This dissertation investigates four methods for attacking stream ciphers that are based on nonlinear combining generators: -- Two exhaustive-search correlation attacks, based on the binary derivative and the Lempel-Ziv complexity measure. -- A fast-correlation attack utilizing the Viterbi algorithm -- A decimation attack, that can be combined with any of the above three attacks. These are ciphertext-only attacks that exploit the correlation that occurs between the ciphertext and an internal linear feedback shift-register (LFSR) of a stream cipher. This leads to a so-called divide and conquer attack that is able to reconstruct the secret initial states of all the internal LFSRs within the stream cipher. The binary derivative attack and the Lempel-Ziv attack apply an exhaustive search to find the secret key that is used to initialize the LFSRs. The binary derivative and the Lempel-Ziv complexity measures are used to discriminate between correct and incorrect solutions, in order to identify the secret key. Both attacks are ideal for implementation on parallel processors. Experimental results show that the Lempel-Ziv correlation attack gives successful results for correlation levels of p = 0.482, requiring approximately 62000 ciphertext bits. And the binary derivative attack is successful for correlation levels of p = 0.47, using approximately 24500 ciphertext bits. The fast-correlation attack, utilizing the Viterbi algorithm, applies principles from convolutional coding theory, to identify an embedded low-rate convolutional code in the pn-sequence that is generated by an internal LFSR. The embedded convolutional code can then be decoded with a low complexity Viterbi algorithm. The algorithm operates in two phases: In the first phase a set of suitable parity check equations is found, based on the feedback taps of the LFSR, which has to be done once only once for a targeted system. In the second phase these parity check equations are utilized in a Viterbi decoding algorithm to recover the transmitted pn-sequence, thereby obtaining the secret initial state of the LFSR. Simulation results for a 19-bit LFSR show that this attack can recover the secret key for correlation levels of p = 0.485, requiring an average of only 153,448 ciphertext bits. All three attacks investigated in this dissertation are capable of attacking LFSRs with a length of approximately 40 bits. However, these attacks can be extended to attack much longer LFSRs by making use of a decimation attack. The decimation attack is able to reduce (decimate) the size of a targeted LFSR, and can be combined with any of the three above correlation attacks, to attack LFSRs with a length much longer than 40 bits. / Dissertation (MEng (Electronic Engineering))--University of Pretoria, 2007. / Electrical, Electronic and Computer Engineering / unrestricted

No description available.

The progess in NMRQIP can be outlined in to four parts.1) Implementation of theoretical protocols on small number of qubits. 2) Demonstration of QIP on various NMR systems. 3) Designing and implementing the algorithms for mixed initial states. 4) Developing the techniques for coherent and decoherent control on higher number(up to 15) of qubits. This thesis contains some efforts in the direction of first three points. Quantum-state discrimination has important applications in the context of quantum communication and quantum cryptography. One of the characteristic features of quantum mechanics is that it is impossible to devise a measurement that can distinguish nonorthogonal states perfectly. However, one can distinguish them with a finite probability by an appropriate measurement strategy. In Chapter 2, we describe the implementation of a theoretical protocol of programmable quantum-state discriminator, on a two-qubit NMR System. The projective measurement is simulated by adding two experiments. This device does the unambiguous discrimination of a pair of states of the data qubit that are symmetrically located about a fixed state. The device is used to discriminate both linearly polarized states and eillipitically polarized states. The maximum probability of successful discrimination is achieved by suitably preparing the ancilla quubit. The last step of any QIP protocol is the readout. In NMR-QIP the readout is done by using density matrix tomography. It was first proposed by Ernst and co-workers that a two-dimensional method can be used to correlate input and output states. This method uses an extra (aniclla) qubit, whose transitions indicate the quantum states of the remaining qubits. The 2D spectrum of ancilla qubit represent the input and output states along F1 and F2 dimensions respectively. However the 2D method requires several t1 increments to achieve the required spectral width and resolution in the indirect dimension, hence leads to large experimental time. In chapter 3, the conventional 2D NMRQIP method is speeded-up by using Hadamard spectroscopy. The Hadamard method is used to implement various two-, three-qubit gates and qutrit gates. We also use Hadamard spectroscopy for information storage under spatial encoding and to implement a parallel search algorithm. Various slices of water sample can be spatially encoded by using a multi-frequency pulse under the field gradient. Thus the information of each slice is projected to the frequency space. Each slice represents a classical bit, where excitation and no excitation corresponds to the binary values 0 and 1 respectively. However one has to do the experiment for each binary information, by synthesizing a suitable multi-frequency pulse. In this work we show that by recording the data obtained by various Hadamard encoded multi-frequency pulses, one can suitably decode it to obtain any birnary information, without doing further experiments. Geometric phases depend only on the geometry of the path executed in the projective Hilbert space, and are therefore resilient to certain types of errors. This leads to the possibility of an intrinsically fault-tolerant quantum computation. In liquid state NMRQIP. Controlled phase shift gates are achieved by using qubit selective pulses and J evolutions, and also by using geometir phases. In order to achieve higher number of qubits in NMR, one explores dipolar couplings which are larger in magnitude, yielding strongly coupled spectra. In such systems since the Hamiltonian consists of terms, it is difficult to apply qubit selective pulses. However such systems have been used for NMRQIP by considering 2n eigen states as basis states of an n-qubit system. In chapter 4, it is shown that non-adiabatic geometric phases can be used to implement controlled phase shift gates in strongly dipolar coupled systems. A detailed theoretical explanation of non-adiabatic geometric phases in NMR is given, by using single transition operators. Using such controlled phase shift gates, the implementation of Deutsch-Jozsa and parity algorithms are demonstrated. Search algorithms play an important role in the filed of information processing. Grovers quantum search algorithm achieves polynomial speed-up over the classical search algorithm. Bruschweiler proposed a Liouville space search algorithm which achieve polymonial speed-up. This algorithm requires a weakly coupled system with a mixed initial state. In chapter 5 we modified the Bruschweiler’s algorithm, so that it can be implemented on a weakly as well as strongly coupled system. The experiments are performed on a strongly dipolar coupled four-qubit system. The experiments from four spin-1/2 nuclei of a molecule oriented in a liquid crystal matrix. Chapter 6 describes the implementation of controlled phase shift gates on a quadrupolar spin-7/2 nucleus, using non-adiabatic geometric phases. The eight energy levels of spin-7/2 nucleus, form a three qubit system. A general procedure is given, for implementing a controlled phase shift gate on a system consisting of any number of energy levels. Finally Collin’s version of three-qubit DJ algorithm using multi-frequency pulses, is implemented in the spin-7/2 system.

