Jarmo Ilonen

Differential Evolution Training Algorithm for Feed-Forward Neural Networks

An evolutionary optimization method over continuous search spaces, differential evolution, has recently been successfully applied to real world and artificial optimization problems and proposed also for neural network training. However, differential evolution has not been comprehensively studied in the context of training neural network weights, i.e., how useful is differential evolution in finding the global optimum for expense of convergence speed. In this study, differential evolution has been analyzed as a candidate global optimization method for feed-forward neural networks. In comparison to gradient based methods, differential evolution seems not to provide any distinct advantage in terms of learning rate or solution quality. Differential evolution can rather be used in validation of reached optima and in the development of regularization terms and non-conventional transfer functions that do not necessarily provide gradient information.

Feature representation and discrimination based on Gaussian mixture model probability densities-Practices and algorithms

Highly active hydrofining catalysts are prepared by ion exchanging a silica-alumina hydrogel with an ammoniacal solution of a cobalt and/or nickel compound, and thereafter compositing the undried product with an alumina hydrogel and a molybdenum component, followed by drying and calcining. The resulting catalysts are particularly active for the denitrogenation of mineral oil feedstocks.

Diagnosis tool for motor condition monitoring

In the modern industrial environment there is increasing demand for automatic condition monitoring. With reliable condition monitoring, faults such as mechanical motor failures could be identified in their early stages and further damage to the system could be prevented. Successful monitoring is a complex and application-specific problem, but a generic tool would be useful in preliminary analysis of new signals and in verification of known theories. A generic condition diagnosis tool is introduced in this paper. The tool is based on discriminative energy functions which reveal discriminative frequency-domain regions where failures can be identified. The tool was applied to induction motor bearing fault detection and succeeded in finding characteristic frequencies which allow accurate detection of bearing faults.

Image Feature Localization by Multiple Hypothesis Testing of Gabor Features

Several novel and particularly successful object and object category detection and recognition methods based on image features, local descriptions of object appearance, have recently been proposed. The methods are based on a localization of image features and a spatial constellation search over the localized features. The accuracy and reliability of the methods depend on the success of both tasks: image feature localization and spatial constellation model search. In this paper, we present an improved algorithm for image feature localization. The method is based on complex-valued multiresolution Gabor features and their ranking using multiple hypothesis testing. The algorithm provides very accurate local image features over arbitrary scale and rotation. We discuss in detail issues such as selection of filter parameters, confidence measure, and the magnitude versus complex representation, and show on a large test sample how these influence the performance. The versatility and accuracy of the method is demonstrated on two profoundly different challenging problems (faces and license plates).

Improving similarity measures of histograms using smoothing projections

Selection of a proper similarity measure is an essential consideration for a success of many methods. In this study, similarity measures are analyzed in the context of ordered histogram type data, such as gray-level histograms of digital images or color spectra. Furthermore, the performance of the studied similarity measures can be improved using a smoothing projection, called neighbor-bank projection. Especially, with distance functions utilizing statistical properties of data, e.g., the Mahalanobis distance, a significant improvement was achieved in the classification experiments on real data sets, resulting from the use of a priori information related to ordered data. The proposed projection seems also to be applicable for dimensional reduction of histograms and to represent sparse data in a more tight form in the projection subspace.

Fusing visual and tactile sensing for 3-D object reconstruction while grasping

In this work, we propose to reconstruct a complete 3-D model of an unknown object by fusion of visual and tactile information while the object is grasped. Assuming the object is symmetric, a first hypothesis of its complete 3-D shape is generated from a single view. This initial model is used to plan a grasp on the object which is then executed with a robotic manipulator equipped with tactile sensors. Given the detected contacts between the fingers and the object, the full object model including the symmetry parameters can be refined. This refined model will then allow the planning of more complex manipulation tasks. The main contribution of this work is an optimal estimation approach for the fusion of visual and tactile data applying the constraint of object symmetry. The fusion is formulated as a state estimation problem and solved with an iterative extended Kalman filter. The approach is validated experimentally using both artificial and real data from two different robotic platforms.

Gaussian mixture pdf in one-class classification: computing and utilizing confidence values

In this study a confidence measure for probability density functions (pdfs) is presented. The measure can be used in one-class classification to select a pdf threshold for class inclusion. In addition, confidence information can be used to verify correctness of a decision in a multi-class case where for example the Bayesian decision rule reveals which class is the most probable. Additionally, using confidence values - which represent in which quantile of the probability mass a pdf value resides ([0,1]) - is often straightforward compared to using arbitrarily scaled pdf values. As the main contributions, use of confidence information in classification is described and a method for confidence estimation is presented

Three-dimensional object reconstruction of symmetric objects by fusing visual and tactile sensing

In this work, we propose to reconstruct a complete three-dimensional (3-D) model of an unknown object by fusion of visual and tactile information while the object is grasped. Assuming the object is symmetric, a first hypothesis of its complete 3-D shape is generated. A grasp is executed on the object with a robotic manipulator equipped with tactile sensors. Given the detected contacts between the fingers and the object, the initial full object model including the symmetry parameters can be refined. This refined model will then allow the planning of more complex manipulation tasks. The main contribution of this work is an optimal estimation approach for the fusion of visual and tactile data applying the constraint of object symmetry. The fusion is formulated as a state estimation problem and solved with an iterated extended Kalman filter. The approach is validated experimentally using both artificial and real data from two different robotic platforms.

Fast extraction of multi-resolution Gabor features

Gabor filter responses are general purpose features for computer vision and image processing and have been very successful in many application areas, for example in bio- metric authentication (fingerprint matching, face detection, face recognition and iris recognition). In a typical feature construction, filters are utilised as a multi-resolution structure of several filters tuned to different frequencies and orientations. The multi-resolution structure is similar to wavelets, but the non-orthogonality of Gabor functions implies the main weakness: computational heaviness. The high computational complexity prevents their use in many real-time or near real-time tasks. In this study, an efficient sequential computation method for multi-resolution Gabor features is presented.

Object Localisation Using Generative Probability Model for Spatial Constellation and Local Image Features

In this paper we apply state-of-the-art approach to object detection and localisation by incorporating local descriptors and their spatial configuration into a generative probability model. In contrast to the recent semi- supervised methods we do not utilise interest point detectors, but apply a supervised approach where local image features (landmarks) are annotated in a training set and therefore their appearance and spatial variation can be learnt. Our method enables working in purely probabilistic search spaces providing a MAP estimate of object location, and in contrast to the recent methods, no background class needs to be formed. Using the training set we can estimate pdfs for both spatial constellation and local feature appearance. By applying an inference bias that the largest pdf mode has probability one, we are able to combine prior information (spatial configuration of the features) and observations (image feature appearance) into posterior distribution which can be generatively sampled, e.g. using MCMC techniques. The MCMC methods are sensitive to initialisation, but as a solution, we also propose a very efficient and accurate RANSAC-based method for finding good initial hypotheses of object poses. The complete method can robustly and accurately detect and localise objects under any homography.