Most neural networks use a normal convolutional layer that assumes that all input pixels are valid pixels. However, pixels added to the input through padding result in adding extra information that was not initially present. This extra information can be considered invalid. Invalid pixels can also be inside the image where they are referred to as holes in completion tasks like image inpainting. In this work, we look for a method that can handle both types of invalid pixels. We compare on the same test bench two methods previously used to handle invalid pixels outside the image (Partial and Edge convolutions) and one method that was designed for invalid pixels inside the image (Gated convolution). We show that Partial convolution performs the best in image classification while Gated convolution has the advantage on semantic segmentation. As for hotel recognition with masked regions, none of the methods seem appropriate to generate embeddings that leverage the masked regions. / Master of Science / A module at the heart of deep neural networks built for Artificial Intelligence is the convolutional layer. When multiple convolutional layers are used together with other modules, a Convolutional Neural Network (CNN) is obtained. These CNNs can be used for tasks such as image classification where they tell if the object in an image is a chair or a car, for example. Most CNNs use a normal convolutional layer that assumes that all parts of the image fed to the network are valid. However, most models zero pad the image at the beginning to maintain a certain output shape. Zero padding is equivalent to adding a black frame around the image. These added pixels result in adding information that was not initially present. Therefore, this extra information can be considered invalid. Invalid pixels can also be inside the image where they are referred to as holes in completion tasks like image inpainting where the network is asked to fill these holes and give a realistic image. In this work, we look for a method that can handle both types of invalid pixels. We compare on the same test bench two methods previously used to handle invalid pixels outside the image (Partial and Edge convolutions) and one method that was designed for invalid pixels inside the image (Gated convolution). We show that Partial convolution performs the best in image classification while Gated convolution has the advantage on semantic segmentation. As for hotel recognition with masked regions, none of the methods seem appropriate to generate embeddings that leverage the masked regions.

As science and technology has developed, the manner by which drugs can be delivered has grown. This research explores an alternative method for the delivery of therapeutic compounds to the body. The basis of the study involves the application of traditional medicines on textiles. Boswellia Serrata Extract (B.S.E.) is a common traditional medicine used for curing body pains. The most common form of B.S.E. based products are creams that are applied directly to the skin. Experiments show that these creams were not suitable as a basis for applying to textile materials because the creams contain highly volatile compounds, which on drying the treated textile, post application, cause almost total loss of the B.S.E. The approach used was the application on textiles of a 'commercial' topical medicine applied as a cream for, where B.S.E. is a major constituent. Cotton woven fabric was padded with this cream and tested for washing and rubbing fastness. The presence of highly volatile substances in the topical cream resulted in a negligible amount of the medicine on the dried treated fabric. Another approach was used for the application of B.S.E. onto the textile substrate. A commercially available B.S.E. powder was applied to woven fabric using a pad mangle. Tests were carried out to validate the

Voice Over IP (VoIP) systems and transmission technologies have now become the norm for many communications applications. However, whether they are used for personal communication or priority business conferences and talks, privacy and confidentiality of the communication is of utmost priority. The present industry standard is to encrypt VoIP calls using Secure Real-time Transport Protocol (SRTP), aided by ZRTP, but this methodology remains vulnerable to traffic analysis attacks, some of which utilize the length of the encrypted packets to infer the language and spoken phrases of the conversation. Secure Real-time Padding Protocol (SRPP) is a new RTP profile which pads all VoIP sessions in a unique way to thwart traffic analysis attacks on encrypted calls. It pads every RTP or SRTP packet to a predefined packet size, adds dummy packets at the end of every burst in a controllable way, adds dummy bursts to hide silence spurts, and hides information about the packet inter-arrival timings. This thesis discusses a few practical approaches and a theoretical optimization approach to packet size padding. SRPP has been implemented in the form of a library, libSRPP, for VoIP application developers and as an application, SQRKal, for regular users. SQRKal also serves as an extensive platform for implementation and verification of new packet padding techniques.

