In this master thesis project a LiDAR-based, depth image-based and semantic segmentation image-based reinforcement learning agent is investigated and compared forlearning in simulation and performing in real-time. The project utilize the Deep Deterministic Policy Gradient architecture for learning continuous actions and was designed to control a RC car. One of the first project to deploy an agent in a real scenario after training in a similar simulation. The project demonstrated that with a proper reward function and by tuning driving parameters such as restricting steering, maximum velocity, minimum velocity and performing input data scaling a LiDAR-based agent could drive indefinitely on a simple but completely unseen track in real-time.

With the exponential growth of virtualization and cloud computing over the last decade, many companies in the telecommunications sector have started their journey towards cloud migration by exchanging a lot of specialized hardware for virtualized solutions. With more and more applications running in a cloud environment, it became essential to run these applications on heterogeneous systems with shared underlying hardware and software resources. However, running these applications in a heterogeneous cloud environment often leads to  unpredictable and non-deterministic performance, as all the applications compete for the shared resources to improve their individual performance. This becomes a problem when the interference from the co-hosted applications starts affecting the performance of the critical applications running on the same server. Ericsson is therefore investigating a solution to dynamically manage the low-level hardware and software resources to get deterministic performance on applications deployed using Kubernetes.  In this thesis, the Intent Driven Orchestration (IDO) model developed by Intel is used as the baseline model. This model was then extended by adding another tool to the setup called Container Runtime Interface-Resource Manager (CRI-RM), which is used to manipulate low-level software and hardware resources managed by a Kubernetes cluster at runtime. The results achieved in this thesis suggest that it is possible to get deterministic performance for an application deployed using Kubernetes, by identifying and isolating the CPU cores in the cluster on which the application is running.

Millions of lives have been lost over the years as a result of the effects of earthquakes. One of these devastating effects is slope failure, more commonly known as landslide. Over the years, seismologists and engineers have teamed up to better record data during an earthquake. As technology has advanced, the data obtained have become more refined, allowing engineers to use the data in their efforts to estimate earthquakes where they have not yet occurred. Several methods have been proposed over time to utilize the earthquake data and estimate slope displacements. A pioneer in the development of methods to estimate slope displacements, Nathan Newmark, proposed what is now called the Newmark sliding block method. This method explained in very simple ways how a mass, in this case a rigid block, would slide over an incline given that the acceleration of the block surpassed the frictional resistance created between the bottom of the block and the surface of the incline. Because many of the assumptions from this method were criticized by scientists over time, modified Newmark sliding block methods were proposed. As the original and modified Newmark sliding block methods were introduced, the need to account for the uncertainty in the way soil would behave under earthquake loading became a big challenge. Deterministic and probabilistic methods have been used to incorporate parameters that would account for some of the uncertainty in the analysis. In an attempt to use a probabilistic approach in understanding how slopes might fail, the Pacific Earthquake Engineering Research Center proposed a performance-based earthquake engineering framework that would allow decision-makers to use probabilistically generated information to make decisions based on acceptable risk. Previous researchers applied this framework to simplified Newmark sliding block models, but the approach is difficult for engineers to implement in practice because of the numerous probability calculations that are required. The work presented in this thesis provides a solution to the implementation of the performance-based approach by providing a simplified procedure for the performance-based determination of seismic slope displacements using the Rathje & Saygili (2009) and the Bray and Travasarou (2007) simplified Newmark sliding block models. This document also includes hazard parameter maps, which are an important part of the simplified procedure, for five states in the United States. A validation of the method is provided, as well as a comparison of the simplified method against other commonly used approaches such as deterministic and pseudo-probabilistic.

