-

<p>Deep ‘Analog Artificial Neural Networks’ (AANNs) perform complex classification problems with high accuracy. However, they rely on humongous amount of power to perform the calculations, veiling the accuracy benefits. The biological brain on the other hand is significantly more powerful than such networks and consumes orders of magnitude less power, indicating some conceptual mismatch. Given that the biological neurons are locally connected, communicate using energy efficient trains of spikes, and the behavior is non-deterministic, incorporating these effects in Artificial Neural Networks (ANNs) may drive us few steps towards a more realistic neural networks. </p> <p> </p> <p>Emerging devices can offer a plethora of benefits including power efficiency, faster operation, low area in a vast array of applications. For example, memristors and Magnetic Tunnel Junctions (MTJs) are suitable for high density, non-volatile Random Access Memories when compared with CMOS implementations. In this work, we analyze the possibility of harnessing the characteristics of such emerging devices, to achieve neuro-inspired solutions to intricate problems.</p> <p> </p> <p>We propose how the inherent stochasticity of nano-scale resistive devices can be utilized to realize the functionality of spiking neurons and synapses that can be incorporated in deep stochastic Spiking Neural Networks (SNN) for image classification problems. While ANNs mainly dwell in the aforementioned classification problem solving domain, they can be adapted for a variety of other applications. One such neuro-inspired solution is the Cellular Neural Network (CNN) based Boolean satisfiability solver. Boolean satisfiability (k-SAT) is an NP-complete (k≥3) problem that constitute one of the hardest classes of constraint satisfaction problems. We provide a proof of concept hardware based analog k-SAT solver that is built using MTJs. The inherent physics of MTJs, enhanced by device level modifications, is harnessed here to emulate the intricate dynamics of an analog, CNN based, satisfiability (SAT) solver. </p> <p> </p> <p>Furthermore, in the effort of reaching human level performance in terms of accuracy, increasing the complexity and size of ANNs is crucial. Efficient algorithms for evaluating neural network performance is of significant importance to improve the scalability of networks, in addition to designing hardware accelerators. We propose a scalable approach for evaluating Liquid State Machines: a bio-inspired computing model where the inputs are sparsely connected to a randomly interlinked reservoir (or liquid). It has been shown that biological neurons are more likely to be connected to other neurons in the close proximity, and tend to be disconnected as the neurons are spatially far apart. Inspired by this, we propose a group of locally connected neuron reservoirs, or an ensemble of liquids approach, for LSMs. We analyze how the segmentation of a single large liquid to create an ensemble of multiple smaller liquids affects the latency and accuracy of an LSM. In our analysis, we quantify the ability of the proposed ensemble approach to provide an improved representation of the input using the Separation Property (SP) and Approximation Property (AP). Our results illustrate that the ensemble approach enhances class discrimination (quantified as the ratio between the SP and AP), leading to improved accuracy in speech and image recognition tasks, when compared to a single large liquid. Furthermore, we obtain performance benefits in terms of improved inference time and reduced memory requirements, due to lower number of connections and the freedom to parallelize the liquid evaluation process.</p>

It`s blameful to do nothing. You should catch the day and its possibilities. Every second, minute and day, you have the potential to be constantly productive. Do not lose a second of time that you can use to create an imprint of your existence. Do not waste your time when you have the opportunity to create something measurably valuable. Soon you are gone and the competition is great. Doing nothing is a task that fewer and fewer are devoted to, to a lesser extent. We are expected to always be accessible and keep up to date with what is happening around us. There is a constant occupation of our sense of mind, thoughtfulness and rest is unprioritized. We try to maximize the part of our time when we are productive and active to absurdity. The goal seems to make as big imprint of our existence as possible. Fast and more is applauded until the day we burn ourselves out. In my work, I investigate passivity as backlash in a faster world, in a world where focus is on production, efficiency and results. A world where everything worth valuing is measurable. What is really a desired result and is it always possible to measure? Who really gets that gain from our fast life? Can passivity and slowness act as an activist act? Can one passively create something measurable? Is burning time perhaps our time´s most provocative act? With my work, I want to question my constant bad conscience about being insufficient and my feeling of not doing enough. If I can´t do enough, I do nothing instead. / Det är skuldbelagt att göra ingenting. Du bör fånga dagen och dess möjligheter. Varje sekund, minut och dag har du potential att vara konstant produktiv. Förlora inte en sekund av tid som du kan använda till att skapa ett avtryck av din existens. Slösa inte med din tid när du har möjlighet att skapa något mätbart värdefullt. Snart är du borta och konkurrensen är stor. Att göra ingenting är en syssla som allt färre ägnar sig åt. Vi förväntas alltid vara tillgängliga och hålla oss uppdaterade om vad som händer runt oss. Det sker ständigt en ockupation av våra sinnesintryck. Eftertänksamhet och vila prioriteras bort. Vi försöker att maximera den del av vår tid då vi är produktiva och aktiva till absurditet. Målet verkar vara att hinna sätta ett så stort avtryck av vår existens som möjligt. Snabbare och mer applåderas till den dagen då vi bränner ut oss. I mitt arbete undersöker jag passivitet som motreaktion till en snabbare värld, i en värld där fokus ligger på produktion, effektivitet och resultat. I en värld där endast det som är mätbart ges ett värde. Vad är egentligen ett önskat resultat och är det alltid möjligt att mäta? Vem är det egentligen som får vinsten av vårt snabba liv? Kan passivitet och långsamhet fungera som en aktivistisk handling? Kan man passivt skapa något mätbart? Är ”att bränna tid”  kanske vår tids mest provokativa handling? Med mitt arbete vill jag ifrågasätta mitt konstanta dåliga samvete om att vara otillräcklig och min känsla av att jag inte gör tillräckligt. Om jag inte kan göra nog så gör jag istället ingenting.

