On Recurrent and Deep Neural Networks

L'apprentissage profond est un domaine de recherche en forte croissance en apprentissage automatique qui est parvenu à des résultats impressionnants dans différentes tâches allant de la classification d'images à la parole, en passant par la modélisation du langage. Les réseaux de neurones récurrents, une sous-classe d'architecture profonde, s'avèrent particulièrement prometteurs. Les réseaux récurrents peuvent capter la structure temporelle dans les données. Ils ont potentiellement la capacité d'apprendre des corrélations entre des événements éloignés dans le temps et d'emmagasiner indéfiniment des informations dans leur mémoire interne. Dans ce travail, nous tentons d'abord de comprendre pourquoi la profondeur est utile. Similairement à d'autres travaux de la littérature, nos résultats démontrent que les modèles profonds peuvent être plus efficaces pour représenter certaines familles de fonctions comparativement aux modèles peu profonds. Contrairement à ces travaux, nous effectuons notre analyse théorique sur des réseaux profonds acycliques munis de fonctions d'activation linéaires par parties, puisque ce type de modèle est actuellement l'état de l'art dans différentes tâches de classification. La deuxième partie de cette thèse porte sur le processus d'apprentissage. Nous analysons quelques techniques d'optimisation proposées récemment, telles l'optimisation Hessian free, la descente de gradient naturel et la descente des sous-espaces de Krylov. Nous proposons le cadre théorique des méthodes à région de confiance généralisées et nous montrons que plusieurs de ces algorithmes développés récemment peuvent être vus dans cette perspective. Nous argumentons que certains membres de cette famille d'approches peuvent être mieux adaptés que d'autres à l'optimisation non convexe. La dernière partie de ce document se concentre sur les réseaux de neurones récurrents. Nous étudions d'abord le concept de mémoire et tentons de répondre aux questions suivantes: Les réseaux récurrents peuvent-ils démontrer une mémoire sans limite? Ce comportement peut-il être appris? Nous montrons que cela est possible si des indices sont fournis durant l'apprentissage. Ensuite, nous explorons deux problèmes spécifiques à l'entraînement des réseaux récurrents, à savoir la dissipation et l'explosion du gradient. Notre analyse se termine par une solution au problème d'explosion du gradient qui implique de borner la norme du gradient. Nous proposons également un terme de régularisation conçu spécifiquement pour réduire le problème de dissipation du gradient. Sur un ensemble de données synthétique, nous montrons empiriquement que ces mécanismes peuvent permettre aux réseaux récurrents d'apprendre de façon autonome à mémoriser des informations pour une période de temps indéfinie. Finalement, nous explorons la notion de profondeur dans les réseaux de neurones récurrents. Comparativement aux réseaux acycliques, la définition de profondeur dans les réseaux récurrents est souvent ambiguë. Nous proposons différentes façons d'ajouter de la profondeur dans les réseaux récurrents et nous évaluons empiriquement ces propositions.

Semiautomated 3D liver segmentation using computed tomography and magnetic resonance imaging

Le foie est un organe vital ayant une capacité de régénération exceptionnelle et un rôle crucial dans le fonctionnement de l’organisme. L’évaluation du volume du foie est un outil important pouvant être utilisé comme marqueur biologique de sévérité de maladies hépatiques. La volumétrie du foie est indiquée avant les hépatectomies majeures, l’embolisation de la veine porte et la transplantation. La méthode la plus répandue sur la base d'examens de tomodensitométrie (TDM) et d'imagerie par résonance magnétique (IRM) consiste à délimiter le contour du foie sur plusieurs coupes consécutives, un processus appelé la «segmentation». Nous présentons la conception et la stratégie de validation pour une méthode de segmentation semi-automatisée développée à notre institution. Notre méthode représente une approche basée sur un modèle utilisant l’interpolation variationnelle de forme ainsi que l’optimisation de maillages de Laplace. La méthode a été conçue afin d’être compatible avec la TDM ainsi que l' IRM. Nous avons évalué la répétabilité, la fiabilité ainsi que l’efficacité de notre méthode semi-automatisée de segmentation avec deux études transversales conçues rétrospectivement. Les résultats de nos études de validation suggèrent que la méthode de segmentation confère une fiabilité et répétabilité comparables à la segmentation manuelle. De plus, cette méthode diminue de façon significative le temps d’interaction, la rendant ainsi adaptée à la pratique clinique courante. D’autres études pourraient incorporer la volumétrie afin de déterminer des marqueurs biologiques de maladie hépatique basés sur le volume tels que la présence de stéatose, de fer, ou encore la mesure de fibrose par unité de volume.