Camouflaging is about making something invisible or less visible. Network camouflaging is about hiding certain traffic information (e.g. traffic pattern, traffic flow identity, etc.) from internal and external eavesdroppers such that important information cannot be deduced from it for malicious use. It is one of the most challenging security requirements to meet in computer networks. Existing camouflaging techniques such as traffic padding, MIX-net, etc., incur significant performance degradation when protected networks are wireless networks, such as sensor networks and mobile ad hoc networks. The reason is that wireless networks are typically subject to resource constraints (e.g. bandwidth, power supply) and possess some unique characteristics (e.g. broadcast, node mobility) that traditional wired networks do not possess. This necessitates developing new techniques that take account of properties of wireless networks and are able to achieve a good balance between performance and security. In this three-part dissertation we investigate techniques for providing network camouflaging services in wireless networks. In the first part, we address a specific problem in a hierarchical multi-task sensor network, i.e. hiding the links between observable traffic patterns and user interests. To solve the problem, a temporally constant traffic pattern, called cover traffic pattern, is needed. We describe two traf- fic padding schemes that implement the cover traffic pattern and provide algorithms for achieving the optimal energy efficiencies with each scheme. In the second part, we explore the design of a MIX-net based anonymity system in mobile ad hoc networks. The objective is to hide the source-destination relationship with respect to each connection. We survey existing MIX route determination algorithms that do not account for dynamic network topology changes, which may result in high packet loss rate and large packet latency. We then introduce adaptive algorithms to overcome this problem. In the third part, we explore the notion of providing anonymity support at MAC layer in wireless networks, which employs the broadcast property of wireless transmission. We design an IEEE 802.11-compliant MAC protocol that provides receiver anonymity for unicast frames and offers better reliability than pure broadcast protocol.

Our landfills are growing and constantly expanding with post-consumer waste in the form of cheap fast fashion garments. One way to deal with this problem is to upcycle the discarded material into new things with a higher value. There is one type of material though, that has a tendency to be forgotten about. This material is baby garments. Mother Elizabeth investigates how these discarded garments can be upcycled into womenswear fashion. The title of this work reflects the two areas investigated, baby garments and the Elizabethan era. With an aim of broadening the perspective of what type of discarded material can be upcycled into womenswear fashion, the areas of investigation are combined. Throughout the study, methods for exploring garments from the Elizabethan era were applied in order to create a reference library used as a base when making design decisions. The outcomes presented in this work aim to inspire other designers to broaden their perspective on what type of material can be used when working with upcycling. In addition to stressing the importance of using discarded textiles, this project also showcases how history can be used as a tool to make sure there is a future.

No description available.

La sécurité prouvée est une branche relativement jeune de la cryptologie dont l'objectif est d'analyser formellement le but ultime des schémas cryptographiques: la sécurité. Elle ne cherche pas à atteindre la sécurité absolue mais plutôt à identifier les conditions suffisantes, au sens de la théorie de la complexité, pour garantir la sécurité. La sécurité prouvée est en étroite relation avec les trois principaux mouvements de la cryptographie: la formalisation des notions de sécurité; la construction des schémas formellement prouvés sûrs et la recherche de nouvelles fonctionnalités pour la cryptographie. Dans cette thèse, nous abordons dans un premier temps l'analyse des notions de sécurité, à la fois pour le chiffrement asymétrique et pour le chiffrement symétrique. D'une part, nous étudions en détails les modèles d'attaque et les relations entre eux pour le chiffrement asymétrique. D'autre part, pour le chiffrement par bloc, nous mettons en évidence la relation entre la notion traditionnelle du chiffrement par bloc, i.e. permutation (super) pseudo–aléatoire, et avec la notion de base de la confidentialité, i.e. sécurité sémantique. Dans un deuxième temps, nous proposons de nouveaux schémas efficaces et prouvés sûrs pour la cryptographie asymétrique dans le modèle de l'oracle aléatoire (de nouveaux paddings pour le chiffrement et des paddings universels pour le chiffrement et la signature). Nous présentons, de plus, une nouvelle catégorie de schémas prouvés sûrs: les schémas de chiffrement sans redondance. Jusqu'à présent, la redondance était nécessaire pour prouver la sécurité. Nous proposons, dans la troisième partie, une nouvelle fonctionnalité du traçage de pirates pour la diffusion des données chiffrées: la traçabilité publique. Nous présentons aussi un schéma satisfaisant cette propriété. Ce schéma est ensuite généralisé à un schéma quasi optimal selon le critère du taux entre de la taille des données chiffrés et celle des données originelles.

Currently, there is a lack of research behind the security of Virtual Reality against fingerprinting attacks and how these affect the Quality of Experience (QoE) and Quality of Service (QoS) for a user. With practical testing in a game which implements traffic shaping methods as security defences, this thesis aims to take the first step towards changing this. Here, tests were made testing QoS and QoE of countermeasures in a VR game using the game-engine Unity. The countermeasures utilized were random padding, random delays and VPNs. The conclusion reached was that using a delay had a significant impact on QoE, creating a high Round-Trip Time, while changing the packet size had minimal impact to both QoE and QoS. Additionally, utilizing a VPN yielded a minimal impact to both the QoE and QoS.

