETP) with the aid of an informed genetic algorithm. I will present the basic model under which the genetic algorithms operate and some information about the ETP and general parallel systems. To conclude we will present our serial ANSI-C code and compare it with the parallel MPI-C code that we build and compare the two results.

[pt] O volume de dados RDF publicados na Web aumentou consideravelmente, o que ressaltou a importância de seguir os princípios de dados interligados para promover a interoperabilidade. Um dos princípios afirma que todo novo conjunto de dados deve ser interligado com outros conjuntos de dados publicados na Web. Esta tese contribui para abordar este princípio de duas maneiras. Em primeiro lugar, utiliza algoritmos de detecção de comunidades e técnicas de criação de perfis para a criação e análise automática de um diagrama da nuvem da LOD (Linked Open Data), o qual facilita a localização de conjuntos de dados na nuvem da LOD. Em segundo lugar, descreve três abordagens, apoiadas por ferramentas totalmente implementadas, para recomendar conjuntos de dados a serem interligados com um novo conjunto de dados, um problema conhecido como problema de recomendação de interligação de conjunto de dados. A primeira abordagem utiliza medidas de previsão de links para produzir recomendações de interconexão. A segunda abordagem emprega algoritmos de aprendizagem supervisionado, juntamente com medidas de previsão de links. A terceira abordagem usa algoritmos de agrupamento e técnicas de criação de perfil para produzir recomendações de interconexão. Essas abordagens são implementadas, respectivamente, pelas ferramentas TRT, TRTML e DRX. Por fim, a tese avalia extensivamente essas ferramentas, usando conjuntos de dados do mundo real. Os resultados mostram que estas ferramentas facilitam o processo de criação de links entre diferentes conjuntos de dados. / [en] The volume of RDF data published on the Web increased considerably, which stressed the importance of following the Linked Data principles to foster interoperability. One of the principles requires that a new dataset should be interlinked with other datasets published on the Web. This thesis contributes to addressing this principle in two ways. First, it uses community detection algorithms and profiling techniques for the automatic creation and analysis of a Linked Open Data (LOD) diagram, which facilitates locating datasets in the LOD cloud. Second, it describes three approaches, backed up by fully implemented tools, to recommend datasets to be interlinked with a new dataset, a problem known as the dataset interlinking recommendation problem. The first approach uses link prediction measures to provide a list of datasets recommendations for interlinking. The second approach employs supervised learning algorithms, jointly with link prediction measures. The third approach uses clustering algorithms and profiling techniques to produce dataset interlinking recommendations. These approaches are backed up, respectively, by the TRT, TRTML and DRX tools. Finally, the thesis extensively evaluates these tools, using real-world datasets, reporting results that show that they facilitate the process of creating links between disparate datasets.

Pedestrian detection is an important field with applications in active safety systems for cars as well as autonomous driving. Since autonomous driving and active safety are becoming technically feasible now the interest for these applications has dramatically increased.The aim of this thesis is to investigate convolutional neural networks (CNN) for pedestrian detection. The reason for this is that CNN have recently beensuccessfully applied to several different computer vision problems. The main applications of pedestrian detection are in real time systems. For this reason,this thesis investigates strategies for reducing the computational complexity offorward propagation for CNN.The approach used in this thesis for extracting pedestrians is to use a CNN tofind a probability map of where pedestrians are located. From this probabilitymap bounding boxes for pedestrians are generated. A method for handling scale invariance for the objects of interest has also been developed in this thesis. Experiments show that using this method givessignificantly better results for the problem of pedestrian detection.The accuracy which this thesis has managed to achieve is similar to the accuracy for some other works which use CNN.

abstract: Internet memes have become a widespread tool used by people for interacting and exchanging ideas over social media, blogs, and open messengers. Internet memes most commonly take the form of an image which is a combination of image, text, and humor, making them a powerful tool to deliver information. Image memes are used in viral marketing and mass advertising to propagate any ideas ranging from simple commercials to those that can cause changes and development in the social structures like countering hate speech. This work proposes to treat automatic image meme generation as a translation process, and further present an end to end neural and probabilistic approach to generate an image-based meme for any given sentence using an encoder-decoder architecture. For a given input sentence, a meme is generated by combining a meme template image and a text caption where the meme template image is selected from a set of popular candidates using a selection module and the meme caption is generated by an encoder-decoder model. An encoder is used to map the selected meme template and the input sentence into a meme embedding space and then a decoder is used to decode the meme caption from the meme embedding space. The generated natural language caption is conditioned on the input sentence and the selected meme template. The model learns the dependencies between the meme captions and the meme template images and generates new memes using the learned dependencies. The quality of the generated captions and the generated memes is evaluated through both automated metrics and human evaluation. An experiment is designed to score how well the generated memes can represent popular tweets from Twitter conversations. Experiments on Twitter data show the efficacy of the model in generating memes capable of representing a sentence in online social interaction. / Dissertation/Thesis / Masters Thesis Computer Science 2020

El texto completo de este trabajo no está disponible en el Repositorio Académico UPC por restricciones de la casa editorial donde ha sido publicado. / Problems related to public transport and private transport in Peru are persistent. New proposals to solve them arise, currently the world of data analysis is starting in Peru, there are not many open datasets useful that allow proposing solutions in each environment. In this paper, we will collect relevant data of the transport located in a province of Peru with more than 1000 users involved, restricted by a delimited geographic area and with 2 years of operations and more than 3000 transport services tracked. In this way, we highlight the importance of the data, the possible potential uses within the transport, and a case of use of the collected dataset. / Revisión por pares

No description available.

Sensors, as a crucial part of autonomous driving, are primarily used for perceiving the environment. The recent deep learning development of different sensors has demonstrated the ability of machines recognizing and understanding their surroundings. Automotive radar, as a primary sensor for self-driving vehicles, is well-known for its robustness against variable lighting and weather conditions. Compared with camera-based deep learning development, Object detection using automotive radars has not been explored to its full extent. This can be attributed to the lack of public radar datasets. In this thesis, we collect a novel radar dataset that contains radar data in the form of Range-Azimuth-Doppler tensors along with the bounding boxes on the tensor for dynamic road users, category labels, and 2D bounding boxes on the Cartesian Bird-EyeView range map. To build the dataset, we propose an instance-wise auto-annotation algorithm. Furthermore, a novel Range-Azimuth-Doppler based multi-class object detection deep learning model is proposed. The algorithm is a one-stage anchor-based detector that generates both 3D bounding boxes and 2D bounding boxes on Range-AzimuthDoppler and Cartesian domains, respectively. Our proposed algorithm achieves 56.3% AP with IOU of 0.3 on 3D bounding box predictions, and 51.6% with IOU of 0.5 on 2D bounding box predictions. Our dataset and the code can be found at https://github.com/ZhangAoCanada/RADDet.git.