A popular computer puzzle, the game ofMinesweeperrequires its human players to have a mix of both luck and strategy to succeed. Analyzing these aspects more formally, in our research, we assessed the feasibility of a novel methodology based on reinforcement learning as an adequate approach to tackle the problem presented by this game. For this purpose, we employed multi-armed bandit algorithms which were carefully adapted in order to enable their use to define autonomous computational players, targeting to make the best use of some game peculiarities. After experimental evaluation, results showed that this approach was indeed successful, especially in smaller game boards, such as the standard beginner level. Despite this fact, the main contribution of this work is a detailed examination ofMinesweeperfrom a learning perspective, which led to various original insights which are thoroughly discussed.

The choice of a fixed step size in adaptive filtering algorithms implies a conflict between the convergence rate and the steady-state performance. In order to address this tradeoff more effectively, variable step-size schemes have been proposed. The efficiency evaluation of such techniques requires comparisons of the resulting step-size values with theoretical optimum values obtained from a statistical analysis of the adaptive algorithm convergence. The analysis generally employs statistical approximations, the most critical being the assumption of independence between the input signal and the filter coefficients. In this letter, it is argued that such a comparison can be misleading because the supposedly optimal step-size sequence sometimes induces divergence in the initial phase of learning. This occurs most often when the input signal is colored and/or heavy tailed. This instability trend can be explained by a convergence analysis that does not employ the independence hypothesis. In addition, the use of this exact analysis implies an optimal step-size sequence that can be significantly different from that obtained with standard analysis methods. This approach can be used to improve the design process of variable step-size adaptive filtering algorithms.

Computer models have been an important tool to determine soil bulk density. This soil property is fundamental to estimate soil carbon reserves and consequently to understand the global carbon cycle. The estimation of soil bulk density is not a trivial task since it demands an intensive and often impractical work. The purpose of this paper is to evaluate the performance of a pedotransfer function against an Artificial Neural Networks to estimate soil bulk density for soils at Brazilian biomes. The first one consists of a linear model composed of a Least Square method. The latter employs a robust committee of multilayer perceptron networks and a model selection procedure based on k-fold cross-validation. The data are composed of 3404 soil layers distributed in different Brazilian regions and with different uses. The proposed non-linear regressor presents higher precision when compared to the linear model, and requires less information to do so. Additionally, the developed solution brings to light the assumed relationship between soil bulk density and some soil chemical properties.

In this work we consider a scenario in which several dispersed nodes intend to identify a nonlinear Volterra system. Such system is represented by a series that has sparse kernels, with few non-zero coefficients. This sparse feature of Volterra series allows us to propose distributed and sparsity- aware adaptive filtering algorithms, that aim at identifying such nonlinear system. We consider a network composed by agents, called Wireless Sensor Network (WSN), containing sensor nodes that have energy and computational constraints. We evaluate the performance of the proposed algorithm by means of the Mean-Squared Deviation (MSD) metric, and we verified that the distributed scheme provides a higher convergence rate and steady-state performance.

In recent years, the use of Unmanned Aerial Systems (UAS) has become commonplace in a wide variety of tasks due to their relatively low cost and ease of operation. In this paper, we explore the use of UAS in maritime Search And Rescue (SAR) missions by using experimental data to detect and classify objects at the sea surface. The objects are chosen as common objects present in maritime SAR missions: a boat, a pallet, a human, and a buoy. The data consists of thermal images and a Gaussian Mixture Model (GMM) is used to discriminate foreground objects from the background. Then, bounding boxes containing the object are defined and used to train a Convolutional Neural Network (CNN). The CNN achieves the average accuracy of 92.5% when evaluating a testing dataset.

