Jethro Shell

Fuzzy data fusion for fault detection in Wireless Sensor Networks

Wireless Sensor Networks (WSN) can produce decisions that are unreliable due to the large inherent uncertainties in the areas which they are deployed. It is vital for the applications where WSNs are deployed that accurate decisions can be made from the data produced. Fault detection is a vital pursuit, however it is a challenging task. In this paper we present a fuzzy logic data fusion approach to fault detection within a Wireless Sensor Network using a Statistical Process Control and a clustered covariance method. Through the use of a fuzzy logic data fusion approach we have introduced a novel technique into this area to reduce uncertainty and false-positives within the fault detection process.

Fuzzy Transfer Learning: Methodology and application

Abstract Producing a methodology that is able to predict output using a model is a well studied area in Computational Intelligence (CI). However, a number of real-world applications require a model but have little or no data available of the specific environment. Predominantly, standard machine learning approaches focus on a need for training data for such models to come from the same domain as the target task. Such restrictions can severely reduce the data acquisition making it extremely costly, or in certain situations, impossible. This impedes the ability of these approaches to model such environments. It is on this particular problem that this paper is focussed. In this paper two concepts, Transfer Learning (TL) and Fuzzy Logic (FL) are combined in a framework, Fuzzy Transfer Learning (FuzzyTL), to address the problem of learning tasks that have no prior direct contextual knowledge. Through the use of a FL based learning method, uncertainty that is evident in dynamic environments is represented. By applying a TL approach through the combining of labelled data from a contextually related source task, and little or no unlabelled data from a target task, the framework is shown to be able to accomplish predictive tasks using models learned from contextually different data.

A fast and efficient semantic short text similarity metric

The semantic comparison of short sections of text is an emerging aspect of Natural Language Processing (NLP). In this paper we present a novel Short Text Semantic Similarity (STSS) method, Lightweight Semantic Similarity (LSS), to address the issues that arise with sparse text representation. The proposed approach captures the semantic information contained when comparing text to process the similarity. The methodology combines semantic term similarities with a vector similarity method used within statistical analysis. A modification of the term vectors using synset similarity values addresses issues that are encountered with sparse text. LSS is shown to be comparable to current semantic similarity approaches, LSA and STASIS, whilst having a lower computational footprint.

Towards fuzzy transfer learning for intelligent environments

By their very nature, Intelligent Environments (IE’s) are infused with complexity, unreliability and uncertainty due to a combination of sensor noise and the human element. The quantity, type and availability of data to model these applications can be a major issue. Each situation is contextually different and constantly changing. The dynamic nature of the implementations present a challenging problem when attempting to model or learn a model of the environment. Training data to construct the model must be within the same feature space and have the same distribution as the target task data, however this is often highly costly and time consuming. There can even be occurrences were a complete lack of labelled target data occurs. It is within these situations that our study is focussed. In this paper we propose a framework to dynamically model IE’s through the use of data sets from differing feature spaces and domains. The framework is constructed using a novel Fuzzy Transfer Learning (FuzzyTL) process.

Application of Artificial Neural Network and Support Vector Regression in cognitive radio networks for RF power prediction using compact differential evolution algorithm

Cognitive radio (CR) technology has emerged as a promising solution to many wireless communication problems including spectrum scarcity and underutilization. To enhance the selection of channel with less noise among the white spaces (idle channels), the a priory knowledge of Radio Frequency (RF) power is very important. Computational Intelligence (CI) techniques cans be applied to these scenarios to predict the required RF power in the available channels to achieve optimum Quality of Service (QoS). In this paper, we developed a time domain based optimized Artificial Neural Network (ANN) and Support Vector Regression (SVR) models for the prediction of real world RF power within the GSM 900, Very High Frequency (VHF) and Ultra High Frequency (UHF) FM and TV bands. Sensitivity analysis was used to reduce the input vector of the prediction models. The inputs of the ANN and SVR consist of only time domain data and past RF power without using any RF power related parameters, thus forming a nonlinear time series prediction model. The application of the models produced was found to increase the robustness of CR applications, specifically where the CR had no prior knowledge of the RF power related parameters such as signal to noise ratio, bandwidth and bit error rate. Since CR are embedded communication devices with memory constrain limitation, the models used, implemented a novel and innovative initial weight optimization of the ANNs through the use of compact differential evolutionary (cDE) algorithm variants which are memory efficient. This was found to enhance the accuracy and generalization of the ANN model.