[pt] IoT refere-se a um sistema de dispositivos de computação inter-relacionados que visa transferir dados através de uma rede sem exigir interação humanohumano ou humano-para-computador. Esses sistemas de comunicação modernos, exigem restrições de baixo consumo de energia e baixa complexidade no receptor. Nesse sentido, o conversor analógico-digital representa um gargalo para o desenvolvimento das aplicações dessas novas tecnologias, pois apresenta alto consumo de energia devido à sua alta resolução. A pesquisa realizada em relação aos conversores analógico-digitais com quantização grosseira mostrou que esses dispositivos são promissores para o projeto de futuros sistemas de comunicação. Para equilibrar a perda de informações, devido à quantização grosseira, a resolução no tempo é aumentada através da superamostragem. Esta tese considera um sistema com quantização de 1 bit e superamostragem no receptor com um canal de downlink MIMO multiusuário com banda ilimitada e apresenta, como principal contribuição, a nova modulação de cruzamento de zeros que implica que a informação é transmitida no instante de tempo zero-crossings. Este método é usado para a pré-codificação temporal através da otimização do design da forma de onda para dois pré-codificadores diferentes, a maximização temporal da distância mínima até o limiar de decisão com forçamento a zero espacial e a pré-codificação MMSE no espácio-temporal. Os resultados da simulação mostram que a abordagem de cruzamento de zeros proposta supera o estado da arte em termos da taxa de erro de bits para os dois pré-codificadores estudados. Além disso, essa nova modulação reduz a complexidade computacional, permite dispositivos de complexidade muito baixa e economiza recursos de banda em comparação com o método mais avançado. Análises adicionais mostram que a abordagem do cruzamento de zeros é benéfica em comparação com o método mais avançado em termos de maior distância mínima até o limiar de decisão e menor MSE para sistemas com limitações de banda. Além disso, foi desenvolvido um esquema de mapeamento de bits para modulação de cruzamento por zero, semelhante à codificação de Gray para reduzir ainda mais a taxa de erro de bits. / [en] The IoT refers to a system of interrelated computing devises which aims to transfer data over a network without requiring human-to-human or humanto- computer interaction. This Modern communication systems demand restrictions of low energy consumption and low complexity in the receiver. In this sense, the analog-to-digital converter represents a bottleneck for the development of the applications of these new technologies since it has a high energy consumption due to its high resolution. The research carried out concerning to the analog-to-digital converters with coarse quantization has shown that such devices are promising for the design of future communication systems. To balance the loss of information, due to the coarse quantization, the resolution in time is increased through oversampling. This thesis considers a system with 1-bit quantization and oversampling at the receiver with a bandlimited multiuser MIMO downlink channel and introduces, as the main contribution, the novel zero-crossing modulation which implies that the information is conveyed within the time instant of the zero-crossings. This method is used for the temporal precoding through the waveform design optimization for two different precoders, the temporal maximization of the minimum distance to the decision threshold with spatial zero forcing and the space-time MMSE precoding. The simulation results show that the proposed zero-crossing approach outperforms the state-of-theart in terms of the bit error rate for both precoders studied. In addition, this novel modulation reduces the computational complexity, allows very low complexity devices and saves band resources in comparison to the state-ofthe- art method. Additional analyses show that the zero-crossing approach is beneficial in comparison to the state-of-the-art method in terms of greater minimum distance to the decision threshold and lower MSE for systems with band limitations. Moreover, it was devised a bit-mapping scheme for zero-crossing modulation, similar to Gray-coding to further reduce the bit error rate.

