Tesis sobre el tema "Dynamic encoding"

Non-Constant Luminance (NCL) and Constant Luminance (CL) are the two common methods for converting RGB values to luma and chroma for compression efficiency. CL coefficients have been derived from the luminous efficacy of the used gamut color primaries in the light linear domain. NCL applies the same coefficients but on non-linear inputs, which are perceptually encoded values using proper transfer function, thus leading to reduced compression efficiency and color shifts. However, since legacy cameras capture perceptually encoded values of light, it is common practice to use NCL in the existing video distribution pipelines. Although color distortion was not a serious problem with legacy Standard Dynamic Range (SDR) systems, this is not the case with High Dynamic Range (HDR) applications where color shifts become much more visible and prohibitive to delivering high quality HDR. In this thesis, we propose methods that address the inefficiencies of the conventional NCL method by optimizing NCL luma values to be as close as possible to those of CL, thus improving compression performance and color accuracy, while maintaining the current pipeline infrastructure. First, we develop a global optimization method for deriving new optimum coefficients that approximate NCL values to those of the CL approach. Then, we improve upon this approach by conducting content based optimization. This adaptive optimization method takes content pixel density into consideration and optimizes only based on these color distributions. Finally, we propose a weighted global optimization method, which separates chromaticity into three categories (Red, Green, and Blue), and assigns weights based on their contributions to luminance. Evaluations show that the proposed method improves color quality and compression efficiency over NCL.
Applied Science, Faculty of
Electrical and Computer Engineering, Department of
Graduate

The flexibility and self-describing nature of XML has made it the most common mark-up language used for data representation over the Web. XML data is naturally modelled as a tree, where the structural tree information can be encoded into labels via XML labelling scheme in order to permit answers to queries without the need to access original XML files. As the transmission of XML data over the Internet has become vibrant, it has also become necessary to have an XML labelling scheme that supports dynamic XML data. For a large-scale and frequently updated XML document, existing dynamic XML labelling schemes still suffer from high growth rates in terms of their label size, which can result in overflow problems and/or ambiguous data/query retrievals. This thesis considers the compression of XML labels. A novel XML labelling scheme, named “Base-9”, has been developed to generate labels that are as compact as possible and yet provide efficient support for queries to both static and dynamic XML data. A Fibonacci prefix-encoding method has been used for the first time to store Base-9’s XML labels in a compressed format, with the intention of minimising the storage space without degrading XML querying performance. The thesis also investigates the compression of XML labels using various existing prefix-encoding methods. This investigation has resulted in the proposal of a novel prefix-encoding method named “Elias-Fibonacci of order 3”, which has achieved the fastest encoding time of all prefix-encoding methods studied in this thesis, whereas Fibonacci encoding was found to require the minimum storage. Unlike current XML labelling schemes, the new Base-9 labelling scheme ensures the generation of short labels even after large, frequent, skewed insertions. The advantages of such short labels as those generated by the combination of applying the Base-9 scheme and the use of Fibonacci encoding in terms of storing, updating, retrieving and querying XML data are supported by the experimental results reported herein.

Bats are known for their highly capable biosonar systems which make them be able to navigate and forage in dense vegetation. Their biosonar system consists of one emitter (nose or mouth) and two receivers (ears). Some bat species, e.g. in the rhinolophid and hipposiderid families, have complicated pinna motion patterns. It has been shown that these pinna motion patterns fall into two distinct categories: rigid motions and non-rigid motions. In the current work, the pinna of Pratt's leaf-nosed bat (Hipposideros pratti) was used as a biological model system to understand how a sensor could benefit from variability. Hence, the variability in the rigid pinna motions and in the non-rigid pinna motions has been investigated by tracking a dense set of landmarks on the pinna surface with stereo vision. Axis-angle representations have shown that the rigid pinna motions exhibited a large continuous variation with rotation axes covering 180 degrees in azimuth and elevation. Distributions of clusters of the landmarks on the pinna surface have shown that the non-rigid pinna motions fall into at least two subgroups. Besides, the acoustic impact of the rigid pinna motions have been investigated using a biomimetic pinna. Normalized mutual information between the acoustic inputs with different rotation axes has shown that different rotation axes can provide at least 50% new sensory information. These results demonstrate that the variability in the pinna motions is an interesting concept for sensor, and how the bats approach that needs to be further investigated.
Master of Science
Sensors have been developed for a long time, and they can be used to detect the environments and then deliver the required sensing information. There are many different types of sensors, such as vision-based sensors (infrared camera and laser scanner) and sound-based sensors (sonar and radar). Ultrasonic transducers are one of the sound-based sensors, and they are more stable and reliable in environments where smoke or steam is present. Similar to human-made ultrasonic transducers, bats have developed highly capable biosonar systems that consist of one ultrasonic emitter (nose or mouth) and two ultrasonic receivers (ears), and these biosonar systems enable them to fly and hunt in cluttered environments. Some bats, e.g. rhinolophid and hipposiderid bats, have dynamic noseleaves (elaborate baffle shapes surrounding the nostrils) and pinna (outer ear), and these could enhance the sensing abilities of bats. Hence, the purpose of this thesis has been to investigate this variability to improve the human-made sensors by focusing on the dynamic pinna of the bats. It has been shown that bats have two distinct categories of pinna motions: rigid motions which change only the orientation of the pinna, and non-rigid motions which change also the shape of the pinna. However, the variability within the rigid and non-rigid pinna motions has received little attention. Therefore, the present work has investigated the variability in the rigid pinna motions and in the non-rigid pinna motions. Landmark points were placed on the pinna of certain bats and the pinna motions were tracked by high-speed video cameras. The rigid pinna motions exhibit a large continuous variation in where the pinna is orientated during rotation. Distributions of clusters of the landmarks on the pinna have shown that the non-rigid pinna motions fall into at least two subgroups. The acoustic impacts of the rigid pinna motions have been studied by a biomimetic pinna which reproduced the observed range of the rigid pinna motions. Ultrasonic signals mimicking the bats were emitted to be received by the biomimetic pinna. Based on these signals, it has been shown that different rotation axes and even small changes can provide over 50% new sensory information. These findings give engineers a potential way to improve the human-made sensors.

