Example Based Image Analysis and Synthesis

Image analysis and graphics synthesis can be achieved with learning techniques using directly image examples without physically-based, 3D models. In our technique: -- the mapping from novel images to a vector of "pose" and "expression" parameters can be learned from a small set of example images using a function approximation technique that we call an analysis network; -- the inverse mapping from input "pose" and "expression" parameters to output images can be synthesized from a small set of example images and used to produce new images using a similar synthesis network. The techniques described here have several applications in computer graphics, special effects, interactive multimedia and very low bandwidth teleconferencing.

Policy Improvement for POMDPs Using Normalized Importance Sampling

We present a new method for estimating the expected return of a POMDP from experience. The estimator does not assume any knowle ge of the POMDP and allows the experience to be gathered with an arbitrary set of policies. The return is estimated for any new policy of the POMDP. We motivate the estimator from function-approximation and importance sampling points-of-view and derive its theoretical properties. Although the estimator is biased, it has low variance and the bias is often irrelevant when the estimator is used for pair-wise comparisons.We conclude by extending the estimator to policies with memory and compare its performance in a greedy search algorithm to the REINFORCE algorithm showing an order of magnitude reduction in the number of trials required.

An artificial economy based on reinforcement learning and agent based modeling

In this paper, we employ techniques from artificial intelligence such as reinforcement learning and agent based modeling as building blocks of a computational model for an economy based on conventions. First we model the interaction among firms in the private sector. These firms behave in an information environment based on conventions, meaning that a firm is likely to behave as its neighbors if it observes that their actions lead to a good pay off. On the other hand, we propose the use of reinforcement learning as a computational model for the role of the government in the economy, as the agent that determines the fiscal policy, and whose objective is to maximize the growth of the economy. We present the implementation of a simulator of the proposed model based on SWARM, that employs the SARSA(λ) algorithm combined with a multilayer perceptron as the function approximation for the action value function.

Statistical deconvolution of enthalpic energetic contributions to MHC-peptide binding affinity

Background: MHC Class I molecules present antigenic peptides to cytotoxic T cells, which forms an integral part of the adaptive immune response. Peptides are bound within a groove formed by the MHC heavy chain. Previous approaches to MHC Class I-peptide binding prediction have largely concentrated on the peptide anchor residues located at the P2 and C-terminus positions. Results: A large dataset comprising MHC-peptide structural complexes was created by remodelling pre-determined x-ray crystallographic structures. Static energetic analysis, following energy minimisation, was performed on the dataset in order to characterise interactions between bound peptides and the MHC Class I molecule, partitioning the interactions within the groove into van der Waals, electrostatic and total non-bonded energy contributions. Conclusion: The QSAR techniques of Genetic Function Approximation (GFA) and Genetic Partial Least Squares (G/PLS) algorithms were used to identify key interactions between the two molecules by comparing the calculated energy values with experimentally-determined BL50 data. Although the peptide termini binding interactions help ensure the stability of the MHC Class I-peptide complex, the central region of the peptide is also important in defining the specificity of the interaction. As thermodynamic studies indicate that peptide association and dissociation may be driven entropically, it may be necessary to incorporate entropic contributions into future calculations.

An approximate dynamic programming approach for improving accuracy of lossy data compression by Bloom filters

Bloom filters are a data structure for storing data in a compressed form. They offer excellent space and time efficiency at the cost of some loss of accuracy (so-called lossy compression). This work presents a yes-no Bloom filter, which as a data structure consisting of two parts: the yes-filter which is a standard Bloom filter and the no-filter which is another Bloom filter whose purpose is to represent those objects that were recognised incorrectly by the yes-filter (that is, to recognise the false positives of the yes-filter). By querying the no-filter after an object has been recognised by the yes-filter, we get a chance of rejecting it, which improves the accuracy of data recognition in comparison with the standard Bloom filter of the same total length. A further increase in accuracy is possible if one chooses objects to include in the no-filter so that the no-filter recognises as many as possible false positives but no true positives, thus producing the most accurate yes-no Bloom filter among all yes-no Bloom filters. This paper studies how optimization techniques can be used to maximize the number of false positives recognised by the no-filter, with the constraint being that it should recognise no true positives. To achieve this aim, an Integer Linear Program (ILP) is proposed for the optimal selection of false positives. In practice the problem size is normally large leading to intractable optimal solution. Considering the similarity of the ILP with the Multidimensional Knapsack Problem, an Approximate Dynamic Programming (ADP) model is developed making use of a reduced ILP for the value function approximation. Numerical results show the ADP model works best comparing with a number of heuristics as well as the CPLEX built-in solver (B&B), and this is what can be recommended for use in yes-no Bloom filters. In a wider context of the study of lossy compression algorithms, our researchis an example showing how the arsenal of optimization methods can be applied to improving the accuracy of compressed data.