Dengue and Zika are two arboviral viruses that affect a significant portion of the world population. The principal vector species of both viruses are Aedes aegypti and Aedes albopictus mosquitoes. They breed in very slow flowing or standing pools of water. It is important to reduce and control such potential breeding grounds to contain the spread of these diseases. This thesis investigates a model for the detection of water bodies using high-resolution images collected by Unmanned Aerial Vehicles (UAVs) in tropical countries, exemplified by Sri Lanka, and their multispectral information to help detect water bodies where larvae are most likely to breed quickly and accurately. Although machine learning has been studied in previous work to process multispectral image information to obtain the location of water bodies, different machine learning methods have not been compared, only random forest algorithms have been used. Because Convolutional Neural Networks (CNNs) are known to provide advanced classification performance for visual recognition tasks, in this thesis, faster region-based CNNs are introduced to perform fast and accurate identification of water body locations. In order to better evaluate the experimental results, this thesis introduces Intersection over Union (IoU) as a criterion for evaluating the results. On the one hand, IoU can judge the success rate of the model for water region recognition, and on the other hand, analysis of the model recall rate under different IoU values can also evaluate the model’s ability to detect the range of water regions. Meanwhile, the basic CNN network and random forest algorithm in the previous work are also implemented to compare the results of faster region-based CNNs. In conclusion, the faster region-based CNN model achieves the best results with a 98.33% recognition success rate for water bodies in multispectral images, compared to 95.80% for the CNN model and 95.74% for the random forest model. In addition, the faster region-based CNN model significantly outperformed the CNN model and the random forest model for training speed. / Dengue och zika är två arbovirala virus som drabbar en stor del av världens befolkning. De viktigaste vektorerna för båda virusen är myggorna Aedes aegypti och Aedes albopictus. De förökar sig i mycket långsamt rinnande eller stående vattensamlingar. Det är viktigt att minska och kontrollera sådana potentiella grogrunder för att begränsa spridningen av dessa sjukdomar. I denna avhandling undersöks en modell för att upptäcka vattenområden med hjälp av högupplösta bilder som samlas in av Unmanned Aerial Vehicles (UAV) i tropiska länder, exemplifierat av Sri Lanka, och deras multispektrala information för att hjälpa till att upptäcka vattenområden där larverna sannolikt förökar sig snabbt och noggrant. Även om maskininlärning har studerats i tidigare arbeten för att bearbeta multispektral information från bilder för att få fram platsen för vattenförekomster, har olika metoder för maskininlärning inte jämförts, utan endast random forest-algoritmer har använts. Eftersom Convolutional Neural Networks (CNN) är kända för att erbjuda avancerade klassificeringsprestanda för visuella igenkänningsuppgifter i denna avhandling introduceras snabbare regionbaserade CNN för att utföra snabb och exakt identifiering av vattenkropparnas läge. För att bättre kunna utvärdera de experimentella resultaten införs i denna avhandling Intersection over Union (IoU) som ett kriterium för utvärdering av resultaten. Å ena sidan kan IoU bedöma modellens framgång för igenkänning av vattenområden, och å andra sidan kan analysen av modellens återkallningsfrekvens under olika IoU-värden också utvärdera modellens förmåga att upptäcka olika vattenområden. Samtidigt genomförs även det grundläggande CNN-nätverket och algoritmen för slumpmässig skog i det tidigare arbetet för att jämföra resultaten av Faster regionbaserad CNN. Sammanfattningsvis ger den snabbare regionbaserade CNN-modellen de bästa resultaten med 98,33% av alla igenkänningsresultat för vattenkroppar i multispektrala bilder, jämfört med 95,80% för CNN-modellen och 95,74% för modellen med slumpmässig skog. Dessutom överträffade den snabbare regionbaserade CNN-modellen CNN-modellen och random forest-modellen avsevärt när det gäller träningshastighet.

This master’s thesis investigates the application of advanced machine learning models for the identification and classification of Swedish predators using camera trap images. With the growing threats to biodiversity, there is an urgent need for innovative and non-intrusive monitoring techniques. This study focuses on the development and evaluation of object detection models, including YOLOv5, YOLOv8, YOLOv9, and Faster R-CNN, aiming to enhance the surveillance capabilities of Swedish predatory species such as bears, wolves, lynxes, foxes, and wolverines. The research leverages a dataset from the NINA database, applying data preprocessing and augmentation techniques to ensure robust model training. The models were trained and evaluated using various dataset sizes and conditions, including day and night images. Notably, YOLOv8 and YOLOv9 underwent extended training for 300 epochs, leading to significant improvements in performance metrics. The performance of the models was evaluated using metrics such as mean Average Precision (mAP), precision, recall, and F1-score. YOLOv9, with its innovative Programmable Gradient Information (PGI) and GELAN architecture, demonstrated superior accuracy and reliability, achieving an F1-score of 0.98 on the expanded dataset. The research found that training models on images captured during both day and night jointly versus separately resulted in only minor differences in performance. However, models trained exclusively on daytime images showed slightly better performance due to more consistent and favorable lighting conditions. The study also revealed a positive correlation between the size of the training dataset and model performance, with larger datasets yielding better results across all metrics. However, the marginal gains decreased as the dataset size increased, suggesting diminishing returns. Among the species studied, foxes were the least challenging for the models to detect and identify, while wolves presented more significant challenges, likely due to their complex fur patterns and coloration blending with the background.