With the rapid growth of XML documents to serve as a popular and major media for storage and interchange of the data on the Web, there is an increasing interest in using existing traditional relational database techniques to store and/or query XML data. Since XQuery is becoming a standard XML query language, significant effort has been made in developing an efficient and comprehensive XQuery-to-SQL query processor. In this thesis, we design and implement an XQuery-to-SQL Query Processor based on the Dynamic Intervals approach. We also provide a comprehensive translation for XQuery basic operations and FLWR expressions. The query processor is able to translate a complex XQuery query, which might include arbitrarily composed and nested FLWR expressions, basic functions, and element constructors, into a single SQL query for RDBMS and a physical plan for the XQuery-enhanced Relational Engine. In order to produce efficient and concise SQL queries, succinct XQuery to SQL translation templates and the optimization algorithms for the SQL query generation are proposed and implemented. The preferable merge-join approach is also proposed to avoid the inefficient nested-loop evaluation for FLWR expressions. Merge-join patterns and query rewriting rules are designed to identify XQuery fragments that can utilize the efficient merge-join evaluation. Proofs of correctness of the approach are provided in the thesis. Experimental results justify the correctness of our work.

Diese Doktorarbeit besteht aus zwei Teilen. In dem ersten Teil der Doktorarbeit behandele ich die Kodierung von Kommunikationssignalen in einem burstenden Interneuron im auditorischen System des Grashuepfers Chorthippus biguttulus. Mit der Anzahl der Aktionspotentialen im Burst wird eine zeitliche Komponente der Kommunikationssignale - die Pausendauer - wiedergegeben. Ein Modell basierend auf schneller Exzitation und langsamer Inhibition kann diese spezielle Kodierung erklaeren. Ich zeige, dass eine zeitliche Integration der Aktionspotentiale dieses burstenden Interneurons dazu genutzt werden kann, die Signale zeitskaleninvariant zu dekodieren. Dieser Mechanismus kann in ein umfassenderes Modell eingebaut werden, dass die Verhaltensantwort des Grashuepfers auf Kommunikationssignale widerspiegelt. Im zweiten Teil der Doktorarbeit benutze ich Konzepte aus der Informationstheorie und der Theorie linearer dynamisches Systeme, um den Begriff der ''vorhersagenden Information'' zu operationalisieren. Im einfachen Fall der informations-theoretisch optimalen Vorhersage des naechsten Zeitschrittes, erhalte ich Eigenvektoren, die denjenigen eines anderen etablierten Algorithmuses, der sogenannten ''Slow Feature Analysis'', entsprechen. Im allgemeinen Fall optimiere ich die vorhersagenden Information, die die Vergangenheit des Inputs eines dynamischen Systems ueber die Zukunft des Outputs enthaelt. Dabei gelange ich zu einer informations-theoretisch optimalen Charakterisierung eines reduzierten Systems, die auf den Eigenvektoren der konditionalen Kovarianzmatrix zwischen Inputvergangenheit und Outputzukunft basiert.
This thesis consists of two parts. In the first part, I investigate the coding of communication signal in a bursting interneuron in the auditory system of the grasshopper Chorthippus biguttulus. The intra-burst spike count codes one temporal feature of the communication signal - pause duration. I show that this code can be understood by a model of parallel fast excitation and slow inhibition. Furthermore, temporal integration of the spike train of this bursting interneuron results in a desirable time-scale invariant read-out of the communication signal. This mechanism can be integrated into a more comprehensive model that can explain behavioural response of grasshoppers. In the second part of this thesis, I combine concepts from information theory and linear system theory to operationalize the notion of ''predictive information''. In the simple case of predicting the next time-step of a signal in an information-theoretic optimal sense, I obtain a description by eigenvectors that are identical to another established algorith, the so-called ''Slow Feature Analysis''. In the general case I optimize a dynamical system such that the predictive information in the input past about the output future is optimalle compressed into the state space. Thereby, I obtain an information-theoretically optimal characterization of reduced system, based on the eigenvectors of the conditional covariance matrix between input past and output future.

Every animal on our planet wanders around when looking for something. Whether it is in search of food, a mate, or home, navigation is one of the most performed cognitive behaviors in Nature. Nevertheless, our understanding of how the brain is capable of solving such a simple problem − to move from one point to another − is still incomplete. The decomposition of navigation into cognitive components reveals the complexity of such behavior. To reach a goal, one has to first understand its current position, then to estimate the target position, followed by identifying a route towards the goal-location and, finally, to physically orchestrate a set of motor-actions leading to the desired location. Extensive research on the mammalian hippocampus has revealed its critical role in spatial navigation, memory, and learning. However, the mechanisms for spatial memory encoding, episodic representation, and their behavioral counterparts are still not fully understood. Moreover, we do not know if the mechanisms involved in spatial re- presentation also scale up to conceptual representation from a purely spatial domain, such as task cognitive demands. In this thesis, we present a set of studies focused on spatial and cognitive representation in the insect and mammalian organisms. We show that the problem of spatial representation requires multi-level solutions working simultaneously: from biophysical neuronal mechanisms to behavioral aspects of navigation. Furthermore, with physiological studies of the human medial temporal lobe, we propose that the mechanisms involved in spatial representation also extend to higher-order cognitive representations, therefore suggesting that the hippocampus handles information that is dimension independent.
Tots els animals del nostre planeta passegen mentre busquen alguna cosa. Ja sigui per trobar menjar, parella o un lloc per viure, la nave- gacio` `es un dels comportaments cognitius més realitzats en la natura. No obstant, la nostra comprensió de com el cervell és capaç de resoldre aquest senzill problema − moure’s d’un punt a un altre − encara és incompleta. La descomposició de la navegació en diferents components cognitius revela la complexitat d’aquest comportament. Per assolir un objectiu, hom ha de primer conèixer la seva posició, llavors estimar la posició destí, seguidament identificar una ruta o camí fins a aquesta i, finalment, orquestrar un conjunt d’accions motores que portin fins a la posició desitjada. L’àmplia recerca de l’hipocamp en mamífers ha revelat el seu paper fonamental en la navegació espacial, la memòria i l’aprenentatge. Tot i això, els mecanismes de codificació de la memòria espacial, la representació episódica i els seus homólegs encara no s’han pogut entendre completament. Tanmateix, encara no sabem si els mecanismes involucrats en la representació espacial també escalen des d’un domini purament espacial a la representació de conceptes, com ara les necessitats cognitives d’una tasca. En aquesta tesi presentem un conjunt d’estudis centrats en la representació espacial i cognitiva en el cervell d’insectes i mamífers. Mostrem que el problema de la representació espacial requereix de solucions amb múltiples nivells treballant simultàniament: des dels mecanismes neuronals biofísics fins als aspectes conductuals de la navegació. Per últim, a partir d’estudis fisiológics del lobul temporal mitjà de l’ésser humà, proposem que els mecanismes involucrats en la representació espacial també s’extenen a representacions cognitives d’alt nivell, suggerint que l’hipocamp s’encarrega de la informació independentment de la seva dimensió.

