To enhance the protection against cyberattacks, significant research is directed towards anomaly-based host intrusion detection systems (HIDS), which particularly appear suited for detecting zero-day attacks. This thesis addresses two problems in HIDS training sets that are often neglected in other publications: unclean and incomplete data. First, using the Leipzig Intrusion Detection - Data Set (LID-DS), a methodology to measure HIDS robustness against contaminated training data is presented. Furthermore, three baseline HIDS approaches (STIDE, SCG, and SOM) are evaluated, and robustness improvements are proposed for them. The results indicate that the baselines are not robust if test and training data share identical attacks. However, the suggested modifications, particularly the removal of anomalous threads from the training set, can enhance robustness significantly. For the problem of incomplete training data, the thesis leverages machine learning models to predict a training set’s suitability, quantified by either data drift measures or the STIDE performance. The thesis then presents rules, extracted from the best models, for assessing the suitability of new training data. Given the practical significance of both issues, for contaminated training data emphasized by the results, further research is essential. This involves examining the robustness of other HIDS algorithms, refining the proposed robustness improvements, and validating the suitability rules on other datasets, preferably real-world data.

Selection strategies to increase productive traits of farm animals have been effective and highly specialized breeds and strains were obtained. At the same time, the effort made to obtain extremely high producing animals was accompanied by undesired effects, such as a reduced ability to sustain production, reproduction and health; especially under constrained conditions. The perception that selection was degrading robustness, lead to selection strategies aiming to improve the ability of animals to perform in a wider range of environmental constrain. However, at the present moment, the physiological mechanisms allowing farm animals to perform well in a wide range of environments, while others succumb, have not been described. The present thesis intended to address this question by describing the evolution of traits related to fitness, survival and to the adaptability to environmental constraints. Two maternal rabbit lines differing in their ability to face the environmental constraints, i.e. a `specialist¿ and a `generalist¿ maternal rabbit line were available. Additionally, two generations (20 generations apart) of the specialized line were simultaneously available. During the first two consecutive reproductive cycles, female rabbits were simultaneously subjected to three environmental conditions differing in the intensity and in the physiological constrain imposed. Digestive capacity, the acquisition of resources and the partitioning of resources into different function (i.e. litter size, milk yield, growth, body reserves, etc.) was also assessed. Results showed a greater acquisition capacity of `generalist¿ females in constrained conditions with respect to `specialist¿ females. Moreover, the greater acquisition capacity was not accompanied by a reduction in the digestive efficiency, allowing the `generalist¿ females a relative greater acquisition of digestible energy. The maintenance of reproductive performance by having a greater acquisition capacity, together with the avoidance of making an intensive use of body reserves were both related to the capacity of `generalist¿ females to sustain reproduction in a wide range of environmental conditions. Twenty generations of selection exclusively for reproduction (specialized line), was not accompanied by a higher acquisition capacity, but by a change in the relative priority between the litter being nursed (actual) and the litter being gestate (future litter). In this sense, females from the actual generation of selection for litter size at weaning had a greater milk yield in the first week of lactation (period of great importance to kits survival), reducing it by the end of lactation. The present thesis also evidenced the importance of the environment where the animals are being selected in the evolution of the interplay between competing functions. / Savietto, D. (2014). Environmental and genetic factors driving robustness in reproductive rabbit does [Tesis doctoral]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/37198 / Premios Extraordinarios de tesis doctorales

