Cost function adaptation: a stochastic gradient algorithm for data echo cancellation

A family of stochastic gradient algorithms and their behaviour in the data echo cancellation work platform are presented. The cost function adaptation algorithms use an error exponent update strategy based on an absolute error mapping, which is updated at every iteration. The quadratic and nonquadratic cost functions are special cases of the new family. Several possible realisations are introduced using these approaches. The noisy error problem is discussed and the digital recursive filter estimator is proposed. The simulation outcomes confirm the effectiveness of the proposed family of algorithms.

Segmentation of optic disc in retinal images using an improved gradient vector flow algorithm

Image segmentation plays an important role in the analysis of retinal images as the extraction of the optic disk provides important cues for accurate diagnosis of various retinopathic diseases. In recent years, gradient vector flow (GVF) based algorithms have been used successfully to successfully segment a variety of medical imagery. However, due to the compromise of internal and external energy forces within the resulting partial differential equations, these methods can lead to less accurate segmentation results in certain cases. In this paper, we propose the use of a new mean shift-based GVF segmentation algorithm that drives the internal/external energies towards the correct direction. The proposed method incorporates a mean shift operation within the standard GVF cost function to arrive at a more accurate segmentation. Experimental results on a large dataset of retinal images demonstrate that the presented method optimally detects the border of the optic disc.

Mean shift based gradient vector flow for image segmentation

In recent years, gradient vector flow (GVF) based algorithms have been successfully used to segment a variety of 2-D and 3-D imagery. However, due to the compromise of internal and external energy forces within the resulting partial differential equations, these methods may lead to biased segmentation results. In this paper, we propose MSGVF, a mean shift based GVF segmentation algorithm that can successfully locate the correct borders. MSGVF is developed so that when the contour reaches equilibrium, the various forces resulting from the different energy terms are balanced. In addition, the smoothness constraint of image pixels is kept so that over- or under-segmentation can be reduced. Experimental results on publicly accessible datasets of dermoscopic and optic disc images demonstrate that the proposed method effectively detects the borders of the objects of interest.

Semi batch learning with store management using enhanced conjugate gradient

This paper explores the performance of sliding-window based training, termed as semi batch, using multilayer perceptron (MLP) neural network in the presence of correlated data. The sliding window training is a form of higher order instantaneous learning strategy without the need of covariance matrix, usually employed for modeling and tracking purposes. Sliding-window framework is implemented to combine the robustness of offline learning algorithms with the ability to track online the underlying process of a function. This paper adopted sliding window training with recent advances in conjugate gradient direction with application of data store management e.g. simple distance measure, angle evaluation and the novel prediction error test. The simulation results show the best convergence performance is gained by using store management techniques. © 2012 Springer-Verlag.

Training neuro-fuzzy models using evolution based algorithms

The normal design process for neural networks or fuzzy systems involve two different phases: the determination of the best topology, which can be seen as a system identification problem, and the determination of its parameters, which can be envisaged as a parameter estimation problem. This latter issue, the determination of the model parameters (linear weights and interior knots) is the simplest task and is usually solved using gradient or hybrid schemes. The former issue, the topology determination, is an extremely complex task, especially if dealing with real-world problems.

Dynamic Facility Location with Modular Capacities : Models, Algorithms and Applications in Forestry

