Federated Learning (FL) is a sub-domain of machine learning (ML) that enforces privacy by allowing the user's local data to reside on their device. Instead of having users send their personal data to a server where the model resides, FL flips the paradigm and brings the model to the user's device for training. Existing works share model parameters or use distillation principles to address the challenges of data heterogeneity. However, these methods ignore some of the other fundamental challenges in FL: device heterogeneity and communication efficiency. In practice, client devices in FL differ greatly in their computational power and communication resources. This is exacerbated by unbalanced data distribution, resulting in an overall increase in training times and the consumption of more bandwidth. In this work, we present a novel approach for resource-efficient FL called emph{REFT} with variable pruning and knowledge distillation techniques to address the computational and communication challenges faced by resource-constrained devices. Our variable pruning technique is designed to reduce computational overhead and increase resource utilization for clients by adapting the pruning process to their individual computational capabilities. Furthermore, to minimize bandwidth consumption and reduce the number of back-and-forth communications between the clients and the server, we leverage knowledge distillation to create an ensemble of client models and distill their collective knowledge to the server. Our experimental results on image classification tasks demonstrate the effectiveness of our approach in conducting FL in a resource-constrained environment. We achieve this by training Deep Neural Network (DNN) models while optimizing resource utilization at each client. Additionally, our method allows for minimal bandwidth consumption and a diverse range of client architectures while maintaining performance and data privacy. / Master of Science / In a world driven by data, preserving privacy while leveraging the power of machine learning (ML) is a critical challenge. Traditional approaches often require sharing personal data with central servers, raising concerns about data privacy. Federated Learning (FL), is a cutting-edge solution that turns this paradigm on its head. FL brings the machine learning model to your device, allowing it to learn from your data without ever leaving your device. While FL holds great promise, it faces its own set of challenges. Existing research has largely focused on making FL work with different types of data, but there are still other issues to be resolved. Our work introduces a novel approach called REFT that addresses two critical challenges in FL: making it work smoothly on devices with varying levels of computing power and reducing the amount of data that needs to be transferred during the learning process. Imagine your smartphone and your laptop. They all have different levels of computing power. REFT adapts the learning process to each device's capabilities using a proposed technique called Variable Pruning. Think of it as a personalized fitness trainer, tailoring the workout to your specific fitness level. Additionally, we've adopted a technique called knowledge distillation. It's like a student learning from a teacher, where the teacher shares only the most critical information. In our case, this reduces the amount of data that needs to be sent across the internet, saving bandwidth and making FL more efficient. Our experiments, which involved training machines to recognize images, demonstrate that REFT works well, even on devices with limited resources. It's a step forward in ensuring your data stays private while still making machine learning smarter and more accessible.

The braided composite actuator is a pressure-driven muscle-like actuator capable of large displacements as well as large blocking forces. It consists of an elastomeric tube reinforced by a sleeve braided by high performance fibers. In addition to the actuation properties, this actuator can also exhibit a large change in stiffness through simple valve control when the working fluid has a high bulk modulus. Several analytical models have been previously developed that capture the geometrical and material nonlinearities, the compliance of the inner liner, and entrapped air in the fluid. The inter fiber yarn compaction in the fiber layer, which is shown to reduce the effective closed-valve stiffness, is studied. A new analytical model for uniformly deformed actuators is developed to capture the compaction effect. This model considers the inter fiber yarn compaction effect and the fiber extensibility as well as the material and geometric nonlinearities. Analysis and experimental results demonstrate that the new compaction model can improve the prediction of the response behavior of the actuator. The compaction model is improved by considering the yarn bending stiffness. The governing equations are derived and the solution algorithm is presented. / Ph. D.