Tesis por compendio / Despite the continuous evolution in computers and information technology, real-world combinatorial optimization problems are NP-problems, in particular in the domain of planning and scheduling. Thus, although exact techniques from the Operations Research (OR) field, such as Linear Programming, could be applied to solve optimization problems, they are difficult to apply in real-world scenarios since they usually require too much computational time, i.e: an optimized solution is required at an affordable computational time. Furthermore, decision makers often face different and typically opposing goals, then resulting multi-objective optimization problems. Therefore, approximate techniques from the Artificial Intelligence (AI) field are commonly used to solve the real world problems. The AI techniques provide richer and more flexible representations of real-world (Gomes 2000), and they are widely used to solve these type of problems. AI heuristic techniques do not guarantee the optimal solution, but they provide near-optimal solutions in a reasonable time. These techniques are divided into two broad classes of algorithms: constructive and local search methods (Aarts and Lenstra 2003). They can guide their search processes by means of heuristics or metaheuristics depending on how they escape from local optima (Blum and Roli 2003). Regarding multi-objective optimization problems, the use of AI techniques becomes paramount due to their complexity (Coello Coello 2006). Nowadays, the point of view for planning and scheduling tasks has changed. Due to the fact that real world is uncertain, imprecise and non-deterministic, there might be unknown information, breakdowns, incidences or changes, which become the initial plans or schedules invalid. Thus, there is a new trend to cope these aspects in the optimization techniques, and to seek robust solutions (schedules) (Lambrechts, Demeulemeester, and Herroelen 2008). In this way, these optimization problems become harder since a new objective function (robustness measure) must be taken into account during the solution search. Therefore, the robustness concept is being studied and a general robustness measure has been developed for any scheduling problem (such as Job Shop Problem, Open Shop Problem, Railway Scheduling or Vehicle Routing Problem). To this end, in this thesis, some techniques have been developed to improve the search of optimized and robust solutions in planning and scheduling problems. These techniques offer assistance to decision makers to help in planning and scheduling tasks, determine the consequences of changes, provide support in the resolution of incidents, provide alternative plans, etc. As a case study to evaluate the behaviour of the techniques developed, this thesis focuses on problems related to container terminals. Container terminals generally serve as a transshipment zone between ships and land vehicles (trains or trucks). In (Henesey 2006a), it is shown how this transshipment market has grown rapidly. Container terminals are open systems with three distinguishable areas: the berth area, the storage yard, and the terminal receipt and delivery gate area. Each one presents different planning and scheduling problems to be optimized (Stahlbock and Voß 2008). For example, berth allocation, quay crane assignment, stowage planning, and quay crane scheduling must be managed in the berthing area; the container stacking problem, yard crane scheduling, and horizontal transport operations must be carried out in the yard area; and the hinterland operations must be solved in the landside area. Furthermore, dynamism is also present in container terminals. The tasks of the container terminals take place in an environment susceptible of breakdowns or incidences. For instance, a Quay Crane engine stopped working and needs to be revised, delaying this task one or two hours. Thereby, the robustness concept can be included in the scheduling techniques to take into consideration some incidences and return a set of robust schedules. In this thesis, we have developed a new domain-dependent planner to obtain more effi- cient solutions in the generic problem of reshuffles of containers. Planning heuristics and optimization criteria developed have been evaluated on realistic problems and they are applicable to the general problem of reshuffling in blocks world scenarios. Additionally, we have developed a scheduling model, using constructive metaheuristic techniques on a complex problem that combines sequences of scenarios with different types of resources (Berth Allocation, Quay Crane Assignment, and Container Stacking problems). These problems are usually solved separately and their integration allows more optimized solutions. Moreover, in order to address the impact and changes that arise in dynamic real-world environments, a robustness model has been developed for scheduling tasks. This model has been applied to metaheuristic schemes, which are based on genetic algorithms. The extension of such schemes, incorporating the robustness model developed, allows us to evaluate and obtain more robust solutions. This approach, combined with the classical optimality criterion in scheduling problems, allows us to obtain, in an efficient in way, optimized solution able to withstand a greater degree of incidents that occur in dynamic scenarios. Thus, a proactive approach is applied to the problem that arises with the presence of incidences and changes that occur in typical scheduling problems of a dynamic real world. / Rodríguez Molins, M. (2015). Optimization and Robustness in Planning and Scheduling Problems. Application to Container Terminals [Tesis doctoral]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/48545 / Compendio

