There has been an exponential increase in the usage of the internet, particularly among students since the introduction of e-learning and Bring Your Own Device (BYO) initiatives into the education system. In New Zealand the percentage of the population using the internet is now 93.8% and this increase in internet usage has increased the risk of cybersecurity attacks. This makes it necessary to provide awareness and education on cybersecurity to students who are potential targets for exploitation. However, to provide this awareness it is necessary to understand what their current knowledge on cybersecurity is which forms the basis of this paper. This paper presents the results of a survey conducted on internet usage and cybersecurity awareness among three age groups between 8 years and 21 years. A questionnaire consisting of various questions on internet usage and cybersecurity concepts was prepared. For this survey, we considered both computers (desktops & laptops) and mobile devices (tablets & smartphones). The results of the survey showed that cybersecurity awareness among the surveyed students was generally low with the lowest level in the 8-12-year age group. The students of 8-12 age group were able to answer only 19% of survey questions. Furthermore, most of the students were not familiar with common cybersecurity terms and did not demonstrate enough awareness of common threats such as phishing. The results further show that the majority of the students were not aware of cybersecurity tools for tablets and smartphones which are frequently used devices for BYOD. The key contribution of this paper is to emphasise the necessity to create cybersecurity awareness among students.

Deep learning (DL) is becoming an increasingly interesting and powerful machine learning method with successful applications in many domains, such as natural language processing, image recognition, hand-written character recognition, and computer vision. Despite of its eminent success, limitations of traditional learning approach may still prevent deep learning from achieving a wide range of realistic learning tasks. DL approaches has shown success in speech recognition and speaker identification over traditional approaches such as those that use Mel Frequency Cepstrum Coefficients for feature extraction with Gaussian Mixture Models. However, speaker identification research community are not fully aware of the DL process and its application with respect to speaker identification. This paper is motivated to reduce this knowledge gap and to promote the research of implementing deep learning techniques for speaker identification. In this paper, we present a review of the DL methodologies used for speaker identification and surveys important DL algorithms that can potentially be explored for future works. We categorised the applications of DL for speaker identification according to the process of speaker identification and presented a review of these implementations.

In real world situations every model has some weaknesses and will make errors on training data. Given the fact that each model has certain limitations, the aim of ensemble learning is to supervise their strengths and weaknesses, leading to best possible decision in general. Ensemble based machine learning is a solution of minimizing risk in decision making. Bagging, boosting, stacked generalization and mixture of expert methods are the most popular techniques to construct ensemble systems. For the purpose of combining outputs of class labels, weighted majority voting, behaviour knowledge space and border count methods are used to construct independent classifiers and to achieve diversity among the classifiers which is important in ensemble learning. It was found that an ideal ensemble method should work on the principal of achieving six paramount characteristics of ensemble learning; accuracy, scalability, computational cost, usability, compactness and speed of classification. In addition, the ideal ensemble method would be able to handle large huge image size and long term historical data particularly of spatial and temporal. In this paper we reveal that ensemble models have obtained high acceptability in terms of accuracy than single models. Further, we present an analogy of various ensemble techniques, their applicability, measuring the solution diversity, challenges and proposed methods to overcome these challenges without diverting from the original concepts.

The success of Deep Neural Networks (DNNs) for various applications like language processing (NLP), image processing, character recognition inspired to use machine learning (ML) and Evolutionary Computation (EC) techniques for improving learning process. Using evolutionary algorithms to improve the efficiency of deep learning attained some success. However, these techniques are unable to reduce the learning time which is the key concern for deep learning. The main problem with DNN is that, it uses a random topology to start with (similar to artificial neural networks). If the topology is not suitable, training procedure will restart with a new topology and this process continues till expected results are obtained. In this paper, we propose, for the first time, a new prospect for evolving optimized deep neural networks which can provide a warm start to the training process compared to heuristic random initial architecture. We discuss the theoretical approach towards possibility of optimizing the learning process inspired from the existing un-conventional approaches. The training process of DNN with EC approach is faster than regular approach by a considerable difference of over 6 hours for MNIST data set. Further, we observed a considerable improvement in the classification accuracies. Our approaches resulted in an improved classification accuracy of 2% and 4.4%for MNIST data set and 1.2% and 1.4% for IRIS data set compared to heuristic random weights approach. Our initial experimental results prove that evolutionary approaches provides a warm start to the deep learning, thus, reducing the training time.