Algoritmos inteligentes para auto-correção, auto-verificação e auto-validação de medidas em sensores

This work presents a set of intelligent algorithms with the purpose of correcting calibration errors in sensors and reducting the periodicity of their calibrations. Such algorithms were designed using Artificial Neural Networks due to its great capacity of learning, adaptation and function approximation. Two approaches willbe shown, the firstone uses Multilayer Perceptron Networks to approximate the many shapes of the calibration curve of a sensor which discalibrates in different time points. This approach requires the knowledge of the sensor s functioning time, but this information is not always available. To overcome this need, another approach using Recurrent Neural Networks was proposed. The Recurrent Neural Networks have a great capacity of learning the dynamics of a system to which it was trained, so they can learn the dynamics of a sensor s discalibration. Knowingthe sensor s functioning time or its discalibration dynamics, it is possible to determine how much a sensor is discalibrated and correct its measured value, providing then, a more exact measurement. The algorithms proposed in this work can be implemented in a Foundation Fieldbus industrial network environment, which has a good capacity of device programming through its function blocks, making it possible to have them applied to the measurement process

Estratégias baseadas em aprendizado para coordenação de uma frota de robôs em tarefas cooperativas

In multi-robot systems, both control architecture and work strategy represent a challenge for researchers. It is important to have a robust architecture that can be easily adapted to requirement changes. It is also important that work strategy allows robots to complete tasks efficiently, considering that robots interact directly in environments with humans. In this context, this work explores two approaches for robot soccer team coordination for cooperative tasks development. Both approaches are based on a combination of imitation learning and reinforcement learning. Thus, in the first approach was developed a control architecture, a fuzzy inference engine for recognizing situations in robot soccer games, a software for narration of robot soccer games based on the inference engine and the implementation of learning by imitation from observation and analysis of others robotic teams. Moreover, state abstraction was efficiently implemented in reinforcement learning applied to the robot soccer standard problem. Finally, reinforcement learning was implemented in a form where actions are explored only in some states (for example, states where an specialist robot system used them) differently to the traditional form, where actions have to be tested in all states. In the second approach reinforcement learning was implemented with function approximation, for which an algorithm called RBF-Sarsa($lambda$) was created. In both approaches batch reinforcement learning algorithms were implemented and imitation learning was used as a seed for reinforcement learning. Moreover, learning from robotic teams controlled by humans was explored. The proposal in this work had revealed efficient in the robot soccer standard problem and, when implemented in other robotics systems, they will allow that these robotics systems can efficiently and effectively develop assigned tasks. These approaches will give high adaptation capabilities to requirements and environment changes.

Comparative study between RBF and radial-PPS neural networks

The study of function approximation is motivated by the human limitation and inability to register and manipulate with exact precision the behavior variations of the physical nature of a phenomenon. These variations are referred to as signals or signal functions. Many real world problem can be formulated as function approximation problems and from the viewpoint of artificial neural networks these can be seen as the problem of searching for a mapping that establishes a relationship from an input space to an output space through a process of network learning. Several paradigms of artificial neural networks (ANN) exist. Here we will be investigated a comparative of the ANN study of RBF with radial Polynomial Power of Sigmoids (PPS) in function approximation problems. Radial PPS are functions generated by linear combination of powers of sigmoids functions. The main objective of this paper is to show the advantages of the use of the radial PPS functions in relationship traditional RBF, through adaptive training and ridge regression techniques.

Multidimensional polynomial powers of sigmoid (PPS) Wavelet neural networks

Wavelet functions have been used as the activation function in feedforward neural networks. An abundance of R&D has been produced on wavelet neural network area. Some successful algorithms and applications in wavelet neural network have been developed and reported in the literature. However, most of the aforementioned reports impose many restrictions in the classical backpropagation algorithm, such as low dimensionality, tensor product of wavelets, parameters initialization, and, in general, the output is one dimensional, etc. In order to remove some of these restrictions, a family of polynomial wavelets generated from powers of sigmoid functions is presented. We described how a multidimensional wavelet neural networks based on these functions can be constructed, trained and applied in pattern recognition tasks. As an example of application for the method proposed, it is studied the exclusive-or (XOR) problem.

Feedforward neural networks based on PPS-wavelet activation functions

Function approximation is a very important task in environments where computation has to be based on extracting information from data samples in real world processes. Neural networks and wavenets have been recently seen as attractive tools for developing efficient solutions for many real world problems in function approximation. In this paper, it is shown how feedforward neural networks can be built using a different type of activation function referred to as the PPS-wavelet. An algorithm is presented to generate a family of PPS-wavelets that can be used to efficiently construct feedforward networks for function approximation.