Wide-bandgap gallium nitride (GaN) high electron mobility transistors (HEMTs) are gaining increased adoption in applications like mobile electronics and data centers. Benefitting from the high channel mobility and the high breakdown field of GaN, GaN power HEMTs enable low specific on-resistance and small capacitance and thus become attractive for high-frequency applications. In addition, most commercial GaN power HEMTs are fabricated on Si substrates up to 8 inches, allowing for a remarkable cost advantage. However, a by-product of the low-cost GaN-on-Si wafer (and conductive Si substrate) is the high voltage drop and high electric field (E-field) in the GaN buffer layers and transition layers sandwiched between the GaN channel and Si substrate. To boost the vertical blocking capability and minimize the leakage current, the GaN buffer layer is usually doped with carbon or iron, which can introduce complex carrier traps. This can further lead to the dynamic shifts of various parameters in GaN-on-Si HEMTs, which can cause their stability and robustness issues in practical circuit operations. This dissertation work studies the robustness and stability of GaN power HEMTs in dynamic power switching. The structures of most GaN power devices are fundamentally different from Si or Silicon Carbide (SiC) power devices, leading to numerous open questions on GaN power device robustness and stability. Simple equipment-level static characterization may not reflect the real device characteristics in circuit-level operation. Based on the relevance between the stress condition and the device's safe operating area (SOA), this dissertation is divided into two parts. In each part, two representative GaN power devices, the standalone GaN HEMT, and the GaN-Si cascode HEMT, are studied. The dissertation's first half discusses the GaN HEMT behavior outside of SOA, with a focus on the robustness of GaN HEMTs in overvoltage power switching. This focus is motivated by the lack of avalanche capability of GaN HEMTs, which is a unique device physics distinct from SiC/Si power transistors. Instead of withstanding the surge energy through avalanching, GaN HEMTs rely on their high breakdown voltage margin to withstand the surge energy, which can trigger new degradation and failure mechanisms. Therefore, investigating the GaN HEMTs' robustness in overvoltage switching is of great interest. The robustness study begins with a standalone depletion-mode (D-mode) MIS (Metal-Insulator-Semiconductor) HEMT in an overvoltage hard-switching. The device is found to show a decreased threshold voltage and increased saturation current after stress. These parametric shifts increase as switching cycles increase but reach a saturation point before one million cycles. The root cause is believed to be the impact-ionization-generated holes trapped underneath the insulated gate. This is verified by the physics-based TCAD (Technology Computer-Aided Design) simulation. After the stress, MIS-HEMT cannot fully recover naturally. Applying at positive gate-to-source bias (VGS) is found to be able to accelerate the threshold voltage recovery but not the saturation current recovery, while a 50-V substrate bias is shown to fully recover both parameters. These findings provide new insight into the hole trapping/de-trapping dynamics and the benefits of substrate voltage control in GaN MIS-HEMTs. Then, a cascode GaN HEMT, which contains a D-mode GaN MIS-HEMT and an enhancement-mode (E-mode) Si MOSFET, is studied similarly in overvoltage stress produced by an inductive switching circuit. Parametric shifts are found in cascode GaN HEMTs, including the unstable breakdown voltage and increased on-resistance. The crosstalk between Si MOSFET and GaN HEMT is believed to account for these parametric shifts. A decapsulated device is developed based on the commercial part to monitor the Si MOSFET behavior. Si MOSFET is found to avalanche during the overvoltage switching. The parametric shifts are believed to be due to the avalanche-generated electrons, which are injected into the GaN HEMTs and trapped in the GaN buffer layer. These electron traps alter the E-field distribution of the GaN HEMT and induce parametric shifts. The second half of the dissertation focuses on the GaN HEMT's stability inside the SOA, with a focus on the non-ideal power loss generated in high-frequency switching. The output capacitance (COSS) loss has recently been found to be the dominant loss in soft switching, which is the loss associated with GaN HEMT's COSS when it is charged and discharged. This process should be lossless for an ideal capacitor, but GaN HEMT experiences a hysteresis COSS loss during each charging-discharging cycle due to the COSS instability in dynamic power switching. The COSS loss study starts with an accurate and easy-to-implement test platform, which is proven to have good robustness and repeatability. The measured COSS loss of different types of GaN HEMTs is modeled, followed by the investigation of the COSS loss origin. TCAD simulation reveals the fundamental role of trappings in the cause of COSS loss in standalone GaN HEMTs. For the cascode GaN HEMT, two additional loss mechanisms are involved as compared to the standalone GaN HEMTs: Si avalanche energy loss and GaN early turn-on loss. This makes cascode GaN HEMT experiences much higher COSS loss than standalone GaN HEMTs. The COSS loss of cascode GaN HEMT is quantitively analyzed, and a mitigation strategy is proposed for suppressing the COSS loss of cascode GaN HEMTs. Then, a circuit-level method is proposed to reduce the COSS loss of standalone GaN HEMT by dynamically tuning the substrate bias, which is verified with a standalone D-mode GaN HEMT. The Si substrate bias can follow the drain voltage in a certain ratio by tuning the capacitance ratio between the drain, substrate, and source. It is found that with a substrate bias of 1/4 to 1/2 of the drain voltage, the COSS loss can be reduced by 86%. This result removes a critical roadblock for deploying GaN HEMTs in high-frequency, soft-switching applications. Finally, the COSS loss of similarly rated Si and SiC power transistors is characterized using the developed test platform. The capability of the setup is further broadened to testing power diodes. Some similarities and distinctions are found in the COSS loss behavior between GaN HEMTs and Si/SiC devices. Also, an EDISS validation process is provided for the UIS-based method in an operating class-E converter, verifying the effectiveness and accuracy of the proposed method. This provides important references for selecting the optimal power devices for high-frequency applications. / Doctor of Philosophy / Gallium Nitride (GaN) high electron mobility transistors (HEMTs) are reshaping the power electronics field. They have become increasingly popular in many applications like smartphones, electric vehicles, and data centers. They offer smaller on-resistance and can handle higher voltages compared to traditional silicon-based devices. GaN transistors are built on large-diameter silicon substrates, making them cost-effective but can lead to unique stability and robustness issues. This dissertation investigates the stability and robustness of GaN power HEMTs in high-voltage and high-frequency power switching. Based on the relevance of the studied stress to the device safe-operating-area, the discussion is divided into two parts: The first part looks at how GaN transistors handle situations where they are pushed beyond their safe operating limits, such as during power surges and overvoltage events. These transistors are found to experience changes in their electrical properties after being stressed, which might affect their performance across their lifetime. In addition to unveiling the physics and evolution of such parametric shifts, this work also discovers ways to recover the device parameters and maintain the device functionality. The second part of the research focuses on the stability and non-ideal power loss of GaN transistors within their safe operating area. The high-frequency soft-switching application is being investigated, as it has become a common trend for future power electronics. The study reveals that GaN transistors can produce additional power loss due to the intrinsic electrical instabilities. In addition to unveiling the key impact factors and physics of this loss, this work also develops device designs to suppress this non-ideal power loss significantly, improving the device efficiency in high-frequency applications. Overall, this work provides valuable insights into improving the robustness and efficiency of GaN transistors, which provide guidelines and insights for GaN designers and users to achieve optimal device and system performance.