Neurons can respond to input by firing isolated action potentials or spikes. Sequences of spikes have been linked to the encoding of neuron input. However, many neurons also fire bursts; mechanistically distinct responses consisting of brief high-frequency spike firing. Bursts form separate response symbols but historically have not been thought to encode input. However, recent experimental evidence suggests that bursts can encode input in parallel with tonic spikes. The recognition of bursts as distinct encoding symbols raises important questions; these form the basic aims of this thesis: (1) What inputs do bursts encode? (2) Does burst structure provide extra information about different inputs. (3) Is burst coding robust against the presence of noise; an inherent property of all neural systems? (4) What mechanisms are responsible for burst input encoding? (5) How does burst coding manifest in in-vivo neurons. To answer these questions, bursting is studied using a combination of neuron models and in-vivo hippocampal neuron recordings. Models ranged from neuron-specific cell models to models belonging to three fundamentally different burst dynamic classes (unspecific to any neural region). These classes are defined using concepts from non-linear system theory. Together, analysing these model types with in-vivo recordings provides a specific and general analysis of burst encoding. For neuron-specific and unspecific models, a number of model types expressing different levels of biological realism are analysed. For the study of thalamic encoding, two models containing either a single simplified burst-generating current or multiple currents are used. For models simulating three burst dynamic classes, three further models of different biological complexity are used. The bursts generated by models and real neurons were analysed by assessing the input they encode using methods such as information theory, and reverse correlation. Modelled bursts were also analysed for their resilience to simulated neural noise. In all cases, inputs evoking bursts and tonic spikes were distinct. The structure of burst-evoking input depended on burst dynamic class rather than the biological complexity of models. Different n-spike bursts encoded different inputs that, if read by downstream cells, could discriminate complex input structure. In the thalamus, this n-spike burst code explains informative responses that were not due to tonic spikes. In-vivo hippocampal neurons and a pyramidal cell model both use the n-spike code to mark different LFP features. This n-spike burst may therefore be a general feature of bursting relevant to both model and in-vivo neurons. Bursts can also encode input corrupted by neural noise, often outperforming the encoding of single spikes. Both burst timing and internal structure are informative even when driven by strongly noise-corrupted input. Also, bursts induce input-dependent spike correlations that remain informative despite strong added noise. As a result, bursts endow their constituent spikes with extra information that would be lost if tonic spikes were considered the only informative responses.

Cette thèse traite du problème de l'intégration d'une nouvelle information dans un système d'argumentation abstrait. Un tel système est un graphe orienté dont les nœuds représentent les arguments, et les arcs représentent les attaques entre arguments. Il existe divers moyen de décider quels arguments sont acceptés par l'agent qui utilise un tel système pour représenter ses croyances.Il peut arriver dans la vie d'un agent qu'il soit confronté à une information du type "tel argument devrait être accepté", alors que c'est en contradiction avec ses croyances actuelles, représentées par son système d'argumentation.Nous avons étudié dans cette thèse diverses approches pour intégrer une information à un système d'argumentation.Notre première contribution est une adaptation du cadre AGM pour la révision de croyances, habituellement utilisé lorsque les croyances de l'agent sont représentées dans un formalisme logique. Nous avons notamment adapté les postulats de rationalité proposés dans le cadre AGM pour pouvoir caractériser des opérateurs de révision de systèmes d'argumentation, et nous avons proposé différents moyens de générer les systèmes d'argumentation résultant de la révision.Nous avons ensuite proposé d'utiliser la révision AGM comme un outil pour réviser les systèmes d'argumentation. Il s'agit cette fois-ci d'une approche par encodage en logique du système d'argumentation, qui permet d'utiliser les opérateurs de révision usuels pour obtenir le résultat souhaité.Enfin, nous avons étudié le problème du forçage d'un ensemble d'arguments (comment modifier le système pour qu'un ensemble donné soit une extension). Nous avons proposé une nouvelle famille d'opérateurs qui garantissent le succès de l'opération, contrairement aux opérateurs de forçage existants, et nous avons montré qu'une traduction de nos approches en problèmes de satisfaction ou d'optimisation booléenne permet de développer des outils efficaces pour calculer le résultat du forçage
This thesis tackles the problem of integrating a new piece of information in an abstract argumentation framework. Such a framework is a directed graph such that its nodes represent the arguments, and the directed edges represent the attacks between arguments. There are different ways to decide which arguments are accepted by the agent who uses such a framework to represent her beliefs.An agent may be confronted with a piece of information such that "this argument should be accepted", which is in contradiction with her current beliefs, represented by her argumentation framework.In this thesis, we have studied several approaches to incorporate a piece of information in an argumentation framework.Our first contribution is an adaptation of the AGM framework for belief revision, which has been developed for characterizing the incorporation of a new piece of information when the agent's beliefs are represented in a logical setting. We have adapted the rationality postulates from the AGM framework to characterize the revision operators suited to argumentation frameworks, and we have identified several ways to generate the argumentation frameworks resulting from the revision.We have also shown how to use AGM revision as a tool for revising argumentation frameworks. Our approach uses a logical encoding of the argumentation framework to take advantage of the classical revision operators, for deriving the expected result.At last, we have studied the problem of enforcing a set of arguments (how to change an argumentation framework so that a given set of arguments becomes an extension). We have developed a new family of operators which guarantee the success of the enforcement process, contrary to the existing approaches, and we have shown that a translation of our approaches into satisfaction and optimization problems makes possible to develop efficient tools for computing the result of the enforcement

