Sevki S. Erdogan

Measurement criteria for neural network pruning

A new measure based on hidden-output node activation is proposed for measuring the relevance of hidden nodes in a neural network. The concept has been successfully applied for pruning in several classification problems. The experiments indicate that redundant nodes are pruned down resulting in optimal network topologies. The measure has been compared to the one proposed by Kamimura-Nakanishi (see IEICE Trans. Inf. & Syst., vol.E78-D, no.4, p.484-9, 1995) and also used in the context of a modified cost function where an additional penalty function is added to steer the direction of the hidden nodes activation in the process of learning.

Quaternion number based vanilla recognition system for automating polination

Vanilla is the second most expensive spice worldwide. The high cost of vanilla has led to the problem of dangerous adulterated substitutes. Its high cost is attributed largely to the labor intensive hand pollination required where the melipona bee is not present. This article is the first known to present a machine vision system for the recognition of vanilla flowers and their stamen for pollination. We present a hypercomplex numbers based system. The present system is based on a rotational invariant basis more fitted to fast quaternion Fourier transform processing than Zernike Polynomial basis, and computer simulations. It could be used for other applications as well.

Towards net-centric system of systems robotics in air, sea and land

This paper investigates how we can work towards building net-centric swarms of land, sea and air robots working together to accomplish a common goal. The goal is to bring together swarms of robots from all three sectors to safely benefit mankind without increasing dangers in the process. Several platforms are explored for simulation to investigate swarm robotics within heterogeneous environments.

An experimental study of the extended NRBF regression model and its enhancement for classification problem

As an extension of the traditional normalized radial basis function (NRBF) model, the extended normalized RBF (ENRBF) model was proposed by Xu [RBF nets, mixture experts, and Bayesian Ying-Yang learning, Neurocomputing 19 (1998) 223-257]. In this paper, we perform a supplementary study on ENRBF with several properly designed experiments and some further theoretical discussions. It is shown that ENRBF is able to efficiently improve the learning accuracies under some circumstances. Moreover, since the ENRBF model is initially proposed for the regression and function approximation problems, a further step is taken in this work to modify the ENRBF model to deal with the classification problems. Both the original ENRBF model and the new proposed ENRBF classifier (ENRBFC) can be viewed as the special cases of the mixture-of-experts (ME) model that is discussed in Xu et al. [An alternative model for mixtures of experts, in: Advances in Neural Information Processing Systems, MIT Press, Cambridge, MA, 1995]. Experimental results show the potentials of ENRBFC compared to some other related classifiers.

Using FPGA technology towards the design of an adaptive fault tolerant framework

In this paper we propose architecture for a reconfigurable, adaptive, fault-tolerant (RAFT) framework for application in real time systems with require multiple levels of redundancy and protection. Typical application environments include distributed processing, fault-tolerant computation, and mission and safety-critical systems. The framework uses field programmable gate array (FPGA) technologies with on the fly partial programmability achieving reconfiguration of a system component when the existing components fail or to provide extra reliability as required in the specification. The framework proposes the use an array of FPGA devices to implement a system that, after detecting an error caused by a fault, can adoptively reconfigure itself to achieve fault tolerance. The FPGAs that are becoming widely available at a low cost are exploited by defining a system model that allows the system user to define various levels of reliability choices, providing a monitoring layer for the system engineer.

An installable version control file system for Unix

We present a file system named VCFS, a Version Control File System that provides transparent version control of files. VCFS is implemented as an installable file system in the Unix operating system. The facility is transparent to users in the sense that VCFS automatically stores changes between incremental versions of a file without user intervention. Access to the latest version of the file retains traditional Unix semantics, enabling all existing Unix applications to be run modified under VCFS if so desired. VCFS provides system call enhancements and supports application level utilities for version number retrieval, displaying changes between two versions of a file, upgrading to a new release number, and related version control functions. Performance measurements show that VCFS provides a significant increase in version control operations compared to application level change management systems such as SCCS, and that the space utilization is significantly better than of the existing version control systems. The primitives and facilities provided by VCFS can be exploited for use in a wide range of applications, ranging from immutable files, transaction processing and file replication support in a distributed file system, to the development of enhanced change management control systems for software project management.

Involving Undergraduate Students in SoS Engineering

Undergraduate Computer Science Department students at the University of Hawaii at Hilo participated in design, development, and deployment of a low-cost system of systems to collect and report near real-time field data from various sources including sensors linked by satellite or directly connected to the Internet. The project has been implemented as a working prototype for investigating an integrated model for environmental research, which concentrates on the application spatial, and temporal analysis techniques to large-scale ecological data sets and includes satellite imagery and ground-based sensor data. The design is extensible so that it can accommodate data sharing among diverse groups such as marine science, biological, geological, and climate control researchers. A Web interface is provided to visualize data collected from isolated geographical stations using satellite technology. Students experimented with alternative implementations of interface equipment using custom field programmable gate array circuits and stamp processors.

INSIGHT OF FUZZY NEURAL SYSTEMS IN THE APPLICATION OF HANDWRITTEN DIGITS CLASSIFICATION

There have been many applications in the area of handwritten character recognition. Over the last decade much research has gone into developing algorithms to accurately convert captured images of handwriting to text. At the same time, research into neuro fuzzy classification models has proven to solve many complex problems. In this paper, Adaptive Neuro Fuzzy Inference System (ANFIS) and Evolving Fuzzy Neural Network (EFuNN) was investigated and studied in detail on how these two models can be used to perform handwritten digits classification. Results of the experiments show great potential of the EFuNN over the ANFIS for practical implementation of the handwritten digit recognition.

Integrated Knowledge Base for Environmental Research

This paper deals with networking and knowledge base design to support researchers from various fields. The project is been implemented as a working prototype for investigating an integrated model for environmental research, which concentrates on the application spatial, and temporal analysis techniques to large-scale ecological data sets and includes satellite imagery and ground-based sensor data. The primary objective of the paper is to present the current state of the art for capturing and displaying live location information of a moving target in a wildlife environment as well as environmental parameters from static monitoring stations

VHDL modeling and simulation of the back-propagation algorithm and its mapping to the RM

The Reconfigurable Machine (RM) is a parallel architecture that is built using Xilinxs 4005 field programmable gate array (FPGA) and associated fast static random-access memories (SRAMs). VHDL modeling and simulation of a fully connected three-layer back-propagation (BP) are attempted in the present work with a view towards its mapping to a reconfigurable parallel architecture. The mapping encompasses both the forward and backward passes. A bottom-up approach is used which starts from the configuration of processing elements to achieve effective computation of floating-point sum-of-products. The FPGAs perform the floating-point multiplication, addition, and function evaluation, while the local SRAMs are used for storing input/output (I/O) data for the RM.