Les décisions de localisation sont souvent soumises à des aspects dynamiques comme des changements dans la demande des clients. Pour y répondre, la solution consiste à considérer une flexibilité accrue concernant l’emplacement et la capacité des installations. Même lorsque la demande est prévisible, trouver le planning optimal pour le déploiement et l'ajustement dynamique des capacités reste un défi. Dans cette thèse, nous nous concentrons sur des problèmes de localisation avec périodes multiples, et permettant l'ajustement dynamique des capacités, en particulier ceux avec des structures de coûts complexes. Nous étudions ces problèmes sous différents points de vue de recherche opérationnelle, en présentant et en comparant plusieurs modèles de programmation linéaire en nombres entiers (PLNE), l'évaluation de leur utilisation dans la pratique et en développant des algorithmes de résolution efficaces. Cette thèse est divisée en quatre parties. Tout d’abord, nous présentons le contexte industriel à l’origine de nos travaux: une compagnie forestière qui a besoin de localiser des campements pour accueillir les travailleurs forestiers. Nous présentons un modèle PLNE permettant la construction de nouveaux campements, l’extension, le déplacement et la fermeture temporaire partielle des campements existants. Ce modèle utilise des contraintes de capacité particulières, ainsi qu’une structure de coût à économie d’échelle sur plusieurs niveaux. L'utilité du modèle est évaluée par deux études de cas. La deuxième partie introduit le problème dynamique de localisation avec des capacités modulaires généralisées. Le modèle généralise plusieurs problèmes dynamiques de localisation et fournit de meilleures bornes de la relaxation linéaire que leurs formulations spécialisées. Le modèle peut résoudre des problèmes de localisation où les coûts pour les changements de capacité sont définis pour toutes les paires de niveaux de capacité, comme c'est le cas dans le problème industriel mentionnée ci-dessus. Il est appliqué à trois cas particuliers: l'expansion et la réduction des capacités, la fermeture temporaire des installations, et la combinaison des deux. Nous démontrons des relations de dominance entre notre formulation et les modèles existants pour les cas particuliers. Des expériences de calcul sur un grand nombre d’instances générées aléatoirement jusqu’à 100 installations et 1000 clients, montrent que notre modèle peut obtenir des solutions optimales plus rapidement que les formulations spécialisées existantes. Compte tenu de la complexité des modèles précédents pour les grandes instances, la troisième partie de la thèse propose des heuristiques lagrangiennes. Basées sur les méthodes du sous-gradient et des faisceaux, elles trouvent des solutions de bonne qualité même pour les instances de grande taille comportant jusqu’à 250 installations et 1000 clients. Nous améliorons ensuite la qualité de la solution obtenue en résolvent un modèle PLNE restreint qui tire parti des informations recueillies lors de la résolution du dual lagrangien. Les résultats des calculs montrent que les heuristiques donnent rapidement des solutions de bonne qualité, même pour les instances où les solveurs génériques ne trouvent pas de solutions réalisables. Finalement, nous adaptons les heuristiques précédentes pour résoudre le problème industriel. Deux relaxations différentes sont proposées et comparées. Des extensions des concepts précédents sont présentées afin d'assurer une résolution fiable en un temps raisonnable.

From Genetic Algorithms to Efficient Organization

The work described in this thesis began as an inquiry into the nature and use of optimization programs based on "genetic algorithms." That inquiry led, eventually, to three powerful heuristics that are broadly applicable in gradient-ascent programs: First, remember the locations of local maxima and restart the optimization program at a place distant from previously located local maxima. Second, adjust the size of probing steps to suit the local nature of the terrain, shrinking when probes do poorly and growing when probes do well. And third, keep track of the directions of recent successes, so as to probe preferentially in the direction of most rapid ascent. These algorithms lie at the core of a novel optimization program that illustrates the power to be had from deploying them together. The efficacy of this program is demonstrated on several test problems selected from a variety of fields, including De Jong's famous test-problem suite, the traveling salesman problem, the problem of coordinate registration for image guided surgery, the energy minimization problem for determining the shape of organic molecules, and the problem of assessing the structure of sedimentary deposits using seismic data.

Policy gradient based Reinforcement Learning for real autonomous underwater cable tracking

This paper proposes a field application of a high-level reinforcement learning (RL) control system for solving the action selection problem of an autonomous robot in cable tracking task. The learning system is characterized by using a direct policy search method for learning the internal state/action mapping. Policy only algorithms may suffer from long convergence times when dealing with real robotics. In order to speed up the process, the learning phase has been carried out in a simulated environment and, in a second step, the policy has been transferred and tested successfully on a real robot. Future steps plan to continue the learning process on-line while on the real robot while performing the mentioned task. We demonstrate its feasibility with real experiments on the underwater robot ICTINEU AUV

Gradient-based reinforcement learning techniques for underwater robotics behavior learning

Darrerament, l'interès pel desenvolupament d'aplicacions amb robots submarins autònoms (AUV) ha crescut de forma considerable. Els AUVs són atractius gràcies al seu tamany i el fet que no necessiten un operador humà per pilotar-los. Tot i això, és impossible comparar, en termes d'eficiència i flexibilitat, l'habilitat d'un pilot humà amb les escasses capacitats operatives que ofereixen els AUVs actuals. L'utilització de AUVs per cobrir grans àrees implica resoldre problemes complexos, especialment si es desitja que el nostre robot reaccioni en temps real a canvis sobtats en les condicions de treball. Per aquestes raons, el desenvolupament de sistemes de control autònom amb l'objectiu de millorar aquestes capacitats ha esdevingut una prioritat. Aquesta tesi tracta sobre el problema de la presa de decisions utilizant AUVs. El treball presentat es centra en l'estudi, disseny i aplicació de comportaments per a AUVs utilitzant tècniques d'aprenentatge per reforç (RL). La contribució principal d'aquesta tesi consisteix en l'aplicació de diverses tècniques de RL per tal de millorar l'autonomia dels robots submarins, amb l'objectiu final de demostrar la viabilitat d'aquests algoritmes per aprendre tasques submarines autònomes en temps real. En RL, el robot intenta maximitzar un reforç escalar obtingut com a conseqüència de la seva interacció amb l'entorn. L'objectiu és trobar una política òptima que relaciona tots els estats possibles amb les accions a executar per a cada estat que maximitzen la suma de reforços totals. Així, aquesta tesi investiga principalment dues tipologies d'algoritmes basats en RL: mètodes basats en funcions de valor (VF) i mètodes basats en el gradient (PG). Els resultats experimentals finals mostren el robot submarí Ictineu en una tasca autònoma real de seguiment de cables submarins. Per portar-la a terme, s'ha dissenyat un algoritme anomenat mètode d'Actor i Crític (AC), fruit de la fusió de mètodes VF amb tècniques de PG.