Encoding behaviorally relevant stimuli in a noisy background is critical for animals to survive in their natural environment. We identify core biophysical and synaptic mechanisms that permit the encoding of low frequency signals in pyramidal neurons of the weakly electric fish Apteronotus leptorhynchus, an animal that can accurately encode miniscule (0.1%) amplitude modulations of its self-generated electric field. We demonstrate that slow NMDA-R mediated EPSPs are able to summate over many interspike intervals of the primary electrosensory afferents (EAs), effectively eliminating the EA spike train serial correlations from the pyramidal cell input. This permits stimulus-evoked changes in EA spiking to be transmitted efficiently to downstream ELL pyramidal cells, where a dynamic balance of NMDA-R and GABA-A-R currents is critical for encoding low frequency signals. Interestingly, AMPA-R activity is depressed and plays a negligible role in the generation of action potentials; instead, cell intrinsic membrane noise implements voltage-dependent stochastic resonance to amplify weak sensory input and appears to drive a significant proportion of pyramidal cell spikes. Together, these mechanisms may be sufficient for the ELL to encode signals near the threshold of behavioral detection.

Horseshoe bats possess a sophisticated biosonar system that helps them to negotiate complex unstructured environments by relying primarily on the sound as the far sense. For this, the bats emit brief ultrasonic pulses and listen to incoming echoes to learn about the environment. The sites of emission and reception in these bats are surrounded by baffle structures called "noseleaves" and "pinnae (outer ears)". These are the the only places in the biosonar system where direction-dependent information gets encoded. These baffle structures in bats unlike the engineering systems like megaphones have complex static geometry and can undergo fast deformations at the time of pulse emission/reception. However, the functional significance of the baffle motions in biosonar system is not known. The current work primarily focuses on: i) the study of the impact of noseleaf dynamics on the outgoing sound waves, ii) the study of the impact of baffle dynamics on encoding of sensory information and localization performance of bats. For this, we take a numerical approach where we use computer-animated digital models of bat noseleaves that mimic noseleaf dynamics as observed in bats. The shapes are acoustically characterized (beampatterns) numerically using a finite element implementation. These beampatterns are then analyzed using an information-theoretic approach. The followings findings were obtained: i) noseleaf dynamics altered the spatial distribution of energy, ii) baffle dynamics results in encoding of new sensory information, and iii) the new sensory information encoded due to baffle dynamics significantly improves the performance of biosonar system on the two target localization tasks evaluated here -- direction resolution and direction estimation accuracy. These results affirm the importance of dynamics in biosonar system of horseshoe bats and point at the possibility of biosonar dynamics as a key factor behind the astounding sensory capabilities of these animals that are not yet matched by engineering systems. Thus, these biosonar dynamic principles can help improve the man-made sensing systems and help close the performance gap between active sensing in biology and in engineering.
Ph. D.

La quantité de données numériques produites dans le monde chaque année augmente exponentiellement et les supports actuels de stockage atteignent leurs limites. Dans ce contexte, le stockage de données sur molécules d'ADN est très prometteur. Stockant jusqu’à 10¹⁸ octets par gramme d'ADN pour une consommation d'énergie quasi nulle, il a une durée de vie 100 fois plus longue que les disques durs. Cette technologie de stockage étant en développement, il est opportun d’y intégrer nativement des mécanismes pour sécuriser les données. C’est l’objet de cette thèse. Notre première contribution est une analyse des risques de l’ensemble de la chaîne de stockage, qui nous a permis d’identifier des vulnérabilités des procédés numériques et biologiques, en termes de confidentialité, d’intégrité, de disponibilité et de traçabilité. Une seconde contribution est l’identification d’opérateurs élémentaires permettant des manipulations simples de l’ADN. Avec ceux-ci, nous avons développé notre troisième contribution, une solution de chiffrement DNACipher qui impose un déchiffrement biomoléculaire des molécules avant de pouvoir lire les données correctement. Cette solution, qui repose sur des enzymes, a nécessité le développement d’un codage des données numériques en séquences ADN appelée DSWE ; notre quatrième contribution. Cet algorithme respecte les contraintes liées aux procédés biologiques (e.g. homopolymères) et à notre DNACipher. Enfin, notre dernière contribution est une validation expérimentale de notre chaîne de stockage sécurisée. C’est la première preuve de concept montrant qu’il est possible de sécuriser ce nouveau support de stockage sur la base de manipulations biomoléculaires
The volume of digital data produced worldwide every year is increasing exponentially, and current storage solutions are reaching their limits. In this context, data storage on DNA molecules holds great promise. Storing up to 10¹⁸ bytes per gram of DNA for almost no energy consumption, it has a lifespan 100 times longer than hard disks. As this storage technology is still under development, the opportunity presents itself to natively integrate data security mechanisms. This is the aim of this thesis. Our first contribution is a risk analysis of the entire storage chain, which has enabled us to identify vulnerabilities in digital and biological processes, particularly in terms of confidentiality, integrity, availability and traceability. A second contribution is the identification of elementary biological operators for simple manipulations of DNA. Using these operators, we have developed a DNACipher encryption solution that requires biomolecular decryption (DNADecipher) of the molecules before the data can be read correctly. This third contribution, based on enzymes, required the development of a coding algorithm for digital data into DNA sequences, a contribution called DSWE. This algorithm respects the constraints of biological processes (e.g. homopolymers) and our encryption solution. Our final contribution is an experimental validation of our secure storage chain. This is the first proof of concept showing that it is possible to secure this new storage medium using biomolecular manipulations