Liquefaction-induced lateral spread displacements cause severe damage to infrastructure, resulting in large economic losses in affected regions. Predicting lateral spread displacements is an important aspect in any seismic analysis and design, and many different methods have been developed to accurately estimate these displacements. However, the inherent uncertainty in predicting seismic events, including the extent of liquefaction and its effects, makes it difficult to accurately estimate lateral spread displacements. Current conventional methods of predicting lateral spread displacements do not completely account for uncertainty, unlike a performance-based earthquake engineering (PBEE) approach that accounts for the all inherent uncertainty in seismic design. The PBEE approach incorporates complex probability theory throughout all aspects of estimating liquefaction-induced lateral spread displacements. A new fully-probabilistic PBEE method, based on results from the cone penetration test (CPT), was created for estimating lateral spread displacements using two different liquefaction triggering procedures. To accommodate the complexity of all probabilistic calculations, a new seismic hazard analysis tool, CPTLiquefY, was developed. Calculated lateral spread displacements using the new fully-probabilistic method were compared to estimated displacements using conventional methods. These comparisons were performed across 20 different CPT profiles and 10 cities of varying seismicity. The results of this comparison show that the conventional procedures of estimating lateral spread displacements are sufficient for areas of low seismicity and for lower return periods. However, by not accounting for all uncertainties, the conventional methods under-predict lateral spread displacements in areas of higher seismicity. This is cause for concern as it indicates that engineers in industry using the conventional methods are likely under-designing structures to resist lateral spread displacements for larger seismic events.

Liquefaction-induced lateral spread displacements cause severe damage to infrastructure, resulting in large economic losses in affected regions. Predicting lateral spread displacements is an important aspect in any seismic analysis and design, and many different methods have been developed to accurately estimate these displacements. However, the inherent uncertainty in predicting seismic events, including the extent of liquefaction and its effects, makes it difficult to accurately estimate lateral spread displacements. Current conventional methods of predicting lateral spread displacements do not completely account for uncertainty, unlike a performance-based earthquake engineering (PBEE) approach that accounts for the all inherent uncertainty in seismic design. The PBEE approach incorporates complex probability theory throughout all aspects of estimating liquefaction-induced lateral spread displacements. A new fully-probabilistic PBEE method, based on results from the cone penetration test (CPT), was created for estimating lateral spread displacements using two different liquefaction triggering procedures. To accommodate the complexity of all probabilistic calculations, a new seismic hazard analysis tool, CPTLiquefY, was developed. Calculated lateral spread displacements using the new fully-probabilistic method were compared to estimated displacements using conventional methods. These comparisons were performed across 20 different CPT profiles and 10 cities of varying seismicity. The results of this comparison show that the conventional procedures of estimating lateral spread displacements are sufficient for areas of low seismicity and for lower return periods. However, by not accounting for all uncertainties, the conventional methods under-predict lateral spread displacements in areas of higher seismicity. This is cause for concern as it indicates that engineers in industry using the conventional methods are likely under-designing structures to resist lateral spread displacements for larger seismic events.

In this thesis, two FPGA testbeds to demonstrate low-latency deterministic Guaranteed- Rate (GR) connections in packet switched networks such as the Internet of Things are developed. Each FPGA testbed consists of multiple simple Input Queued (IQ) switches or routers, interconnected in a given topology to form a forwarding-plane. Each switch has an associated switch controller with several programmable Lookup- Tables (LUTs). A Software Defined Networking (SDN) control plane can configure the switch controllers to establish the GR connections in the forwarding-plane of IP routers or layer- 2 packet switches. According to a recent paper in the IEEE Transactions on Networking; (1) The use of very low jitter GR connections can reduce queuing delays to negligible values, so that the end-to-end delays can be reduced to the buffer latency. (2) The routers, switches and links can operate at 100% loads, while simultaneously guaranteeing very low end- to-end latencies. The goal of the thesis is to evaluate these properties in real hardware clocked at MegaHertz clock rates. In the first testbed, a network of 8 simple IQ switches organized in a linear array is synthesized on an Altera Cyclone IV FPGA. 128 GR traffic flows were routed through the testbed to effectively saturate the switches and links. In the second testbed, a USA backbone topology with 26 simple IQ switches and 88 links is synthesized on the FPGA. Over 300 GR traffic flows were routed through the USA network to achieve utilizations exceeding 90%. In both testbeds, packets move through the forwarding plane at a clock rate of 65 MHz, transferring millions of packets per second, and statistics are recorded. Both testbeds con rm that traffic flows achieve deterministic GR service with minimum buffering, where end-to-end delays are effectively reduced to the fiber latency. These hardware testbeds demonstrate the technical feasibility of achieving deterministic GR services in a packet-switched network such as Internet of Things using simple FPGA switch controllers working with an SDN control plane. The technology also applies to networks of simple optical packet switches with minimal buffering. / Thesis / Master of Applied Science (MASc)

No description available.