Morphing aircraft combine both radical and subtle wing shape changes to improve vehicle performance relative to a rigid airframe. An aircraft wind tunnel model with considerable wing-shape freedom can serve as a tool in learning to model, control, and fully exploit the potential of such vehicles. This work describes the design, development, and initial analysis of a wind tunnel model that combines large and small wing shape variations for fundamental research in modeling and control of morphing air vehicles. The vehicle is designed for five primary purposes: quasi-steady aerodynamic modeling of an aircraft with large planform changes, optimization studies in achieving efficient flight configurations, transient aerodynamic modeling of high-rate planform changes, evaluating planform maneuvering as an control effector, and gimbaled flight control simulation of a morphing aircraft. The knowledge gained from the wind tunnel evaluations will be used to develop general stabilization and optimal control strategies that can be applied to other vehicles with large scale planform changes and morphing flight models. After a brief background on the development of the Morphing Aircraft Program, and previous research ventures, the first phase vehicle development is described. The vehicle function, subsystems, and control are all presented in addition to the results of first phase wind tunnel testing. Deficiencies in the phase one design motivated the phase two development which has led to the current vehicle model: MORPHEUS. The evolution towards the MORPHEUS configuration is presented in detail along with an elementary strength analysis. The new embedded control implementation to permit a rate controllable planform is included. A preliminary aerodynamic analysis is presented to contrast MORPHEUS against the phase one design and an industry morphing concept. In particular, it is shown how the redesigned model has enhanced performance characteristics and the additional degrees of freedom enable greater flexibility in optimizing a configuration, especially with respect to trim characteristics. An expansion of traditional analysis techniques is applied to derive a new optimal twist algorithm for the MORPHEUS model at each planform configuration. The analysis concludes with a hybrid continuous modeling method that combines first-order computational aerodynamic modeling with classic stability expressions and DATCOM enhancements. The elementary aerodynamic coefficients are computed over the range of possible planform configurations and combined with the optimal twist results for preliminary trim analysis. This work precedes phase two wind tunnel testing and transient modeling. Future work involves expansion into the five purposes detailed for the MORPHEUS model. / Master of Science

Los gráficos de control multivariantes son una gran aportación al control de procesos, siendo el gráfico T2 de Hotelling la opción más utilizada por el operario por su fácil aplicación. Por este motivo se busca potenciar su uso, sin complicar o cargar de esfuerzo adicional a los responsables del proceso. Considerando los buenos resultados obtenidos por gráficos predecesores en los cuales se varía el tamaño de la muestra, esta tesis plantea la posibilidad de obtener mejores resultados variando de forma adaptativa el número de variables involucradas en el control del proceso. Con ello lograr la reducción del ARL o promedio de muestras necesarias hasta que aparezca una señal de fuera de control, además reducir los costos asociados al muestreo utilizando la totalidad de variables involucradas en el proceso únicamente cuando sea necesario. Para poder lograr los objetivos planteados se hizo uso de técnicas de simulación, aplicación de cadenas de Markov y métodos heurísticos (algoritmos genéticos). Se desarrollaron programas informáticos que facilitaron el cálculo y la optimización del diseño de los gráficos de control propuestos, los cuales trabajan con dimensiones variables p1 y p (p1 < p), el primero gráfico denominado de Doble Dimensión (DDT2 ) y el segundo de Dimensión Variable (VDT2 ). Para mostrar los resultados se presentan tablas informativas, se realiza análisis comparativos con los resultados de los gráficos T 2 de Hotelling y MCUSUM y se hace un análisis de sensibilidad. Los gráficos propuestos, logran reducir el ARL fuera de control con respecto al gráfico de control T2 de Hotelling. Los ARL1 de los gráficos propuestos para todos los casos analizados presentan mejor rendimiento que los obtenidos por el gráfico T2 calculadocon solo las primeras p1 variables. En muchos casos el rendimiento de los gráficos DDT2 y VDT2 superan al rendimiento del gráfico T 2 conseguido con las p variables y MCUSUM con las p1 variables. Los gráficos de control propuestos logran reducir los costos asociados al muestreo. A medida que aumentan p1 y p el porcentaje de veces que se utilizan todas las variables va incrementando. Con las distancias d y d1 el efecto es contrario. Por este motivo, se puede afirmar que a pequeñas distancias y mayor cantidad de variables, se obtienen los porcentajes más altos (coste alto de muestreo, aunque más económico que utilizar todas las p variables). Por el contrario, cuando se consideran pocos parámetros y distancias mayores, éste porcentaje es bajo, reduciendo considerablemente los costos del muestreo. / Ruiz Barzola, O. (2013). GRÁFICOS DE CONTROL DE CALIDAD MULTIVARIANTES CON DIMENSIÓN VARIABLE [Tesis doctoral]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/29396