Optimized artificial neural network using differential evolution for prediction of RF power in VHF/UHF TV and GSM 900 bands for cognitive radio networks

Cognitive radio (CR) technology has emerged as a promising solution to many wireless communication problems including spectrum scarcity and underutilization. The knowledge of Radio Frequency (RF) power (primary signals and/or interfering signals plus noise) in the channels to be exploited by CR is of paramount importance, not just the existence or absence of primary users. If a channel is known to be noisy, even in the absence of primary users, using such channels will demand large quantities of radio resources (transmission power, bandwidth, etc) in order to deliver an acceptable quality of service to users. Computational Intelligence (CI) techniques can be applied to these scenarios to predict the required RF power in the available channels to achieve optimum Quality of Service (QoS). While most of the prediction schemes are based on the determination of spectrum holes, those designed for power prediction use known radio parameters such as signal to noise ratio (SNR), bandwidth, and bit error rate. Some of these parameters may not be available or known to cognitive users. In this paper, we developed a time domain based optimized Artificial Neural Network (ANN) model for the prediction of real world RF power within the GSM 900, Very High Frequency (VHF) and Ultra High Frequency (UHF) TV bands. The application of the models produced was found to increase the robustness of CR applications, specifically where the CR had no prior knowledge of the RF power related parameters. The models used implemented a novel and innovative initial weight optimization of the ANNs through the use of differential evolutionary algorithms. This was found to enhance the accuracy and generalization of the approach.

Optimized Neural Network using differential evolutionary and swarm intelligence optimization algorithms for RF power prediction in cognitive radio network: A comparative study

Cognitive radio (CR) technology has emerged as a promising solution to many wireless communication problems including spectrum scarcity and underutilization. The a priory knowledge of Radio Frequency (RF) power (primary signals and/ or interfering signals plus noise) in the channels to be exploited by CR is of paramount importance. This will enable the selection of channel with less noise among idle (free) channels. Computational Intelligence (CI) techniques can be applied to these scenarios to predict the required RF power in the available channels to achieve optimum Quality of Service (QoS). In this paper, we developed a time domain based optimized Artificial Neural Network (ANN) model for the prediction of real world RF power within the GSM 900, Very High Frequency (VHF) and Ultra High Frequency (UHF) TV bands. The application of the models produced was found to increase the robustness of CR applications, specifically where the CR had no prior knowledge of the RF power related parameters such as signal to noise ratio, bandwidth and bit error rate. The models used, implemented a novel and innovative initial weight optimization of the ANNs through the use of differential evolutionary and swarm intelligence algorithms. This was found to enhance the accuracy and generalization of the ANN model. For this problem, DE/best/1/bin was found to yield a better performance as compared with the other algorithms implemented.

Towards dynamic accessibility through soft gaze gesture recognition

It is difficult for some sets of users with physical disabilities to operate standard input devices such as a keyboard and mouse. Eye gaze technologies and more specifically gaze gestures are emerging to assist such users. There is a high level of inter and intra user variation in the ability to perform gaze gestures due to the high levels of noise with the gaze patterns. In this paper we use a novel fuzzy transfer learning approach in order to construct a fuzzy system for gaze gesture recognition which can be automatically adapted for different users and/or user groups. We show that the fuzzy system is able to recognise gestures across groups of both able bodied (AB) and disabled users through the use of a base of AB data surpassing an expert constructed classifier.

LOGAN's Run: Lane optimisation using genetic algorithms based on NSGA-II

Whilst bus lanes are an important tool to ensure bus time reliability their presence can be detrimental to urban traffic. In this paper a Non-dominated Sorting Genetic Algorithm (NSGA-II) has been adopted to study the effect of bus lanes on urban traffic in terms of location and time of operation. Due to the complex nature of this problem traditional search would not be feasible. An artificial arterial route has been modelled from real data to evaluate candidate solutions. The results confirm this methodology for the purpose of studying and identifying bus lane locations and times of operation. Additionally it is shown that bus lanes can exist on an arterial link without exclusively occupying a continuous lane for large periods of time. Furthermore results indicate a use for this methodology over a larger scale and potential near real-time operation.