Le travail de thèse a consisté, en priorité, à s’approprier les technologies d’intégration verticale en usage dans l’industrie pour réaliser des mémoires à plusieurs étages, et à en évaluer l’apport pour les capteurs à pixel CMOS (CPS). Cette approche s’appuie sur la capacité de l’industrie à interconnecter des puces amincies empilées les unes sur les autres. Elle ouvre la perspective d’associer plusieurs microcircuits superposés à un même pixel, en dépits de sa taille réduite. L’interconnexion est donc réalisée au niveau du pixel. Ce saut technologique permet de lever la majorité des obstacles à l’obtention de performances optimales des CPS. On peut en particulier combiner des puces réalisées dans des technologies CMOS très différentes, chacune optimale pour une fonctionnalité précise. La collection des charges du signal peut ainsi être réalisée dans une couche dédiée, les microcircuits de conditionnement analogique des signaux peuvent être concentrés dans une autre couche, une troisième couche pouvant héberger les parties numériques assurant la compression puis la transmission des signaux, etc. Ce progrès se traduit notamment par la possibilité de combiner haute résolution spatiale et lecture rapide, avec une amélioration probable de la tolérance aux rayonnements intenses.On s’affranchit de cette manière des limitations provenant des paramètres de fabrication des fondeurs, qui ne permettent pas à l’heure actuelle, de pleinement exploiter le potentiel des CPS à l’aide d’une technologie CMOS unique. / The thesis has been a priority as taking ownership of vertical integration technologies used in the industry to realize a multistage development, and to evaluate the contributions on CMOS pixel sensors (CPS). 3D integration technologies (3DIT) provide a way to mitigate this hampering correlation between speed and resolution, since they allow to staple layers of readout circuitry on top of the sensing layer, which results in a drastic increase of the functionalities located in (the shadow of) each pixel. A multi-layer structure allows for a higher spatial resolution because more and more transistors may be integrated vertically in a relatively small pixel. Moreover, bringing the components of the sensor closer to each other translates in a faster readout, owing to the reduction in the average length of the inner connecting wires. Vertical integration also opens up the possibility of combining different technologies best suited to each of the sensor main functionalities (signal sensing, analog and digital signal processing and transmission). It overcomes the limitations in this way from the foundry manufacturing parameters, which do not allow to fully exploit the potential ofCPS with a single CMOS technology. 3D-CPS are thus expected to overcome most of the limitations of standard 2DCPS, and are therefore suspected to over new perspectives for the innermost layer of the ILC vertex detector.