Theoretical study of magnetism, structure and chemical order in transition-metal alloy clusters

Research on transition-metal nanoalloy clusters composed of a few atoms is fascinating by their unusual properties due to the interplay among the structure, chemical order and magnetism. Such nanoalloy clusters, can be used to construct nanometer devices for technological applications by manipulating their remarkable magnetic, chemical and optical properties. Determining the nanoscopic features exhibited by the magnetic alloy clusters signifies the need for a systematic global and local exploration of their potential-energy surface in order to identify all the relevant energetically low-lying magnetic isomers. In this thesis the sampling of the potential-energy surface has been performed by employing the state-of-the-art spin-polarized density-functional theory in combination with graph theory and the basin-hopping global optimization techniques. This combination is vital for a quantitative analysis of the quantum mechanical energetics. The first approach, i.e., spin-polarized density-functional theory together with the graph theory method, is applied to study the Fe$_m$Rh$_n$ and Co$_m$Pd$_n$ clusters having $N = m+n \leq 8$ atoms. We carried out a thorough and systematic sampling of the potential-energy surface by taking into account all possible initial cluster topologies, all different distributions of the two kinds of atoms within the cluster, the entire concentration range between the pure limits, and different initial magnetic configurations such as ferro- and anti-ferromagnetic coupling. The remarkable magnetic properties shown by FeRh and CoPd nanoclusters are attributed to the extremely reduced coordination number together with the charge transfer from 3$d$ to 4$d$ elements. The second approach, i.e., spin-polarized density-functional theory together with the basin-hopping method is applied to study the small Fe$_6$, Fe$_3$Rh$_3$ and Rh$_6$ and the larger Fe$_{13}$, Fe$_6$Rh$_7$ and Rh$_{13}$ clusters as illustrative benchmark systems. This method is able to identify the true ground-state structures of Fe$_6$ and Fe$_3$Rh$_3$ which were not obtained by using the first approach. However, both approaches predict a similar cluster for the ground-state of Rh$_6$. Moreover, the computational time taken by this approach is found to be significantly lower than the first approach. The ground-state structure of Fe$_{13}$ cluster is found to be an icosahedral structure, whereas Rh$_{13}$ and Fe$_6$Rh$_7$ isomers relax into cage-like and layered-like structures, respectively. All the clusters display a remarkable variety of structural and magnetic behaviors. It is observed that the isomers having similar shape with small distortion with respect to each other can exhibit quite different magnetic moments. This has been interpreted as a probable artifact of spin-rotational symmetry breaking introduced by the spin-polarized GGA. The possibility of combining the spin-polarized density-functional theory with some other global optimization techniques such as minima-hopping method could be the next step in this direction. This combination is expected to be an ideal sampling approach having the advantage of avoiding efficiently the search over irrelevant regions of the potential energy surface.