The entorhinal cortex (EC) in the medial temporal lobe plays a critical role in memory formation and is implicated in several neurological diseases including temporal lobe epilepsy and Alzheimer’s disease. Despite the known importance of this brain region, little is known about the normal bioelectrical activity patterns of the EC in awake, behaving primates. In order to develop effective therapies for diseases affecting the EC, we must first understand its normal properties. To contribute to our understanding of the EC, I monitored the activity of individual neurons and populations of neurons in the EC of rhesus macaque monkeys during free-viewing of photographs using electrophysiological techniques. The results of these experiments help to explain how primates can form memories of, and navigate through, the visual world. These experiments revealed neurons in the EC that represent visual space with triangular grid receptive fields and other neurons that prefer to fire near image borders. These properties are similar to those previously described in the rodent EC, but here the neuronal responses relate to viewing of remote space as opposed to representing the physical location of the animal. The representation of visual space may be aided by another EC neuron type that was discovered, free-viewing saccade direction cells, neurons that signaled the direction of upcoming saccades. Such a signal could be used by other cells to prepare to fire according to the future gaze location. Many of these spatially-responsive neurons also represented memory for images, suggesting that they may be useful for associating items with their locations. I also examined the neuronal circuitry of recognition memory for visual stimuli in the EC, and I found that population synchronization within the gamma-band (30-140 Hz) in superficial layers of the EC was modulated by stimulus novelty, while the strength of memory formation modulated gamma-band synchronization in the deep layers and in layer III. Furthermore, the strength of connectivity in the gamma-band between different layers was correlated with the strength of memory formation, with deep to superficial power transfer being correlated with stronger memory formation and superficial to deep transfer correlated with weaker memory formation. These findings support several previous investigations of hippocampal-entorhinal connectivity in the rodent and advance our understanding of the functional circuitry of the medial temporal lobe memory system. Finally, I explored the design of a device that could be used to investigate properties of brain tissue in vitro, potentially aiding in the development of treatments for disorders of the EC and other brain structures. We designed, fabricated, and validated a novel device for long-term maintenance of thick brain slices and 3-dimensional dissociated cell cultures on a perforated multi-electrode array. To date, most electrical recordings of thick tissue preparations have been performed by manually inserting electrode arrays. This work demonstrates a simple and effective solution to this problem by building a culture perfusion chamber around a planar perforated multi-electrode array. By making use of interstitial perfusion, the device maintained the thickness of tissue constructs and improved cellular survival as demonstrated by increased firing rates of perfused slices and 3-D cultures, compared to unperfused controls. To the best of our knowledge, this is the first thick tissue culture device to combine forced interstitial perfusion for long-term tissue maintenance and an integrated multi-electrode array for electrical recording and stimulation.

Animals have to deal with an environment presenting vast amounts of ever-changing sensory information. What features of this information flow are captured by the sensory system? My thesis work examines how signals experienced during free olfactory behaviors are processed by first-order olfactory sensory neurons (OSNs) of the Drosophila larva. By combining a novel extracellular recording technique with a microfluidics control system to control odor delivery in time and space, the computational principles underlying the encoding of dynamical odor stimuli were explored in a single OSN. An optogenetic approach was used to explore the OSN coding space and mimic naturalistic odor responses. The results described herein suggest that relative changes of the stimulus and their temporal integration are captured in a single OSN. Finally, larval behavior was characterized in closed-loop virtual odor environments and dissected with respect to the influence of dynamic features of the stimulus. It emerged that the neural activity of a single OSN is firmly correlated with dynamic features, notably the derivative of the stimulus intensity. These findings link the neural activity of single sensory neuron to behavioral transitions. Taken together, the results of this work provide an entry point into the understanding of larval action selection during chemotaxis.
Los animales tienen que hacer frente a un entorno caracterizado por un flujo variable y abundante de información sensorial. ¿Qué rasgos de este flujo de información son capturados por el sistema sensorial? Este trabajo examina como las neuronas olfativas sensoriales de primer orden (NOS) de la larva de Drosophila procesan las señales experimentadas durante comportamientos olfativos incondicionados. Mediante la combinación de una novedosa técnica de grabación extracelular y un sistema de control de microfluidos que permite controlar temporal y espacialmente el suministro de olor, se exploraron los principios computacionales que permiten codificar los estímulos dinámicos de olor en una sola NOS. Para explorar el espacio que codifican las NOS y recrear una respuesta olfatoria naturalista se utilizaron técnicas de optogenética. Los resultados de este estudio sugieren que tanto los cambios de estímulos relativos como su integración temporal son capturados en una sola NOS. Por último, se cuantificó la conducta larval en entornos de olor virtual de bucle cerrado (closed-loop virtual odor environment) y se analizó teniendo en cuenta la influencia de las características dinámicas del estímulo. Del análisis resultó que la actividad neuronal de una única NOS está firmemente correlacionada con características dinámicas, en particular con la derivada de la intensidad del estímulo. Estos hallazgos ponen de manifiesto vinculan la actividad neuronal de una sola neurona sensorial con transiciones del comportamiento. En conjunto, los resultados de este trabajo proporcionan un punto de partida para comprender la toma de decisiones de las larvas durante la quimiotaxis.

Single Ventricle congenital heart defects with cyanotic mixing between systemic and pulmonary circulations afflict 2 per 1000 live births. The total cavo-pulmonary connection (TCPC), where the superior and inferior vena cavae are sutured to the left and right pulmonary arteries, is the current procedure of choice. It is believed that reducing the fluid mechanical power losses in the TCPC will relieve strain on the single functional ventricle. It is hypothesized that a proposed idealized TCPC design, decreases power losses to a level below that of any other TCPC designs, while providing other advantages and increased flexibility. Physical models with slightly different geometries of the proposed design were created, and in vitro experiments carried out with particle image velocimetry (PIV), phase contrast magnetic resonance imaging (PC-MRI), and control volume flow analysis at physiological flow rates. Computational fluid dynamics (CFD) was used for numerical studies of the same geometries as in the physical models. Power losses were calculated using the control volume method and the viscous power dissipation function. The latter method incorporated registration of high-resolution PC-MRI velocity vectors to tetrahedral meshes followed by inverse interpolation of the vectors onto the meshes. Detailed flow structures were analyzed. Results show that the new design is more energy efficient than any other idealized models. Furthermore, a tool was developed to extract flow and vessel information from PC-MRI datasets obtained from patients with Fontan connections. The tool utilized a display algorithm, which was developed for optimal noise detection in PC-MRI images. This enabled accurate segmentation. Comparing PC-MRI images before and after this accurate segmentation showed that the standard deviations of the pixels at the perimeter of the segmented vessel were statistically significantly smaller after the segmentation in 94.1% of the datasets investigated. The developed tool was able to extract flow, flow in the quadrants of vessels, area of the segmented vessel, velocities and pulsatility indices. The velocity vectors were exported for use as CFD boundary conditions in models reconstructed from patient anatomies. A database was created with patient PC-MRI data from approximately 140 patients, which is probably the largest database in the world.