La biología sintética se define como la ingeniería de la biología: el (re)diseño y construcción de nuevas partes, dispositivos y sistemas biológicos para realizar nuevas funciones con fines útiles, que se basan en principios elucidados de la biología y la ingeniería. Para facilitar la construcción rápida, reproducible y predecible de estos sistemas biológicos a partir de conjuntos de componentes es necesario desarrollar nuevos métodos y herramientas. La tesis plantea la optimización multiobjetivo como el marco adecuado para tratar los problemas comunes que surgen en el diseño racional y el ajuste óptimo de los circuitos genéticos sintéticos. Utilizando un enfoque clásico de ingeniería de sistemas, la tesis se centra principalmente en: i) el modelado de circuitos genéticos sintéticos basado en los primeros principios, ii) la estimación de parámetros de modelos a partir de datos experimentales y iii) el ajuste basado en modelos para lograr el desempeño deseado de los circuitos. Se han utilizado dos circuitos genéticos sintéticos de diferente naturaleza y con diferentes objetivos y problemas: un circuito de realimentación de tipo 1 incoherente (I1-FFL) que exhibe la importante propiedad biológica de adaptación, y un circuito de detección de quorum sensing y realimentación (QS/Fb) que comprende dos bucles de realimentación entrelazados -uno intracelular y uno basado en la comunicación de célula a célula- diseñado para regular el nivel medio de expresión de una proteína de interés mientras se minimiza su varianza a través de la población de células. Ambos circuitos han sido analizados in silico e implementados in vivo. En ambos casos, se han desarrollado modelos de estos circuitos basado en primeros principios. Se presta especial atención a ilustrar cómo obtener modelos de orden reducido susceptibles de estimación de parámetros, pero manteniendo el significado biológico. La estimación de los parámetros del modelo a partir de los datos experimentales se considera en diferentes escenarios, tanto utilizando modelos determinísticos como estocásticos. Para el circuito I1-FFL se consideran modelos determinísticos. Aquí, la tesis plantea la utilización de modelos locales utilizando la optimización multiobjetivo para realizar la estimación de parámetros del modelo bajo escenarios con estructura de modelo incompleta. Para el circuito QS/Fb, una estructura controlada por realimentación, el problema tratado es la falta de excitabilidad de las señales. La tesis propone una metodología de estimación en dos etapas utilizando modelos estocásticos. La metodología permite utilizar datos de curso temporal promediados de la población y mediciones de distribución en estado estacionario para una sola célula. El ajuste de circuitos basado en modelos para lograr un desempeño deseado también se aborda mediante la optimización multiobjetivo. Para el circuito QS/Fb se realiza un análisis estocástico completo. La tesis aborda cómo tener en cuenta correctamente tanto el ruido intrínseco como el extrínseco, las dos principales fuentes de ruido en los circuitos genéticos. Se analiza el equilibrio entre ambas fuentes de ruido y el papel que desempeñan en el bucle de realimentación intracelular, y en la realimentación extracelular de toda la población. La principal conclusión es que la compleja interacción entre ambos canales de realimentación obliga al uso de la optimización multiobjetivo para el adecuado ajuste del circuito. En esta tesis además del uso adecuado de herramientas de optimización multiobjetivo, la principal preocupación es cómo derivar directrices para el ajuste in silico de parámetros de circuitos que puedan aplicarse de forma realista in vivo en un laboratorio estándar. Como alternativa al análisis de sensibilidad de parámetros clásico, la tesis propone el uso de técnicas de clustering a lo largo de los frentes de Pareto, relacionando el compr / La biologia sintètica es defineix com l'enginyeria de la biologia: el (re) disseny i construcció de noves parts, dispositius i sistemes biològics per a realitzar noves funcions útils que es basen a principis elucidats de la biologia i l'enginyeria. Per facilitar la construcció ràpida, reproduïble i predictible de aquests sistemes biològics a partir de conjunts de components és necessari desenvolupar nous mètodes i eines. La tesi planteja la optimització multiobjectiu com el marc adequat per a tractar els problemes comuns que apareixen en el disseny racional i l' ajust òptim dels circuits genètics sintètics. Utilitzant un enfocament clàssic d'enginyeria de sistemes, la tesi es centra principalment en: i) el modelatge de circuits genètics sintètics basat en primers principis, ii) l' estimació de paràmetres de models a partir de dades experimentals i iii) l' ajust basat en models per aconseguir el rendiment desitjat dels circuits. S'han utilitzat dos circuits genètics sintètics de diferent naturalesa i amb diferents objectius i problemes: un circuit de prealimentació de tipus 1 incoherent (I1-FFL) que exhibeix la important propietat biològica d'adaptació, i un circuit de quorum sensing i realimentació (QS/Fb) que comprèn dos bucles de realimentació entrellaçats -un intracel·lular i un basat en la comunicació de cèl·lula a cèl·lula- dis-senyat per regular el nivell mitjà d'expressió normal d'una proteïna d'interès mentre es minimitza la seua variació al llarg de la població de cèl·lules. Els dos circuits han estat analitzats in silico i implementats in vivo. En tots dos casos, s'han desenvolupat models basats en primers principis d'aquests circuits. Després es presta especial atenció a delinear com obtenir models d'ordre reduït susceptibles de estimació de paràmetres, però mantenint el significat biològic. L' estimació dels paràmetres del model a partir de les dades experimentals es considera en diferents escenaris, tant utilitzant models determinístics com estocàstics. Per al circuit I1-FFL es consideren models determinístics. La tesi planteja la utilització de models locals utilitzant la optimització multiobjectiu per realitzar l'estimació de parametres del model sota escenaris amb estructura de model incompleta (dinàmica no modelada). Per al