On merging gradient estimation with mean-tracking techniques for cluster identification

This paper discusses how numerical gradient estimation methods may be used in order to reduce the computational demands on a class of multidimensional clustering algorithms. The study is motivated by the recognition that several current point-density based cluster identification algorithms could benefit from a reduction of computational demand if approximate a-priori estimates of the cluster centres present in a given data set could be supplied as starting conditions for these algorithms. In this particular presentation, the algorithm shown to benefit from the technique is the Mean-Tracking (M-T) cluster algorithm, but the results obtained from the gradient estimation approach may also be applied to other clustering algorithms and their related disciplines.

Standardized test to evaluate numerical weather prediction algorithms

In order to assist in comparing the computational techniques used in different models, the authors propose a standardized set of one-dimensional numerical experiments that could be completed for each model. The results of these experiments, with a simplified form of the computational representation for advection, diffusion, pressure gradient term, Coriolis term, and filter used in the models, should be reported in the peer-reviewed literature. Specific recommendations are described in this paper.

Structured minimal-memory inexact quasi-Newton method and secant preconditioners for augmented Lagrangian optimization

Augmented Lagrangian methods for large-scale optimization usually require efficient algorithms for minimization with box constraints. On the other hand, active-set box-constraint methods employ unconstrained optimization algorithms for minimization inside the faces of the box. Several approaches may be employed for computing internal search directions in the large-scale case. In this paper a minimal-memory quasi-Newton approach with secant preconditioners is proposed, taking into account the structure of Augmented Lagrangians that come from the popular Powell-Hestenes-Rockafellar scheme. A combined algorithm, that uses the quasi-Newton formula or a truncated-Newton procedure, depending on the presence of active constraints in the penalty-Lagrangian function, is also suggested. Numerical experiments using the Cute collection are presented.

Parallelization of the discrete gradient method of non-smooth optimization and its applications

We investigate parallelization and performance of the discrete gradient method of nonsmooth optimization. This derivative free method is shown to be an effective optimization tool, able to skip many shallow local minima of nonconvex nondifferentiable objective functions. Although this is a sequential iterative method, we were able to parallelize critical steps of the algorithm, and this lead to a significant improvement in performance on multiprocessor computer clusters. We applied this method to a difficult polyatomic clusters problem in computational chemistry, and found this method to outperform other algorithms.

Frequency-domain beamformers using conjugate gradient techniques for speech enhancement

A multiple-iteration constrained conjugate gradient (MICCG) algorithm and a single-iteration constrained conjugate gradient (SICCG) algorithm are proposed to realize the widely used frequency-domain minimum-variance-distortionless-response (MVDR) beamformers and the resulting algorithms are applied to speech enhancement. The algorithms are derived based on the Lagrange method and the conjugate gradient techniques. The implementations of the algorithms avoid any form of explicit or implicit autocorrelation matrix inversion. Theoretical analysis establishes formal convergence of the algorithms. Specifically, the MICCG algorithm is developed based on a block adaptation approach and it generates a finite sequence of estimates that converge to the MVDR solution. For limited data records, the estimates of the MICCG algorithm are better than the conventional estimators and equivalent to the auxiliary vector algorithms. The SICCG algorithm is developed based on a continuous adaptation approach with a sample-by-sample updating procedure and the estimates asymptotically converge to the MVDR solution. An illustrative example using synthetic data from a uniform linear array is studied and an evaluation on real data recorded by an acoustic vector sensor array is demonstrated. Performance of the MICCG algorithm and the SICCG algorithm are compared with the state-of-the-art approaches.

Multistep stochastic mirror descent for risk-averse convex stochastic programs based on extended polyhedral risk measures

We consider risk-averse convex stochastic programs expressed in terms of extended polyhedral risk measures. We derive computable con dence intervals on the optimal value of such stochastic programs using the Robust Stochastic Approximation and the Stochastic Mirror Descent (SMD) algorithms. When the objective functions are uniformly convex, we also propose a multistep extension of the Stochastic Mirror Descent algorithm and obtain con dence intervals on both the optimal values and optimal solutions. Numerical simulations show that our con dence intervals are much less conservative and are quicker to compute than previously obtained con dence intervals for SMD and that the multistep Stochastic Mirror Descent algorithm can obtain a good approximate solution much quicker than its nonmultistep counterpart. Our con dence intervals are also more reliable than asymptotic con dence intervals when the sample size is not much larger than the problem size.