Individual Decision-Making and Inflation Persistence

The traditional task of a central bank is to preserve price stability and, in doing so, not to impair the real economy more than necessary. To meet this challenge, it is of great relevance whether inflation is only driven by inflation expectations and the current output gap or whether it is, in addition, influenced by past inflation. In the former case, as described by the New Keynesian Phillips curve, the central bank can immediately and simultaneously achieve price stability and equilibrium output, the so-called ‘divine coincidence’ (Blanchard and Galí 2007). In the latter case, the achievement of price stability is costly in terms of output and will be pursued over several periods. Similarly, it is important to distinguish this latter case, which describes ‘intrinsic’ inflation persistence, from that of ‘extrinsic’ inflation persistence, where the sluggishness of inflation is not a ‘structural’ feature of the economy but merely ‘inherited’ from the sluggishness of the other driving forces, inflation expectations and output. ‘Extrinsic’ inflation persistence is usually considered to be the less challenging case, as policy-makers are supposed to fight against the persistence in the driving forces, especially to reduce the stickiness of inflation expectations by a credible monetary policy, in order to reestablish the ‘divine coincidence’. The scope of this dissertation is to contribute to the vast literature and ongoing discussion on inflation persistence: Chapter 1 describes the policy consequences of inflation persistence and summarizes the empirical and theoretical literature. Chapter 2 compares two models of staggered price setting, one with a fixed two-period duration and the other with a stochastic duration of prices. I show that in an economy with a timeless optimizing central bank the model with the two-period alternating price-setting (for most parameter values) leads to more persistent inflation than the model with stochastic price duration. This result amends earlier work by Kiley (2002) who found that the model with stochastic price duration generates more persistent inflation in response to an exogenous monetary shock. Chapter 3 extends the two-period alternating price-setting model to the case of 3- and 4-period price durations. This results in a more complex Phillips curve with a negative impact of past inflation on current inflation. As simulations show, this multi-period Phillips curve generates a too low degree of autocorrelation and too early turnings points of inflation and is outperformed by a simple Hybrid Phillips curve. Chapter 4 starts from the critique of Driscoll and Holden (2003) on the relative real-wage model of Fuhrer and Moore (1995). While taking the critique seriously that Fuhrer and Moore’s model will collapse to a much simpler one without intrinsic inflation persistence if one takes their arguments literally, I extend the model by a term for inequality aversion. This model extension is not only in line with experimental evidence but results in a Hybrid Phillips curve with inflation persistence that is observably equivalent to that presented by Fuhrer and Moore (1995). In chapter 5, I present a model that especially allows to study the relationship between fairness attitudes and time preference (impatience). In the model, two individuals take decisions in two subsequent periods. In period 1, both individuals are endowed with resources and are able to donate a share of their resources to the other individual. In period 2, the two individuals might join in a common production after having bargained on the split of its output. The size of the production output depends on the relative share of resources at the end of period 1 as the human capital of the individuals, which is built by means of their resources, cannot fully be substituted one against each other. Therefore, it might be rational for a well-endowed individual in period 1 to act in a seemingly ‘fair’ manner and to donate own resources to its poorer counterpart. This decision also depends on the individuals’ impatience which is induced by the small but positive probability that production is not possible in period 2. As a general result, the individuals in the model economy are more likely to behave in a ‘fair’ manner, i.e., to donate resources to the other individual, the lower their own impatience and the higher the productivity of the other individual. As the (seemingly) ‘fair’ behavior is modelled as an endogenous outcome and as it is related to the aspect of time preference, the presented framework might help to further integrate behavioral economics and macroeconomics.

Exploiting non-Markovianity for quantum control

Quantum technology, exploiting entanglement and the wave nature of matter, relies on the ability to accurately control quantum systems. Quantum control is often compromised by the interaction of the system with its environment since this causes loss of amplitude and phase. However, when the dynamics of the open quantum system is non-Markovian, amplitude and phase flow not only from the system into the environment but also back. Interaction with the environment is then not necessarily detrimental. We show that the back-flow of amplitude and phase can be exploited to carry out quantum control tasks that could not be realized if the system was isolated. The control is facilitated by a few strongly coupled, sufficiently isolated environmental modes. Our paradigmatic example considers a weakly anharmonic ladder with resonant amplitude control only, restricting realizable operations to SO(N). The coupling to the environment, when harnessed with optimization techniques, allows for full SU(N) controllability.

A Concurrent Smalltalk Compiler for the Message-Driven Processor

This thesis describes Optimist, an optimizing compiler for the Concurrent Smalltalk language developed by the Concurrent VLSI Architecture Group. Optimist compiles Concurrent Smalltalk to the assembly language of the Message-Driven Processor (MDP). The compiler includes numerous optimization techniques such as dead code elimination, dataflow analysis, constant folding, move elimination, concurrency analysis, duplicate code merging, tail forwarding, use of register variables, as well as various MDP-specific optimizations in the code generator. The MDP presents some unique challenges and opportunities for compilation. Due to the MDP's small memory size, it is critical that the size of the generated code be as small as possible. The MDP is an inherently concurrent processor with efficient mechanisms for sending and receiving messages; the compiler takes advantage of these mechanisms. The MDP's tagged architecture allows very efficient support of object-oriented languages such as Concurrent Smalltalk. The initial goals for the MDP were to have the MDP execute about twenty instructions per method and contain 4096 words of memory. This compiler shows that these goals are too optimistic -- most methods are longer, both in terms of code size and running time. Thus, the memory size of the MDP should be increased.

