Marek Bundzel

Combining Gradient and Evolutionary Approaches to the Artificial Neural Networks Training According to Principles of Support Vector Machines

A gradient based learning for ANN training in pattern recognition tasks and a genetic approach for ANN pruning are proposed in this paper. The goal is to achieve a wide margin classifier the Vapnik-Chevornenkis (VC) dimension of which is being reduced in order to increase the generalization performance. Inspired by Support Vector Machines the examples closest to the decision boundary contribute to the training the most. The training penalty is rule-based and calculated according to the spatial distribution of the training examples relative to the separating hyperplane. The tendency to saturation of hidden neurons is suppressed. Genetic algorithm based method is proposed for reduction of the size of a trained ANN. The proposed algorithms were tested on artificial and real world data and compared to standard Backpropagation and Support Vector Machine with Gaussian RBF kernel.

Object Identification in Dynamic Images Based on the Memory-Prediction Theory of Brain Function

In 2004, Jeff Hawkins presented a memory-prediction theory of brain function, and later used it to create the Hierar-chical Temporal Memory model. Several of the concepts described in the theory are applied here in a computer vision system for a mobile robot application. The aim was to produce a system enabling a mobile robot to explore its envi-ronment and recognize different types of objects without human supervision. The operator has means to assign names to the identified objects of interest. The system presented here works with time ordered sequences of images. It utilizes a tree structure of connected computational nodes similar to Hierarchical Temporal Memory and memorizes frequent sequences of events. The structure of the proposed system and the algorithms involved are explained. A brief survey of the existing algorithms applicable in the system is provided and future applications are outlined. Problems that can arise when the robot’s velocity changes are listed, and a solution is proposed. The proposed system was tested on a sequence of images recorded by two parallel cameras moving in a real world environment. Results for mono- and ste-reo vision experiments are presented.

Forward control of robotic arm using the information from stereo-vision tracking system

In this paper we present the feed-forward neural network controller of robotic arm, which makes use of tracking method applied to stereo-vision cameras mounted on the head of the humanoid robot Nao, in order to touch the tracked object. The Tracking-Learning-Detection (TLD) method, which we use to detect and track the object, is known for its state-of-art performance and high robustness. This method was adjusted to be usable with a stereo-vision camera system, in order to provide 3D spatial coordinates of the object. These coordinates are used as the input for the feed-forward controller, which controls the arm of a humanoid robot. The goal of the controller is to move the hand of the robot to the object by setting arm joints into position corresponding to the object location. The controller is implemented as an artificial neural network and trained using the error back-propagation algorithm. The experiment, which demonstrates the proof of the concept, is also denoted in this paper.

Using string invariants for prediction searching for optimal parameters

We have developed a novel prediction method based on string invariants. The method does not require learning but a small set of parameters must be set to achieve optimal performance. We have implemented an evolutionary algorithm for the parametric optimization. We have tested the performance of the method on artificial and real world data and compared the performance to statistical methods and to a number of artificial intelligence methods. We have used data and the results of a prediction competition as a benchmark. The results show that the method performs well in single step prediction but the method’s performance for multiple step prediction needs to be improved. The method works well for a wide range of parameters.

Industry IoT Gateway for Cloud Connectivity

New approaches and technologies like Internet of Things (IoT), cloud computing and Big Data are giving rise to another industrial revolution. We propose here an implementation of an industrial gateway architecture adopting the idea of IoT, intelligent methods, Machine-to-Machine and Cyber-Physical Systems. The proposed gateway creates a virtual representation of the physical world scanning the technological layer’s devices in real time. It creates a uniform communication interface for the heterogeneous technological layer, enables self-management of devices, diagnostics and self-reconfiguration to improve Quality of Service aided with cloud SCADA and MES services. We have tested the proposed gateway in an experimental setup with a programmable logic controller.

Application of Tracking-Learning-Detection for Object Tracking in Stereoscopic Images

We use Tracking-Learning-Detection algorithm (TLD) [1]-[3] to localize and track objects in images sensed simultaneously by two parallel cameras in order to determine 3D coordinates of the tracked object. TLD method was chosen for its state-of-art performance and high robustness. TLD stores the object to be tracked as a set of 2D grayscale images that is incrementally built. We have implemented the 3D tracking system into a PC, communicating with the Nao humanoid robot [4][5] equipped with a stereo camera head. Experiments evaluating the accuracy of the 3D tracking system are presented. The robot uses feed-forward control to touch the tracked object. The controller is an artificial neural network trained using the error Back-Propagation algorithm. Experiments evaluating the success rate of the robot touching the object are presented.

Building support vector machine alternative using algorithms of computational geometry

The task of pattern recognition is a task of division of a feature space into regions separating the training examples belonging to different classes. Support Vector Machines (SVM) identify the most borderline examples called support vectors and use them to determine discrimination hyperplanes (hyper–curves). In this paper a pattern recognition method is proposed which represents an alternative to SVM algorithm. Support vectors are identified using selected methods of computational geometry in the original space of features i.e. not in the transformed space determined partially by the kernel function of SVM. The proposed algorithm enables usage of kernel functions. The separation task is reduced to a search for an optimal separating hyperplane or a Winner Takes All (WTA) principle is applied.

Optimization of tangential pivot tonearm using genetic algorithm

Turntables continue to be manufactured and sold today, although in small numbers. Some audiophiles still prefer the sound of vinyl records over that of digital music sources and are ready to pay for high end gramophone systems. We pre-sent here a tonearm constructed so that its rotating cartridge is always nearly tan-gent to the records groove the stylus is reading thus greatly reducing the anti-skating force. The physical dimensions of the tonearm are optimized using genetic algorithm to minimize the deviation of the cartridge from the tangent along its entire trajectory (tangent tracking). We describe the specific optimization algorithm used, report the simulated results and present a working prototype.

Pseudo-distance based artificial neural network training

The goal of this work is to create a model based on an artificial neural network for prediction in a real world domain. Information on prediction errors price function (reflecting the monetary value of a loss) is acquired from a power and heat production expert. Systematic errors and not acceptable types of errors in specific situations are identified and a solution using superposition of the price function on Euclidian distance is proposed. This approach can be considered as a special case of pseudo distance theory implementation. There is a functional prediction system in the observed power and heat producing company which makes the information retrieval in cooperation with the domain expert less difficult. The knowledge on the price and on the not acceptable errors is used for modification of the learning error function. The experimental results of the proposed algorithm are obtained on the heat production data. Our model has approximately the same percentage accuracy as the existing model but using the expensive errors suppression, predictions of the new model lead to production costs reduction.