Proposta de implementação de redes de base radial em tecnologias CMOS e BiCMOS

Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES)

Aeroelastic stability analysis using linear matrix inequalities

The present work describes an alternative methodology for identification of aeroelastic stability in a range of varying parameters. Analysis is performed in time domain based on Lyapunov stability and solved by convex optimization algorithms. The theory is outlined and simulations are carried out on a benchmark system to illustrate the method. The classical methodology with the analysis of the system's eigenvalues is presented for comparing the results and validating the approach. The aeroelastic model is represented in state space format and the unsteady aerodynamic forces are written in time domain using rational function approximation. The problem is formulated as a polytopic differential inclusion system and the conceptual idea can be used in two different applications. In the first application the method verifies the aeroelastic stability in a range of air density (or its equivalent altitude range). In the second one, the stability is verified for a rage of velocities. These analyses are in contrast to the classical discrete analysis performed at fixed air density/velocity values. It is shown that this method is efficient to identify stability regions in the flight envelope and it offers promise for robust flutter identification.

Control of Limit Cycle Oscillation in a Three Degrees of Freedom Airfoil Section Using Fuzzy Takagi-Sugeno Modeling

This work presents a strategy to control nonlinear responses of aeroelastic systems with control surface freeplay. The proposed methodology is developed for the three degrees of freedom typical section airfoil considering aerodynamic forces from Theodorsen's theory. The mathematical model is written in the state space representation using rational function approximation to write the aerodynamic forces in time domain. The control system is designed using the fuzzy Takagi-Sugeno modeling to compute a feedback control gain. It useds Lyapunov's stability function and linear matrix inequalities (LMIs) to solve a convex optimization problem. Time simulations with different initial conditions are performed using a modified Runge-Kutta algorithm to compare the system with and without control forces. It is shown that this approach can compute linear control gain able to stabilize aeroelastic systems with discontinuous nonlinearities.

Aeroelastic stability analysis considering a continuous flight envelope

This paper presents a new methodology to analyze aeroelastic stability in a continuous range of flight envelope with varying parameter of velocity and altitude. The focus of the paper is to demonstrate that linear matrix inequalities can be used to evaluate the aeroelastic stability in a region of flight envelope instead of a single point, like classical methods. The proposed methodology can also be used to study if a system remains stable during an arbitrary motion from one point to another in the flight envelope, i.e., when the problem becomes time-variant. The main idea is to represent the system as a polytopic differential inclusion system using rational function approximation to write the model in time domain. The theory is outlined and simulations are carried out on the benchmark AGARD 445.6 wing to demonstrate the method. The classical pk-method is used for comparing results and validating the approach. It is shown that this method is efficient to identify stability regions in the flight envelope. (C) 2014 Elsevier Ltd. All rights reserved.

Lernen mit einer Hybride aus Lernendem Klassifizierendem System und Selbstorganisierender Karte

Im Forschungsgebiet der Künstlichen Intelligenz, insbesondere im Bereich des maschinellen Lernens, hat sich eine ganze Reihe von Verfahren etabliert, die von biologischen Vorbildern inspiriert sind. Die prominentesten Vertreter derartiger Verfahren sind zum einen Evolutionäre Algorithmen, zum anderen Künstliche Neuronale Netze. Die vorliegende Arbeit befasst sich mit der Entwicklung eines Systems zum maschinellen Lernen, das Charakteristika beider Paradigmen in sich vereint: Das Hybride Lernende Klassifizierende System (HCS) wird basierend auf dem reellwertig kodierten eXtended Learning Classifier System (XCS), das als Lernmechanismus einen Genetischen Algorithmus enthält, und dem Wachsenden Neuralen Gas (GNG) entwickelt. Wie das XCS evolviert auch das HCS mit Hilfe eines Genetischen Algorithmus eine Population von Klassifizierern - das sind Regeln der Form [WENN Bedingung DANN Aktion], wobei die Bedingung angibt, in welchem Bereich des Zustandsraumes eines Lernproblems ein Klassifizierer anwendbar ist. Beim XCS spezifiziert die Bedingung in der Regel einen achsenparallelen Hyperquader, was oftmals keine angemessene Unterteilung des Zustandsraumes erlaubt. Beim HCS hingegen werden die Bedingungen der Klassifizierer durch Gewichtsvektoren beschrieben, wie die Neuronen des GNG sie besitzen. Jeder Klassifizierer ist anwendbar in seiner Zelle der durch die Population des HCS induzierten Voronoizerlegung des Zustandsraumes, dieser kann also flexibler unterteilt werden als beim XCS. Die Verwendung von Gewichtsvektoren ermöglicht ferner, einen vom Neuronenadaptationsverfahren des GNG abgeleiteten Mechanismus als zweites Lernverfahren neben dem Genetischen Algorithmus einzusetzen. Während das Lernen beim XCS rein evolutionär erfolgt, also nur durch Erzeugen neuer Klassifizierer, ermöglicht dies dem HCS, bereits vorhandene Klassifizierer anzupassen und zu verbessern. Zur Evaluation des HCS werden mit diesem verschiedene Lern-Experimente durchgeführt. Die Leistungsfähigkeit des Ansatzes wird in einer Reihe von Lernproblemen aus den Bereichen der Klassifikation, der Funktionsapproximation und des Lernens von Aktionen in einer interaktiven Lernumgebung unter Beweis gestellt.