The present thesis focuses on two appealing paradigms that are expected to characterize the next generation of communication systems: microfluidic networking and Machine to Machine (M2M) Communications. Concerning the former topic, we show how it is possible to introduce switching and routing mechanism in microfluidic systems. We define some simple mathematical models that capture the macroscopic behavior of droplets in microfluidic networks. Then, we use them to implement a simulator that is able to reproduce the motion and predict the path of droplets in a generic microfluidic system. We validate the simulator and apply it to design a network with bus topology. Finally, we prove the feasibility of attaining molecular communication in this domain by describing a simple protocol that exploits droplets length/interdistance modulation to send information. The research activity on M2M, instead, is aimed at the investigation of two critical issues that are expected to affect Machine-Type Communication (MTC), i.e. energy efficiency and massive access. Regarding energy efficiency, we address the problem of delivering a fixed data payload over a Rayleigh fading wireless channel with the purpose of minimizing the average total energy cost, given by the sum of the transmit energy and an overhead circuit energy, to complete it. This scenario is well suited for uplink cellular MTC in future 5G Internet of Things (IoT) use cases, where the focus is more on device energy efficiency than on throughput. We describe the optimal transmission policies to be used under various coordinated access scenarios with different levels of channel state information and transmitter/receiver capabilities, and show the corresponding theoretical bounds. In the last part of the work, we study the asymptotic performance of uncoordinated access schemes with Multi Packet Reception (MPR) and Successive Interference Cancellation (SIC) techniques for contention resolution at the receiver. The corresponding results in terms of throughput in a massive access M2M scenario are finally evaluated and discussed.
La presente tesi si focalizza sullo studio di due importanti paradigmi che si prevede possano caratterizzare i sistemi di comunicazione di prossima generazione: le reti microfluidiche e le comunicazioni Machine to Machine (M2M). Riguardo alle reti microfluidiche, in questo lavoro illustriamo come sia possibile introdurre elementi di switch e meccanismi di routing all’interno di sistemi microfluidici. Definiamo poi alcuni semplici modelli matematici che descrivono il comportamento macroscopico di gocce all’interno di tali reti. Questi ultimi sono quindi sfruttati per implementare un simulatore che è capace di riprodurre il movimento e predire il percorso delle gocce in un generico sistema microfluidico. Dopo averlo validato sperimentalmente, il simulatore è impiegato per progettare una rete microfluidica con topologia a bus. Infine, viene dimostrato come sia possibile realizzare comunicazioni molecolari in questo ambito tramite la formalizzazione e la descrizione di un protocollo che sfrutta la modulazione della lunghezza/interdistanza delle gocce per trasferire informazione. L’attività di ricerca in merito alle comunicazioni M2M, invece, è finalizzata allo studio di due importanti criticità insite nelle Machine-Type Communications (MTCs), ovvero l’efficienza energetica e l’accesso simultaneo di massa (massive access). Per quanto concerne l’efficienza energetica, viene affrontato il problema di trasmettere un payload di una certa lunghezza fissata attraverso un canale wireless affetto da Rayleigh fading con lo scopo di minimizzare il costo totale medio dell’utente finale, dato dalla somma dell’energia di trasmissione e di quella di circuito, per completare l’operazione. Tale scenario ben si applica al contesto di trasmissioni cellulari per applicazioni di tipo IoT nelle future reti 5G, dove l’attenzione è rivolta maggiormente all’efficienza energetica dei dispositivi rispetto al throughput, in quanto le UE hanno tipicamente capacità computazionali ed energetiche esigue e si limitano ad inviare sporadicamente pacchetti molto brevi. Vengono quindi descritte le strategie ottime di trasmissione da adottare in un contesto di accesso coordinato a seconda del livello di dettaglio sulle informazioni di canale e delle potenzialità di trasmettitore/ricevitore, illustrando i corrispondenti limiti teorici. Nell’ultima parte del lavoro vengono studiate le prestazioni asintotiche di schemi di accesso non coordinati quando si utilizzano tecniche di Multi Packet Reception (MPR) e Successive Interference Cancellation (SIC) per la risoluzione delle collisioni al ricevitore. I risultati corrispondenti, in termini di throughput, per uno scenario M2M con massive access sono infine ricavati e discussi.