Self-organized sociopolitical interactions as the best way to achieve organized patterns in human social systems: going beyond the top-down control of classical political regimes

La monografía presenta la auto-organización sociopolítica como la mejor manera de lograr patrones organizados en los sistemas sociales humanos, dada su naturaleza compleja y la imposibilidad de las tareas computacionales de los regímenes políticos clásico, debido a que operan con control jerárquico, el cual ha demostrado no ser óptimo en la producción de orden en los sistemas sociales humanos. En la monografía se extrapola la teoría de la auto-organización en los sistemas biológicos a las dinámicas sociopolíticas humanas, buscando maneras óptimas de organizarlas, y se afirma que redes complejas anárquicas son la estructura emergente de la auto-organización sociopolítica.

Desarrollo de un estado de la cuestión acerca de las interacciones entre el control de gestión y el aprendizaje organizacional.

Las organizaciones en la actualidad deben encontrar diferentes maneras de sobrevivir en un tiempo de rápida transformación. Uno de los mecanismos usados por las empresas para adaptarse a los cambios organizacionales son los sistemas de control de gestión, que a su vez permiten a las organizaciones hacer un seguimiento a sus procesos, para que la adaptabilidad sea efectiva. Otra variable importante para la adaptación es el aprendizaje organizacional siendo el proceso mediante el cual las organizaciones se adaptan a los cambios del entorno, tanto interno como externo de la compañía. Dado lo anterior, este proyecto se basa en la extracción de documentación soporte valido, que permita explorar las interacciones entre estos dos campos, los sistemas de control de gestión y el aprendizaje organizacional, además, analizar el impacto de estas interacciones en la perdurabilidad organizacional. ​

Optimal management of natural resources. Accounting for heterogeneity

Dynamic optimization methods have become increasingly important over the last years in economics. Within the dynamic optimization techniques employed, optimal control has emerged as the most powerful tool for the theoretical economic analysis. However, there is the need to advance further and take account that many dynamic economic processes are, in addition, dependent on some other parameter different than time. One can think of relaxing the assumption of a representative (homogeneous) agent in macro- and micro-economic applications allowing for heterogeneity among the agents. For instance, the optimal adaptation and diffusion of a new technology over time, may depend on the age of the person that adopted the new technology. Therefore, the economic models must take account of heterogeneity conditions within the dynamic framework. This thesis intends to accomplish two goals. The first goal is to analyze and revise existing environmental policies that focus on defining the optimal management of natural resources over time, by taking account of the heterogeneity of environmental conditions. Thus, the thesis makes a policy orientated contribution in the field of environmental policy by defining the necessary changes to transform an environmental policy based on the assumption of homogeneity into an environmental policy which takes account of heterogeneity. As a result the newly defined environmental policy will be more efficient and likely also politically more acceptable since it is tailored more specifically to the heterogeneous environmental conditions. Additionally to its policy orientated contribution, this thesis aims making a methodological contribution by applying a new optimization technique for solving problems where the control variables depend on two or more arguments --- the so-called two-stage solution approach ---, and by applying a numerical method --- the Escalator Boxcar Train Method --- for solving distributed optimal control problems, i.e., problems where the state variables, in addition to the control variables, depend on two or more arguments. Chapter 2 presents a theoretical framework to determine optimal resource allocation over time for the production of a good by heterogeneous producers, who generate a stock externalit and derives government policies to modify the behavior of competitive producers in order to achieve optimality. Chapter 3 illustrates the method in a more specific context, and integrates the aspects of quality and time, presenting a theoretical model that allows to determine the socially optimal outcome over time and space for the problem of waterlogging in irrigated agricultural production. Chapter 4 of this thesis concentrates on forestry resources and analyses the optimal selective-logging regime of a size-distributed forest.