In data science, machine learning methods, such as deep learning and other AI algorithms, have been widely used in many applications. These machine learning methods often have complicated model structures with a large number of model parameters and a set of hyperparameters. Moreover, these machine learning methods are data-driven in nature. Thus, it is not easy to provide a comprehensive evaluation on the performance of these machine learning methods with respect to the data quality and hyper-parameters of the algorithms. In the statistical literature, design of experiments (DoE) is a set of systematical methods to effectively investigate the effects of input factors for the complex systems. There are few works focusing on the use of DoE methodology for evaluating the quality assurance of AI algorithms, while an AI algorithm is naturally a complex system. An understanding of the quality of Artificial Intelligence (AI) algorithms is important for confidently deploying them in real applications such as cybersecurity, healthcare, and autonomous driving. In this proposal, I aim to develop a set of novel methods on the interface between experimental design and machine learning, providing a systematical framework of using DoE methodology for AI algorithms. This proposal contains six chapters. Chapter 1 provides a general introduction of design of experiments, machine learning, and surrogate modeling. Chapter 2 focuses on investigating the robustness of AI classification algorithms by conducting a comprehensive set of mixture experiments. Chapter 3 proposes a so-called Do-AIQ framework of using DoE for evaluating the AI algorithm’s quality assurance. I establish a design-of-experiment framework to construct an efficient space-filling design in a high-dimensional constraint space and develop an effective surrogate model using additive Gaussian process to enable the quality assessment of AI algorithms. Chapter 4 introduces a framework to generate continual learning (CL) datsets for cybersecurity applications. Chapter 5 presents a variable selection method under cumulative exposure model for time-to-event data with time-varying covariates. Chapter 6 provides the summary of the entire dissertation. / Doctor of Philosophy / Artificial intelligence (AI) techniques, including machine learning and deep learning algorithms, are widely used in various applications in the era of big data. While these algorithms have impressed the public with their remarkable performance, their underlying mechanisms are often highly complex and difficult to interpret. As a result, it becomes challenging to comprehensively evaluate the overall performance and quality of these algorithms. The Design of Experiments (DoE) offers a valuable set of tools for studying and understanding the underlying mechanisms of complex systems, thereby facilitating improvements. DoE has been successfully applied in diverse areas such as manufacturing, agriculture, and healthcare. The use of DoE has played a crucial role in enhancing processes and ensuring high quality. However, there are few works focusing on the use of DoE methodology for evaluating the quality assurance of AI algorithms, where an AI algorithm can be naturally considered as a complex system. This dissertation aims to develop innovative methodologies on the interface between experimental design and machine learning. The research conducted in this dissertation can serve as practical tools to use DoE methodology in the context of AI algorithms.