The system capacity of the direct-sequence code division multiple access (DS-CDMA) system is limited mainly due to the multiple access interference (MAI), this is basically due to the incomplete orthogonality of spreading codes between different users. In wireless communication environments, the use of DS-CDMA system over multipath channels will introduce the effect of inter-symbol interference (ISI), thus the system performance might degrade, dramatically. To circumvent the above-mentioned problems many adaptive multiuser detectors are proposed, such as the minimum mean square error (MMSE) criteria subject to certain constraints. Unfortunately, with the MMSE receiver it requires an extra training sequence, which decreases the spectral efficiency. To increase the spectral efficiency, the blind adaptive receivers are adopted. In the conventional approach the blind adaptive receiver is developed based on the linear constrained minimum variance (LCMV) criteria, which can be viewed as the constrained version of the minimum output energy (MOE) criteria. Other alternative of designing the blind adaptive receiver is to use the linear constrained constant modulus (LCCM) criteria. In general, the LCCM receiver could achieve better robustness due to the changing environment of channel. With the above-mentioned adaptive linearly constrained multi-user receivers, we are able to reduce the effects of ISI and MAI and achieve desired system performance. However, for worse communication link, the conventional adaptive multi-user detector might not achieve desired performance and suppress interference effectively. In this thesis, we consider a new approach, in which the pre-coder similar to the Orthogonal Frequency Division Multiplexing (OFDM) systems is introduced in the transmitter of the DS-CDMA system. In the receiver, by using the characteristics of pre-coder we could remove the effect of ISI, effectively, and follows by the adaptive multi-user detector to suppress the MAI. Two most common use pre-coders of the OFDM systems are the Cyclic Prefix (CP) or Zero Padding (ZP). Thus the pre-coded DS-CDMA systems associated with the adaptive blind linearly constrained receiver could be employed to further improve the system performance with the cost of decreasing the spectral efficiency.

Gegenstand der vorliegenden Arbeit ist die Untersuchung des Foulardierprozesses zur chemischen Nassausrüstung von Vliesstoffen bei Warengeschwindigkeiten bis zu 250 m/min. Hintergrund ist die abweisende Ausrüstung von Polypropylen-Spinnvliesstoffen für die Anwendung als Operationskittel. Wo bislang nach dem Stand der Technik eine Veredlung bei Lohnausrüstern bei geringen Warengeschwindigkeiten durchgeführt wurde, zeigt die Tendenz in der Vliesstoffindustrie in Richtung der eigenen Prozessbeherrschung. Eine grundlegende Voraussetzung, um den Foulardierprozess für diese Anwendung nutzbar zu machen, ist die Kenntnis über die Prozesseigenschaften bei den geforderten hohen Warengeschwindigkeiten. Für den abzudeckenden Versuchsraum mit sechs Einflussgrößen bei jeweils drei Faktorstufen wurde mittels der Methodik der statistischen Versuchsplanung ein D-optimaler Versuchsplan erstellt. Die Versuchsdurchführung erfolgte auf einem in eine Technikumsanlage eingebundenen Foulard mit horizontaler Walzenanordnung. Für jede der sieben Zielgrößen wurde auf Grundlage der erhaltenen Messwerte eine lineare Regressionsanalyse erstellt und ausgewertet. Eine detaillierte Analyse und Diskussion der Regressionsmodelle liefert Informationen zu Wirkungsrichtung und Intensität der einzelnen Einflussgrößen sowie zu Faktor-Faktor-Wechselwirkungen. / The subject of the work presented here is the study of the padding process for the chemical wet finishing of nonwovens at web speeds up to 250 m/min. Background to the topic is the repellent treatment of polypropylene spunbond nonwovens applied for surgical gowns. Finishing carried out at subcontractors corresponding to best practice technology up to now, the trend in the nonwovens industry is turning towards an in-house process mastery. Essential requirement to make the padding process technologically exploitable for this kind of application is the knowledge of the process characteristics at the high web speeds claimed. For the experimental scenario to be covered comprising six determining factors at three level steps each, a D-optimal trial plan was defined using the statistic method of the design of experiments (DOE). The realization of the trials carried out on a padder with horizontal roll arrangement installed in a pilot line. For each of the seven responses a linear regression analyses was compiled and evaluated. A detailed analysis and discussion of the regression models provides information on direction of influence as well as intensity of each of the determining factors and factor-factor-interactions.