Implementing a generic systolic array for genetic algorithms

We have designed a highly parallel design for a simple genetic algorithm using a pipeline of systolic arrays. The systolic design provides high throughput and unidirectional pipelining by exploiting the implicit parallelism in the genetic operators. The design is significant because, unlike other hardware genetic algorithms, it is independent of both the fitness function and the particular chromosome length used in a problem. We have designed and simulated a version of the mutation array using Xilinix FPGA tools to investigate the feasibility of hardware implementation. A simple 5-chromosome mutation array occupies 195 CLBs and is capable of performing more than one million mutations per second. I. Introduction Genetic algorithms (GAs) are established search and optimization techniques which have been applied to a range of engineering and applied problems with considerable success [1]. They operate by maintaining a population of trial solutions encoded, using a suitable encoding scheme.

Nonlinear system identification using particle swarm optimisation tuned radial basis function models

A novel particle swarm optimisation (PSO) tuned radial basis function (RBF) network model is proposed for identification of non-linear systems. At each stage of orthogonal forward regression (OFR) model construction process, PSO is adopted to tune one RBF unit's centre vector and diagonal covariance matrix by minimising the leave-one-out (LOO) mean square error (MSE). This PSO aided OFR automatically determines how many tunable RBF nodes are sufficient for modelling. Compared with the-state-of-the-art local regularisation assisted orthogonal least squares algorithm based on the LOO MSE criterion for constructing fixed-node RBF network models, the PSO tuned RBF model construction produces more parsimonious RBF models with better generalisation performance and is often more efficient in model construction. The effectiveness of the proposed PSO aided OFR algorithm for constructing tunable node RBF models is demonstrated using three real data sets.

Expansion quality of epidemic protocols

Epidemic protocols are a bio-inspired communication and computation paradigm for large and extreme-scale networked systems. This work investigates the expansion property of the network overlay topologies induced by epidemic protocols. An expansion quality index for overlay topologies is proposed and adopted for the design of epidemic membership protocols. A novel protocol is proposed, which explicitly aims at improving the expansion quality of the overlay topologies. The proposed protocol is tested with a global aggregation task and compared to other membership protocols. The analysis by means of simulations indicates that the expansion quality directly relates to the speed of dissemination and convergence of epidemic protocols and can be effectively used to design better protocols.

Body-centric modelling, identification, and acceleration tracking control of a quadrotor UAV

This paper presents the mathematical development of a body-centric nonlinear dynamic model of a quadrotor UAV that is suitable for the development of biologically inspired navigation strategies. Analytical approximations are used to find an initial guess of the parameters of the nonlinear model, then parameter estimation methods are used to refine the model parameters using the data obtained from onboard sensors during flight. Due to the unstable nature of the quadrotor model, the identification process is performed with the system in closed-loop control of attitude angles. The obtained model parameters are validated using real unseen experimental data. Based on the identified model, a Linear-Quadratic (LQ) optimal tracker is designed to stabilize the quadrotor and facilitate its translational control by tracking body accelerations. The LQ tracker is tested on an experimental quadrotor UAV and the obtained results are a further means to validate the quality of the estimated model. The unique formulation of the control problem in the body frame makes the controller better suited for bio-inspired navigation and guidance strategies than conventional attitude or position based control systems that can be found in the existing literature.

Lipopeptides: from self-assembly to bioactivity

This Feature Article discusses several classes of lipopeptide with important biomedical applications as antimicrobial and antifungal agents, in immune therapies and in personal care applications among others. Two main classes of lipopeptide are considered: (i) bacterially-expressed lipopeptides with a cyclic peptide headgroup and (ii) linear lipopeptides (with one or more lipid chains) based on bio-derived and bio-inspired amino acid sequences with current clinical applications. The applications are briefly summarized, and the biophysical characterization of the molecules is reviewed, with a particular focus on self-assembly. For several of these types of biomolecule, the formation of micelles above a critical micelle concentration has been observed while others form bilayer structures, depending on conditions of pH and temperature. As yet, there are few studies on the possible relationship between self-assembly into structures such as micelles and bioactivity of this class of molecule although this is likely to attract further attention.