Variable speed limits (VSL) are dynamic traffic management systems designed to increase the efficiency and safety of highways. While the macroscopic performance of VSL systems is well explored in the existing literature, there is a need to further understand the microscopic behavior of vehicles driving in VSL zones. Specifically, driver compliance to advisory VSL systems is quantified based on a driving-simulation experiment and introduced into a broader microscopic behavior model. Statistical analysis indicates that VSL compliance can be predicted based upon several VSL design parameters. The developed two-state microscopic model is calibrated to driving-simulation trajectory data. A calibrated VSL microscopic model can be utilized for new VSL control and macroscopic performance studies, adding an increased dimension of realism to simulation work. As an example, the microscopic model is implemented within VISSIM (overriding the default car-following model) and utilized for a safety-mobility performance assessment of an incident-responsive VSL control algorithm implemented in a MATLAB COM interface. Examination of the multi-objective optimization frontier reveals an inverse relationship between safety and mobility under different control algorithm parameters. Engineers are thus faced with a decision between performing multi-objective optimization and selecting a dominant VSL control objective (e.g. maximizing safety versus mobility performance). / Master of Science / Variable speed limits (VSL) are dynamic traffic management systems designed to increase the efficiency and safety of highways. While the system performance of VSL systems is well explored in previous research, there is a need to further understand the individual behavior of vehicles driving under VSL control. Specifically, driver compliance to advisory VSL systems is modelled based on a driving-simulation experiment. Low compliance equates to poor VSL performance so it is important for engineers to have the ability to predict compliance based on VSL design conditions. The compliance model is introduced into a driver behavior model that quantifies and predicts the driver decision process on VSL controlled highways. The driver behavior model parameters are set using data obtained from the driving-simulation experiment. Utilization of the developed driver behavior model will increase the accuracy of future simulation work on VSL system performance. In this study, the model is implemented within a traffic simulation software to conduct an assessment of the trade-offs between safety and mobility VSL performance for different VSL control designs. An accident is modelled in the simulation software, and VSL is utilized to respond to and alleviate the incident. Simulation results indicate an inverse relationship between safety and mobility performance – indicating that engineers must select a primary objective when selecting VSL control design parameters.

Simulation, whether in-flight or ground-based, is an invaluable tool for testing and evaluating aircraft. Classically, a simulation model is specific to a single particular airframe, only able to model those flying characteristics. Vast information can be gained from a simulation that is able to model a wide range of aircraft, through a comparison of the performance of these aircraft. Such a variable stability simulation model was created based on 46 stability parameters, including natural frequencies, damping ratios, time constants, and gains. The simulation was obtained using transfer functions representing the aircraft state responses to control inputs. These transfer functions were converted into state space systems used to create the linear equations for the model. The model was first developed as a desktop simulation and then converted for use with the Virginia Tech's 2F122A flight simulator. This conversion required a simple dynamic inversion of the body axis force and moment terms. To reduce the error in these terms, a model following scheme was incorporated. A series of canned inputs and real-time pilot-in-the-loop tests were flown to evaluate the variable stability model. Results in this paper have demonstrated the successful creation of a variable stability simulation model. / Master of Science