Forest areas have an important role in the global carbon cycle. They are usually drains of greenhouse gases. From estimates of the soil bulk density Db, it is possible to determine the soil carbon stock. This relation shows the importance of determining Db. The purpose of this paper is to evaluate the performance of pedotransfer functions generated by a Linear Artificial Neural Network (ANN) to estimate Dbfor soils at Amazon Forest area. The raw data used was soil organic carbon (SOC), porosity, pH in water, sum of basic cations (SB) as predictor, Al3+and hydrogen (H). These data were taken from a soil survey report conducted in the Oil Province of the Urucu River, composed by 378 samples. The model proposed present good results when compared with other approaches presented in related works.

The wide availability of mobile devices with embedded microphones opens up opportunities for new applications based on acoustic sensor localization (ASL). Among them, this paper highlights mobile device self-localization relying exclusively on acoustic signals, but with previous knowledge of reference signals and source positions. The problem of finding the sensor position is stated as a function of estimated times-of-flight (TOFs) or time-differences-of-flight (TDOFs) from the sound sources to the target microphone, and the main practical issues involved in TOF estimation are discussed. Least-squares ASL solutions are introduced, followed by other strategies inspired by sound source localization solutions: steered-response power, which improves localization accuracy, and a new region-based search, which alleviates complexity. A set of complementary techniques for further improvement of TOF/TDOF estimates are reviewed: sliding windows, matching pursuit, and TOF selection. The paper proceeds with proposing a novel ASL method that combines most of the previous material, whose performance is assessed in a real-world example: in a typical lecture room, the method achieves accuracy better than 20 cm.

Self-localization of smart portable devices serves as foundation for several novel applications. This work proposes a set of algorithms that enable a mobile device to passively determine its position relative to a known reference with centimeter precision, based exclusively on the capture of acoustic signals emitted by controlled sources around it. The proposed techniques tackle typical practical issues such as reverberation, unknown speed of sound, line-of-sight obstruction, clock skew, and the need for asynchronous operation. After their theoretical developments and off-line simulations, the methods are assessed as real-time applications embedded into off-the-shelf mobile devices operating in real scenarios. When line of sight is available, position estimation errors are at most 4 cm using recorded signals.

The purpose of this paper was to evaluate the performance of pedotransfer functions generated by Radial Base Function (RBF) Artificial Neural Network (ANN) to estimate soil water retention at field capacity (FC, suction at -30 kPa) and Permanent Wilting Point (PWP, -1500 kPa) for soils at PROJIR area -RJ/BR. The raw data used was type of soil horizon, texture, bulk density, soil organic carbon content and porosity. These data were taken from RURALDATA database, composed of 218 soil profiles. The RBF ANN was trained through the Radial Basis Learning algorithm. The ANN generated to predict FC and PWP for PROJIR present better performance than the other approaches presented in the related works. The performance of ANN to predict PWP was higher than to FC. Neural network models present similar performance to the previously developed regression-type and, in general, the addition of porosity, bulk density data and horizon type, did not improve the performance of ANN.

Subband blind source separation methods have been recently proposed with the objective of reducing the computational complexity and improving the convergence rate of the online algorithms. Oversampled subband structures with DFT filter banks are usually employed in order to avoid aliasing effects and keep enough samples to estimate the statistics of the subband signals. In this paper we present a critically sampled subband structure, composed of cosine-modulated filter banks and reduced-order adaptive subfilters, for convolutive blind source separation. Its performance is compared to those of the fullband algorithm, of an oversampled subband algorithm and of a frequency-domain algorithm. We also evaluate, through computer simulations, the impact of reducing the order of the high-frequency subfilters on the separation results for different reverberation characteristics.

Frequency-domain blind source separation (BSS) techniques have been proposed with the objective of increasing the speed and/or reducing the computational complexity of conventional algorithms, mainly for applications that involve large convolutive mixture filters. In this letter, we present a new initialization method for frequency-domain BSS algorithms employing estimates of the directions of arrival and time-frequency masking, that outperforms the classical pre-whitening initialization technique. Performance and convergence time results of the proposed approach are presented using a frequency-domain BSS algorithm that exploits higher-order frequency dependencies, employing real conference room recordings and simulated data.