circuit de QS/Fb, una estructura controlada per realimentació, el problema tractat és la manca d'excitabilitat dels senyals. La tesi proposa una metodologia de estimació en dues etapes utilitzant models estocàstics. La metodologia permet utilitzar dades de curs temporal promediats de la població i mesures de distribució en estat estacionari d'una sola una cèl·lula. L' ajust de circuits basat en models per aconseguir el rendiment desitjat dels circuits també s' aborda mitjançant la optimització multiobjectiu. Per al circuit QS/Fb, es fa un anàlisi estocàstic complet. La tesi aborda com tenir en compte correctament tant el soroll intrínsec com l' extrínsec, les dues principals fonts de soroll en els circuits genètics sintètics. S' analitza l'equilibri entre dues fonts de soroll i el paper que exerceixen en el bucle de realimentació intracel·lular, les i en la realimentació extracel·lular de tota la població. La principal conclusió es que la complexa interacció entre els dos canals de realimentació fa necessari l' ús de la optimització multiobjectiu per al adequat ajust del circuit. En aquesta tesi, a més de l'ús adequat d'eines d'optimització multiobjectiu, la principal preocupació és com derivar directives per al ajust in silico de paràmetres de circuits que puguin aplicar-se de forma realista en viu en un laboratori estàndard. Així, com a alternativa a l'anàlisi de sensibilitat de paràmetres clàssic, la tesi proposa l'ús de l' tècniques de l'agrupació al llarg dels fronts de Pareto, relacionant el compromís de dessempeny amb les regions en l'espai d'paràmetres. / Synthetic biology is defined as the engineering of biology: the deliberate (re)design and construction of novel biological and biologically based parts, devices and systems to perform new functions for useful purposes, that draws on principles elucidated from biology and engineering. Methods and tools are needed to facilitate fast, reproducible and predictable construction of biological systems from sets of biological components. This thesis raises multi-objective optimization as the proper framework to deal with common problems arising in rational design and optimal tuning of synthetic gene circuits. Using a classical systems engineering approach, the thesis mainly addresses: i) synthetic gene circuit modeling based on first principles, ii) model parameters estimation from experimental data and iii) model-based tuning to achieve desired circuit performance. Two gene synthetic circuits of different nature and with different goals and inherent problems have been used throughout the thesis: an Incoherent type 1 feedforward circuit (I1-FFL) that exhibits the important biological property of adaptation, and a Quorum sensing/Feedback circuit (QS/Fb) comprising two intertwined feedback loops -an intracellular one and a cell-to-cell communication-based one-- designed to regulate the mean expression level of a protein of interest while minimizing its variance across the population of cells. Both circuits have been analyzed in silico and implemented in vivo. In both cases, circuit modeling based on first principles has been carried out. Then, special attention is paid to illustrate how to obtain reduced order models amenable for parameters estimation yet keeping biological significance. Model parameters estimation from experimental data is considered in different scenarios, both using deterministic and stochastic models. For the I1-FFL circuit, deterministic models are considered. In this case, the thesis raises ensemble modeling using multi-objective optimization to perform model parameters estimation under scenarios with incomplete model structure (unmodeled dynamics). For the QS/Fb gene circuit, a feedback controlled structure, the lack of excitability of the signals is the problem addressed. The thesis proposes a two-stage estimation methodology using stochastic models. The methodology allows using population averaged time-course data and steady state distribution measurements at the single-cell level. Model-based circuit tuning to achieve desired circuit performance is also addressed using multi-objective optimization. First, for the QS/Fb feedback control circuit, a complete stochastic analysis is performed. Here, the thesis addresses how to correctly take into account both intrinsic and extrinsic noise, the two main sources of noise in gene synthetic circuits. The trade-off between both sources of noise, and the role played by in the intracellular single-cell feedback loop and the extracellular population-wide feedback is analyzed. The main conclusion being that the complex interplay between both feedback channels compel the use of multi-objective optimization for proper tuning of the circuit to achieve desired performance. Thus, the thesis wraps up all the previous results and uses them to address circuit tuning for desired performance. Here, besides the proper use of multi-objective optimization tools, the main concern is how to derive guidelines for circuit parameters tuning in silico that can realistically be applied in vivo in a standard laboratory. Thus, as an alternative to classical parameters sensitivity analysis, the thesis proposes the use of clustering techniques along the optimal Pareto fronts relating the performance trade-offs with regions in the circuits parameters space. / This work has been partially supported by the Spanish Government (CICYT DPI2014- 55276-C5-1) and the European Union (FEDER). The author was recipient of the grant Formación de Personal Investigador by the Universitat Politècnica de València, subprogram 1 (FPI/2013-3242). She was also recipient of the competitive grants for pre-doctoral stays Erasmus Student Placement-European Programme 2015, and FPI Mobility program 2016 of the Universitat Politècnica de València. She also received the competitive grant for a pre-doctoral stay Becas de movilidad para Jóvenes Profesores e Investigadores 2016, Programa de Becas Iberoamérica of the Santander Bank. / Boada Acosta, YF. (2018). A systems engineering approach to model, tune and test synthetic gene circuits [Tesis doctoral]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/112725