Ces dernières années, le domaine de l’apprentissage profond a connu des progrès énormes dans des applications allant de la génération d’images, détection d’objets, modélisation du langage à la réponse aux questions visuelles. Les approches classiques telles que l’apprentissage supervisé nécessitent de grandes quantités de données étiquetées et spécifiques à la tâches. Cependant, celles-ci sont parfois coûteuses, peu pratiques, ou trop longues à collecter. La modélisation efficace en données, qui comprend des techniques comme l’apprentissage few-shot (à partir de peu d’exemples) et l’apprentissage self-supervised (auto-supervisé), tentent de remédier au manque de données spécifiques à la tâche en exploitant de grandes quantités de données plus “générales”. Les progrès de l’apprentissage profond, et en particulier de l’apprentissage few-shot, s’appuient sur les benchmarks (suites d’évaluation), les métriques d’évaluation et les jeux de données, car ceux-ci sont utilisés pour tester et départager différentes méthodes sur des tâches précises, et identifier l’état de l’art. Cependant, du fait qu’il s’agit de versions idéalisées de la tâche à résoudre, les benchmarks sont rarement équivalents à la tâche originelle, et peuvent avoir plusieurs limitations qui entravent leur rôle de sélection des directions de recherche les plus prometteuses. De plus, la définition de métriques d’évaluation pertinentes peut être difficile, en particulier dans le cas de sorties structurées et en haute dimension, telles que des images, de l’audio, de la parole ou encore du texte. Cette thèse discute des limites et des perspectives des benchmarks existants, des fonctions de coût (training losses) et des métriques d’évaluation (evaluation metrics), en mettant l’accent sur la modélisation générative - les Réseaux Antagonistes Génératifs (GANs) en particulier - et la modélisation efficace des données, qui comprend l’apprentissage few-shot et self-supervised. La première contribution est une discussion de la tâche de modélisation générative, suivie d’une exploration des propriétés théoriques et empiriques des fonctions de coût des GANs. La deuxième contribution est une discussion sur la limitation des few-shot classification benchmarks, certains ne nécessitant pas de généralisation à de nouvelles sémantiques de classe pour être résolus, et la proposition d’une méthode de base pour les résoudre sans étiquettes en phase de testing. La troisième contribution est une revue sur les méthodes few-shot et self-supervised de détection d’objets , qui souligne les limites et directions de recherche prometteuses. Enfin, la quatrième contribution est une méthode efficace en données pour la description de vidéo qui exploite des jeux de données texte et vidéo non supervisés. / In recent years, the field of deep learning has seen tremendous progress for applications ranging from image generation, object detection, language modeling, to visual question answering. Classic approaches such as supervised learning require large amounts of task-specific and labeled data, which may be too expensive, time-consuming, or impractical to collect. Data-efficient methods, such as few-shot and self-supervised learning, attempt to deal with the limited availability of task-specific data by leveraging large amounts of general data. Progress in deep learning, and in particular, few-shot learning, is largely driven by the relevant benchmarks, evaluation metrics, and datasets. They are used to test and compare different methods on a given task, and determine the state-of-the-art. However, due to being idealized versions of the task to solve, benchmarks are rarely equivalent to the original task, and can have several limitations which hinder their role of identifying the most promising research directions. Moreover, defining meaningful evaluation metrics can be challenging, especially in the case of high-dimensional and structured outputs, such as images, audio, speech, or text. This thesis discusses the limitations and perspectives of existing benchmarks, training losses, and evaluation metrics, with a focus on generative modeling—Generative Adversarial Networks (GANs) in particular—and data-efficient modeling, which includes few-shot and self-supervised learning. The first contribution is a discussion of the generative modeling task, followed by an exploration of theoretical and empirical properties of the GAN loss. The second contribution is a discussion of a limitation of few-shot classification benchmarks, which is that they may not require class semantic generalization to be solved, and the proposal of a baseline method for solving them without test-time labels. The third contribution is a survey of few-shot and self-supervised object detection, which points out the limitations and promising future research for the field. Finally, the fourth contribution is a data-efficient method for video captioning, which leverages unsupervised text and video datasets, and explores several multimodal pretraining strategies.