Conventional stereo vision systems rely on two spatially fixed cameras to gather depth information about a scene. The cameras typically have a fixed distance between them, known as the baseline. As the baseline increases, the estimated 3D information becomes more accurate, which makes it advantageous to have as large a baseline as possible. However, large baselines have problems whenever objects approach the cameras. The objects begin to leave the field of view of the cameras, making it impossible to determine where they are located in 3D space. This becomes especially important if an object of interest must be actuated upon and is approached by a vehicle. In an attempt to overcome this limitation, this thesis introduces a variable baseline stereo system that can adjust its baseline automatically based on the location of an object of interest. This allows accurate depth information to be gathered when an object is both near and far. The system was designed to operate under, and automatically move to a large range of different baselines. This thesis presents the mechanical design of the stereo boom. This is followed by a derivation of a control scheme that adjusts the baseline based on an estimate object location, which is gathered from stereo vision. This algorithm ensures that a certain incident angle on an object of interest is never surpassed. This maximum angle is determined by where a stereo correspondence algorithm, Semi-Global Block Matching, fails to create full reconstructions. / Master of Science

Analog-to-digital converters (ADCs) are used to convert analog signals to the digital domain in digital communications systems. An ADC used in wireless communications should meet the necessary requirements for the worst-case channel condition. However, the worst-case scenario rarely occurs. As a consequence, a high-resolution and subsequently high power ADC designed for the worst case is not required for most operating conditions. A solution to reduce the power dissipation of ADCs in wireless digital communications systems is to detect the current channel condition and to dynamically vary the resolution of the ADC according to the given channel condition. In this thesis, we investigated an ADC that can change its resolution dynamically and, consequently, its power dissipation. Our ADC is a switched-current, redundant signed-digit (RSD) cyclic implementation that easily incorporates variable resolution. Furthermore, the RSD cyclic algorithm is insensitive to offsets, allowing simple, low-power comparators. Our ADC is implemented in a 0.35 um CMOS technology with a single-ended 3.3 V power supply. Our ADC has a maximum power dissipation of 6.35 mW for a 12-bit resolution and dissipates an average of 10 percent less power when the resolution is decreased by two bits. Simulation results indicate our ADC achieves a bit rate of 1.7 MHz and has a SNR of 84 dB for the maximum input frequency of 8.3 kHz. / Master of Science

We present a utility accrual real-time scheduling algorithm called CIC-VCUA, for tasks whose execution times are functions of their starting times. We model such variable execution times employing variable cost functions (or VCFs). The algorithm considers application activities that are subject to time/utility function time constraints (or TUFs), execution times described using VCFs, and concurrent, mutually exclusive sharing of non-CPU resources. We consider the multi-criteria scheduling objective of (1) assuring that the maximum interval between any two consecutive, successful completions of jobs of a task must not exceed a specified upper bound, and (2) maximizing the system's total accrued utility, while satisfying mutual exclusion resource constraints. Since the scheduling problem is intractable, CIC-VCUA statically computes worst-case sojourn times of tasks, selects tasks for execution based on their potential utility density, and completes them at specific times, in polynomial-time. We establish that CIC-VCUA achieves optimal timeliness during under-loads. Further, we identify the conditions under which timeliness assurances hold. Our simulation experiments illustrate CIC-VCUA's effectiveness and superiority. / Master of Science

Yes / In this paper we study a complex real world workforce scheduling problem. We apply constructive search and variable neighbourhood search (VNS) metaheuristics and enhance these methods by using a variable fitness function. The variable fitness function (VFF) uses an evolutionary approach to evolve weights for each of the (multiple) objectives. The variable fitness function can potentially enhance any search based optimisation heuristic where multiple objectives can be defined through evolutionary changes in the search direction. We show that the VFF significantly improves performance of constructive and VNS approaches on training problems, and "learn" problem features which enhance the performance on unseen test problem instances.