Les objets perceptuels sont les unités élémentaires utilisées par le cerveau pour construire une représentation interne du monde a partir de signaux physiques, comme la lumière ou les ondes sonores. Alors que ces signaux sont d'abord traduit, par les récepteurs dans les organes périphériques, en signaux neuronaux, l'émergence d'objets perceptuels nécessite un traitement intensif dans le système nerveux central qui n'est pas encore entièrement connu. Il est intéressant de noter que les progrès récents de deep learning montrent qu'une séries d'opérations non linéaires et linéaires est très efficace pour catégoriser les objets perceptuels visuels et auditifs de la même manière que les humains. En revanche, la plupart des connaissances actuelles sur le système auditif se concentrent sur les transformations linéaires. Afin de comprendre la contribution des non-linéarités du système auditif à la perception, nous avons étudié l'encodage des sons avec une intensité croissante et une intensité décroissante dans le cortex auditif de la souris. Ces deux sons sont perçus avec une importance inégale malgré le fait qu'ils ont la même énergie physique et le même contenu spectral, un phénomène incompatible avec le traitement linéaire. En enregistrant l'activité de grandes populations corticales pour les sons montants et descendants, nous avons constaté que le cortex les encode avec des populations distinctes qui détectent des caractéristiques non linéaires, ce qui explique l'asymétrie perceptuelle. Nous avons également montré que, dans les modèles de reinforcement learning, la quantité d'activité neuronale déclenchée par un son impacte la vitesse et la stratégie d'apprentissage. Des effets très similaires ont été observés dans plusieurs taches de discrimination ou les sons provoquaient des réponses neuronales de différentes intensités. Ceci établit que les non-linéarités du système auditif ont un impact sur la perception et le comportement. Pour mieux identifier les non-linéarités qui influencent le codage des sons, nous avons ensuite enregistré l'activité d'environ 60 000 neurones échantillonnant toute la superficie du cortex auditif. Au-delà de l'organisation tonotopique à fine échelle découverte avec cet ensemble de données, nous avons identifié et quantifié 7 non-linéarités. Il est aussi intéressant de constater que différentes non-linéarités peuvent interagir entre elles d'une manière non triviale. La connaissance de ces interactions est importante pour affiner le modèle de traitement auditif. Enfin, nous nous sommes demandé si les processus non linéaires sont également importants pour l'intégration multisensorielle. Nous avons mesuré, par imagerie calcique, comment les images et les sons se combinent dans le cortex visuel et auditif. Nous n'avons trouvé aucune modulation du cortex auditif (L2/3) en réponse à des stimuli visuels. Nous avons observé que les entrées du cortex auditif dans le cortex visuel affectent les réponses visuelles concomitantes à un son. Nous avons constaté que les projections du cortex auditif au cortex visuel encode de préférence une caractéristique non linéaire particulière : l'apparition soudaine de sons fort. Par conséquent, l'activité du cortex visuel pour une image et un son fort est plus élevée que pour l'image seule ou combinée à un son faible. Ce résultat suggère que les sons forts sont pertinents du point de vue de comportement multisensoriel, peut-être pour indiquer la présence de nouveaux objets dans le champ visuel, ce qui pourrait représenter des menaces potentielles. En conclusion, nos résultats montrent que les non-linéarités sont omniprésentes dans le traitement du son par le cerveau et jouent également un rôle dans l'intégration de l'information auditive avec l'information visuelle. Il est non seulement crucial de tenir compte de ces non-linéarités pour comprendre comment se forment les représentations perceptuelles, mais aussi pour prédire l'impact de ces représentations sur le comportement
Perceptual objects are the elementary units used by the brain to construct an inner world representation of the environment from multiple physical sources, like light or sound waves. While the physical signals are first encoded by receptors in peripheral organs into neuroelectric signals, the emergence of perceptual object require extensive processing in the central nervous system which is not yet fully characterized. Interestingly, recent advances in deep learning shows that implementing series of nonlinear and linear operations is a very efficient way to create models that categorize visual and auditory perceptual objects similarly to humans. In contrast, most of the current knowledge about the auditory system concentrates on linear transformations. In order to establish a clear example of the contribution of auditory system nonlinearities to perception, we studied the encoding of sounds with an increasing intensity (up ramps) and a decreasing intensity (down ramps) in the mouse auditory cortex. Two behavioral tasks showed evidence that these two sounds are perceived with unequal salience despite carrying the same physical energy and spectral content, a phenomenon incompatible with linear processing. Recording the activity of large cortical populations for up- and down-ramping sounds, we found that cortex encodes them into distinct sets of non-linear features, and that asymmetric feature selection explained the perceptual asymmetry. To complement these results, we also showed that, in reinforcement learning models, the amount of neural activity triggered by a stimulus (e.g. a sound) impacts learning speed and strategy. Interestingly very similar effects were observed in sound discrimination behavior and could be explain by the amount of cortical activity triggered by the discriminated sounds. This altogether establishes that auditory system nonlinearities have an impact on perception and behavior. To more extensively identify the nonlinearities that influence sounds encoding, we then recorded the activity of around 60,000 neurons sampling the entire horizontal extent of auditory cortex. Beyond the fine scale tonotopic organization uncovered with this dataset, we identified and quantified 7 nonlinearities. We found interestingly that different nonlinearities can interact with each other in a non-trivial manner. The knowledge of these interactions carry good promises to refine auditory processing model. Finally, we wondered if the nonlinear processes are also important for multisensory integration. We measured how visual inputs and sounds combine in the visual and auditory cortex using calcium imaging in mice. We found no modulation of supragranular auditory cortex in response to visual stimuli, as observed in previous others studies. We observed that auditory cortex inputs to visual cortex affect visual responses concomitant to a sound. Interestingly, we found that auditory cortex projections to visual cortex preferentially channel activity from neurons encoding a particular non-linear feature: the loud onset of sudden sounds. As a result, visual cortex activity for an image combined with a loud sound is higher than for the image alone or combine with a quiet sound. Moreover, this boosting effect is highly nonlinear. This result suggests that loud sound onsets are behaviorally relevant in the visual system, possibly to indicate the presence of a new perceptual objects in the visual field, which could represent potential threats. As a conclusion, our results show that nonlinearities are ubiquitous in sound processing by the brain and also play a role in the integration of auditory information with visual information. In addition, it is not only crucial to account for these nonlinearities to understand how perceptual representations are formed but also to predict how these representations impact behavior

碩士
臺灣大學
電機工程學研究所
95
Real world luminance ranges from a few cd/m2 at dark night to over thousands of billions cd/m2 under direct sunlight. For typical sceneries, the dynamic range which is defined as the maximum to minimum luminance ratio ranges from 3 to 10 orders of magnitudes. High Dynamic Range Imaging and Video can bring about much more realistic image to the audience. In other fields such as medical imaging (CT, MRI) or space exploration imaging, any details in the original scene have too high a price to discard and High Dynamic Range Video will be a suitable tool that can assist the development in these fields. In scene-referred capturing, we try to capture the whole range of luminance in the real scene with every possible effort to preserve visual details. Preserving every detail causes a lot of bandwidth, storage spaces; and most importantly, we will need a new video/image encoding and compressing method and system. What is lacking on the path to bring High Dynamic Range video to users is High Dynamic Range encoding systems.In this research, we have conducted analysis on the problem about the encoding and compression difficulties in High Dynamic Range Video. We have proposed a backward compatible High Dynamic Range video encoding algorithm and the respective hardware implementation. The proposed algorithm takes into account the high correlation between the normal and High Dynamic Range Image.Apart from the goal of achieving high compression ratio, other aspects such as proportionality of dynamic range versus bit-rate, flexibility for users to use different TMO, dynamic range scalability and dynamic range adjustments capabilities are considered and implemented in our algorithm. In conclusion, we have proposed a novel algorithm in HDR video encoding which fulfills several scalability and flexibility requirements. In the algorithm we have proposed a new HDR-LDR prediction method, a LALEMO method and a flexible automatic multi-level prediction model. Hardware implementation proves that the algorithm is realizable. We believe that with all the above virtues and benefits, the algorithm and methods proposed could become one of the important factors in bringing HDR video to real life enjoyments.