Longitudinal velocity control, or adaptive cruise control (ACC), is a common advanced driving feature aimed at assisting the driver and reducing fatigue. It maintains the velocity of a vehicle and ensures a safe distance from the preceding vehicle. Many models for ACC are available, such as Proportional, Integral, and Derivative (PID) and Model Predictive Control (MPC). However, conventional models have some limitations as they are designed for simplified driving scenarios. Artificial intelligence (AI) and machine learning (ML) have made robust navigation and decision-making possible in complex environments. Recent approaches, such as reinforcement learning (RL), have demonstrated remarkable performance in terms of faster processing and effective navigation through unknown environments. This dissertation explores an RL approach, deep deterministic policy gradient (DDPG), for longitudinal velocity control. The baseline DDPG model has been modified in two different ways. In the first method, an attention mechanism has been applied to the neural network (NN) of the DDPG model. Integrating the attention mechanism into the DDPG model helps in decreasing focus on less important features and enhances overall model effectiveness. In the second method, the inputs of the actor and critic networks of DDPG are replaced with outputs of the self-supervised network. The self-supervised learning process allows the model to accurately predict future states from current states and actions. A custom reward function has been designed for the RL algorithm considering overall safety, efficiency, and comfort. The proposed models have been trained with human car-following data, and evaluated on multiple datasets, including publicly available data, simulated data, and sensor data collected from real-world environments. The analyses demonstrate that the new architectures can maintain strong robustness across various datasets and outperform the current state-of-the-art models.

Modern machine learning methods are dependent on active optimization research to improve the set of methods available for the efficient and effective extraction of information from large datasets. This, in turn, requires an intense and rigorous study of optimization methods and their possible applications to crucial machine learning applications in order to advance the potential benefits of the field. This thesis provides a study of several modern optimization techniques and supplies a mathematical inquiry into the effectiveness of homotopy methods to attack a fundamental machine learning problem, effective clustering under constraints. The first part of this thesis provides an empirical survey of several popular optimization algorithms, along with one approach that is cutting-edge. These algorithms are tested against deeply challenging real-world problems with vast numbers of local minima, and compares and contrasts the benefits of each when confronted with problems of different natures. The second part of this thesis proposes a new homotopy map for use with constrained clustering problems. This thesis explores the connections between the map and the problem, providing several theorems to justify the use of the map and making use of modern homotopy tracking software to compare an optimization that employs the map with several modern approaches to solving the same problem. / Ph. D.