Computer vision algorithms, including image classifiers and object detectors, play a pivotal role in various cyber-physical systems, spanning from facial recognition to self-driving vehicles and security surveillance. However, the emergence of real-world adversarial patches, which can be as simple as stickers, poses a significant threat to the reliability of AI models utilized within these systems. To address this challenge, several defense mechanisms such as PatchGuard, Minority Report, and (De)Randomized Smoothing have been proposed to enhance the resilience of AI models against such attacks. In this thesis, we introduce a novel framework that integrates masking with attribution analysis to robustify AI models against adversarial patch assaults. Attribution analysis identifies the crucial pixels influencing the model's decision-making process. Subsequently, inspired by the Derandomized Smoothing defense strategy, we employ a masking approach to mask these important pixels. Our experimental findings demonstrate improved robustness against adversarial attacks, at the expense of a slight degradation in clean accuracy.

no / This paper identifies the need for thorough experimental based study for Software-over-the-air (SOTA) in an automotive context. The paper outlines the challenges and context for automotive SOTA with an extensive literature review. It then details the early stages of the experimental studies, which aim to identify the key control and noise factors that affect performance of the SOTA in an automotive environment. This contribution establishes a framework for uncertainty modelling of SOTA as a system which highlights the needs to develop solutions requiring big data gathering and analysis as next research opportunities to the scientific community. / Jaguar Land-Rover

So called discrete models have been successfully used in engineering and computational systems biology. This thesis discusses algebraic methods for modeling and analysis of gene regulatory networks within the discrete modeling context. The first chapter gives a background for discrete models and put in context some of the main research problems that have been pursued in this field for the last fifty years. It also outlines the content of each subsequent chapter. The second chapter focuses on the problem of inferring dynamics from the structure (topology) of the network. It also discusses the characterization of the attractor structure of a network when a particular class of functions control the nodes of the network. Chapters~3 and 4 focus on the study of multi-state nested canalyzing functions as biologically inspired functions and the characterization of their dynamics. Chapter 5 focuses on stochastic methods, specifically on the development of a stochastic modeling framework for discrete models. Stochastic discrete modeling is an alternative approach from the well-known mathematical formalizations such as stochastic differential equations and Gillespie algorithm simulations. Within the discrete setting, a framework that incorporates propensity probabilities for activation and degradation is presented. This approach allows a finer analysis of discrete models and provides a natural setup for cell population simulations. Finally, Chapter 6 discusses future research directions inspired by the work presented here. / Ph. D.

Semiconductor–superconductor hybrid systems are promising candidates for the realization of Majorana fermions and topological order, i.e. topologically protected degeneracies, in solid state devices. We show that the topological order is mirrored in the excitation spectra and can be observed in nonlinear Coulomb blockade transport through a ring-shaped nanowire. Especially, the excitation spectrum is almost independent of magnetic flux in the topologically trivial phase but acquires a characteristic h/e magnetic flux periodicity in the non-trivial phase. The transition between the trivial and nontrivial phase is reflected in the closing and reopening of an excitation gap. We show that the signatures of topological order are robust against details of the geometry, electrostatic disorder and the existence of additional subbands and only rely on the topology of the nanowire and the existence of a superconducting gap. Finally, we show that the coherence length in the non-trivial phase is much longer than in the trivial phase. This opens the possibility to coat the nanowire with superconducting nanograins and thereby significantly reduce the current due to cotunnelling of Cooper pairs and to enhance the Coulomb charging energy without destroying the superconducting gap.

<p dir="ltr">Despite their astonishing capacity to fit data, neural networks have difficulties extrapolating beyond training data distribution. When the out-of-distribution prediction task is formalized as a counterfactual query on a causal model, the reason for their extrapolation failure is clear: neural networks learn spurious correlations in the training data rather than features that are causally related to the target label. This thesis proposes to perform a causal search over a known family of causal models to learn robust (maximally invariant) predictors for single- and multiple-environment extrapolation tasks.</p><p dir="ltr">First, I formalize the out-of-distribution task as a counterfactual query over a structural causal model. For single-environment extrapolation, I argue that symmetries of the input data are valuable for training neural networks that can extrapolate. I introduce Asymmetry learning, a new learning paradigm that is guided by the hypothesis that all (known) symmetries are mandatory even without evidence in training, unless the learner deems it inconsistent with the training data. Asymmetry learning performs a causal model search to find the simplest causal model defining a causal connection between the target labels and the symmetry transformations that affect the label. My experiments on a variety of out-of-distribution tasks on images and sequences show that proposed methods extrapolate much better than the standard neural networks.</p><p dir="ltr">Then, I consider multiple-environment out-of-distribution tasks in dynamical system forecasting that arise due to shifts in initial conditions or parameters of the dynamical system. I identify key OOD challenges in the existing deep learning and physics-informed machine learning (PIML) methods for these tasks. To mitigate these drawbacks, I combine meta-learning and causal structure discovery over a family of given structural causal models to learn the underlying dynamical system. In three simulated forecasting tasks, I show that the proposed approach is 2x to 28x more robust than the baselines.</p>