We studied encoding of hand force and its relations with observation-related activity in macaques trained in an isometric hand-force application and recalibration task, that required to move a visual cursor on a screen toward eight peripheral targets by exerting a force on an isometric joystick, in absence or presence of an opposing force field. Monkeys also observed the result of their action in play-back, as motion of a visual cursor on the screen. This approach combined in a single experiment isometric action performance and force adjustment with observation of its consequences in the external world, also allowing to determine whether PMd neuronal populations reflected an inverse model that specified the force necessary to move in different directions a visual object, or a forward computation encoding its desired trajectory in visual space. We found that a population of PMd cells encoded the direction of dynamic force and its recalibration when the force condition changed but did not retain memory of such change, probably reflecting an adaptation rather than a learning process. Cells with observation-related activity also modulated by change in hand force were not modulated when the force conditions changed, suggesting that their activity reflected the motion of the visual cursor on the screen, therefore the consequences of force application in the visual space. These results also allow a direct comparison of the relative contribute of different populations of PMd cells with that of cells with similar activity profile in the encoding of hand force and its consequences in the parieto-frontal system.

碩士
國立臺灣大學
資訊工程學研究所
92
In the years to come, interconnect will be a challenge to surmount for deep sub-micron (DSM) technologies. The use of DSM technology increases the capacitive and inductive coupling which leads to severe crosstalk noise, more power dissipation and malfunction of the chip between neighboring wires. We propose a method to reduce crosstalk noise on buses based on dynamic coding scheme. The proposed method deals with capacitive and inductive effects at the same time by using the realistic RLC table and the segmented bus-invert method. The experimental results show that our approach reduces bus power consumption up to 7%.

碩士
國立清華大學
資訊工程學系
95
A high dynamic range (HDR) image contains a vast amount of luminance and color information at the same time. In order to improve the efficiency in transmission and storage, a proper encoding method is necessary. We propose a perception-based HDR image encoding scheme using the Huffman code to achieve a quantization error curve that is below Ferwerda’s contrast versus intensity (c.v.i). With the discrete wavelet transform (DWT), we propose a sub-band histogram-based HDR image encoding scheme. By assigning different number of bits to the low band and the high band, we show that only an average of 5 bits is required to encode the luminance data under human perceptive range. This method has more than two times quantization efficiency compared with the previous result of encoding luminance data using 10~11 bits [1]. Besides, we verify the correctness of the tone mapping results from the low resolution bit stream because the receiver may only obtain the LL band data in the SVC system. We also represent a simple tone mapping way to ease the influence of DWT-tone mapping.

碩士
國立中央大學
資訊工程學系
102
Record boundary detection plays an important role in wrapper induction and the quality of record boundary detection will affect the precision of alignment and extraction directly. Previous approaches usually focus on calculating similarity between blocksor measure tree similarity in a single page. In this paper, we analyze multiple pages that are generated by the same website. By exploring common parts and different parts in pages, we can overcome the weakness in single-page approaches. Because the computation load will increase when we deal with more pages, the proposed approach only focus on leaf nodes in DOM tree, which are about 30 percent of all nodes. We propose dynamic encoding, which can abstract leaf nodes and emphasize the regularity of every data records. With the dynamic encoding, we reduce the numberof the repeated pattern discovered. Finally, we propose the idea of landmark, which is located in the data record, and detecting the record boundary by segmenting the DOM tree. In the experiment, we evaluate the efficiencyin our approach and compare the effectivenesswith other systems.

博士
國立中央大學
資訊工程學系
107
Automatic data extraction from template pages is an essential task for data integration and analysis. Most researches focus on data extraction from list pages. The problem of data alignment for singleton pages, which contain detail information of a single item is less addressed and is more challenging. In the rst work, we propose a novel Divide-and-Conquer Alignment algorithm (DCA) that works on leaf nodes from the DOM trees of singleton pages. The idea is to detect mandatory templates via the longest increasing sub-sequence from the landmark quivalence class leaf nodes and recursively apply the same procedure to each segment divided by mandatory templates. DCA able aligns each segment efficiently and handles multi-order attribute-value pairs eeffectively with a two-pass procedure. The results on selected items, DCA outperforms TEX and WEIR 2% and 12% respectively. The improvement is more obvious in terms of full schema evaluation, with 0.95 (DCA) versus 0.63 (TEX) F1 measure, on 26 websites from TEX and ExAlg. In the second work, we propose an unsupervised full schema web data extraction via Divide-and-Conquer Alignment with Dynamic Encoding (DCADE) from either multiple list pages or singleton pages with the same template. We dene the Content Equivalence Class and Typeset Equivalence Class based on leaf node content. We then combine HTML attributes (id and class) in the paths for various levels of encoding, so that the proposed algorithm can align leaf nodes by exploring patterns at various levels from specic to general. We conducted experiments on 49 real-world websites used in TEX and ExAlg. The proposed DCADE achieved a 0.962 F1 measure for non-recordset data extraction (FD), and a 0.936 F1 measure for recordset data extraction (FS), which outperformed other page-level web data extraction methods, i.e., DCA (FD=0.660), TEX (FD=0.454 and FS=0.549), RoadRunner (FD=0.396 and FS=0.330), and UWIDE (FD=0.260 and FS=0.081).

Recent findings suggest, that perception of acoustic stimuli in the mouse auditory cortex relies on categorization of object-based representations. Local neuronal populations in L2/3 of the mouse auditory cortex reportedly exhibit a limited number (1-3) of stable modes of response, each possibly evoked by multiple complex sounds of variable acoustic features. Stimulation using linear intensity mixing of sounds evoking different response modes revealed an attractor-like dynamic of the underlying representation. These modes of response were hypothesized to represent the neural correlate of perceptual categorization. We have developed an experimental protocol enabling chronic two-photon imaging of the previously described population coding under awake conditions. Using this protocol we acquired data suggesting that the pattern of population activity underlying a mode of response, is stable during a week-long timeframe. We have also recorded the neural activity of a local subpopulation of somatostatin-positive inhibitory interneurons (SST+ INs) during abrupt changes in cortical representation. Our preliminary results suggest that local SST+ INs exhibit maximal firing when the neural correlate of a mode of response is exhibited by the surrounding population of principal cells. In addition, we observed a...