Ce travail contribue au programme de recherche et de développement d'un capteur CMOS à pixel qui pourrait satisfaire pleinement les spécifications du nouvel ITS (Inner Tracking System : trajectomètre interne) de l'expérience ALICE. Afin de briser les limites de la CPS de pointe, une technologie CMOS 0.18 µm à quatre puits a été explorée. Les capteurs fabriqués dans cette nouvelle technologie ont montré une meilleure tolérance aux radiations que les capteurs réalisés dans une technologie CMOS 0.35 µm plus ancienne. En outre, cette nouvelle technologie offre la possibilité d’implémenter des transistors de type P dans chaque pixel sans dégrader la capacité de collection de la diode. Il devient donc possible d’intégrer un discriminateur dans chaque pixel et obtenir un pixel à sortie binaire. En conséquence, la consommation sera largement réduite. De plus, le temps de traitement de la ligne peut être potentiellement réduit. Un premier prototype de petite taille, intitulé AROM-0, a été conçu et fabriqué afin d’étudier la faisabilité de la discrimination de signal dans un petit pixel. Dans ce prototype, chaque pixel de surface 22 × 33 µm2 contient une diode de détection, un préamplificateur et un discriminateur à tension d’offset compensée. La performance de bruit des différentes versions de pixels dans le capteur AROM-0 a été évaluée. Ensuite sera détaillé le développement des capteurs AROM-1. Ce sont les capteurs intermédiaires vers le capteur final proposé par notre groupe. Ils ont deux objectifs principaux, l’un est de valider les optimisations de conception du pixel et l’autre est de mettre en place une architecture du capteur évolutive intégrant l’intelligence nécessaire dans le circuit. Cette thèse présente en détail la conception et les résultats de mesure de ces capteurs AROM. / This work is part of the R&D program aimed for a CMOS pixel sensor (CPS) complying with the requirements of the upgrade of the inner tracking system (ITS) of the ALICE experiment. In order break the limitations of the state-of-the-art CPS, a 0.18 µm quadruple-well CMOS process was explored. Besides an enhanced radiation tolerance, with respect to the former sensors fabricated in a 0.35 µm process, the sensor based on this new process allows for full CMOS capability inside the pixel without degradation of the detection efficiency. Therefore, the signal discrimination, which was formerly performed at the column level, can be integrated inside the pixel. As a result, the readout speed and power consumption can be greatly improved as compared to the CPS with column-level discrimination. This work addresses the feasibility study of achieving the signal discrimination withina small pixel (i.e. 22 × 33 µm2), via the prototype named AROM-0. The pixel of AROM-0 contains a sensing diode, a pre-amplifier and an offset compensated discriminator. The noise performance of the various pixel versions implemented in AROM-0 was evaluated. The study was further pursued with the AROM-1 prototypes, incorporating the optimized pixel designs and the necessary on-chip intelligence to approach the final sensor we have proposed for the ALICE-ITS upgrade. This thesis presents in detail the design and the measurement results of these AROM sensors.

No description available.

El presente proyecto de investigación muestra el estudio realizado en una red de transporte microondas de un operador local de telecomunicaciones que aplica el uso de la polarización cruzada o también llamado polarización co-canal con el fin de verificar que existe una duplicidad del ancho de banda y por ende un mejoramiento en la red de transporte, a través del manejo de aplicación de XPIC. Para verificar el mejoramiento de un enlace, hay que considerar ciertos parámetros de radio que nos ayudaran a determinar el comportamiento del mismo, como son el nivel de XPD, nivel de RSL, Margen de desvanecimiento y disponibilidad del enlace. Dentro del estudio se está considerando los factores externos que afectan a un enlace microondas con el uso de XPIC como fallas en instalación, climas por región, obstrucciones en afectación de línea de vista, así como también estudiaremos la parte de simulación con los parámetros de radio involucrados que podrían afectar a poder duplicar la capacidad del enlace, y se mostrará un caso real para verificar el tráfico. This research project shows a study in a network of microwave complimentary local telecommunications operator that applies the use of cross polarization also called co-channel polarization in order to verify that there is a duplication of bandwidth and hence an improvement in the transport network, by managing application XPIC. To verify the improvement of a link, consider certain parameters within which we help to determine the behavior of the same, as are the level of XPD, level RSL, fade margin and link availability. Inside the studio is considering external factors affecting a microwave link using XPIC as faulty installation, climates region affectation obstructions in line of sight, the radio parameters were studied in the part that involved simulation could affect the ability to double bond, and show a real case to verify traffic.

Artificial intelligence (AI) has been seeing a great amount of hype around it for a few years but more so now in the field of diagnostic medical imaging. AI-based diagnoses have shown improvements in detecting the smallest abnormalities present in tumors and lesions. This can tremendously help public healthcare. There is a large amount of data present in the field of biomedical imaging with the hospitals but only a small amount is available for the use of research due to data and privacy protection. The task of manually segmenting tumors in this magnetic resonance imaging (MRI) can be quite expensive and time taking. This segmentation and classification would need high precision which is usually performed by medical experts that follow clinical medical standards. Due to this small amount of data when used with machine learning models, the trained models tend to overfit. With advancing deep learning techniques it is possible to generate images using Generative Adversarial Networks (GANs). GANs has garnered a heap of attention towards itself for its power to produce realistic-looking images, videos, and audios. This thesis aims to use the synthetic images generated by progressive growing GANs (PGGAN) along with real images to perform segmentation on brain tumor MRI. The idea is to investigate whether the addition of this synthetic data improves the segmentation significantly or not. To analyze the quality of the images produced by the PGGAN, Multi-scale Similarity Index Measure (MS-SSIM) and Sliced Wasserstein Distance (SWD) are recorded. To exam-ine the segmentation performance, Dice Similarity Coefficient (DSC) and accuracy scores are observed. To inspect if the improved performance by synthetic images is significant or not, a parametric paired t-test and non-parametric permutation test are used. It could be seen that the addition of synthetic images with real images is significant for most cases in comparison to using only real images. However, this addition of synthetic images makes the model uncertain. The models’ robustness is tested using training-free uncertainty estimation of neural networks.

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.