Search

As robots begin to coexist with humans, the need for efficient and safe social robot navigation becomes increasingly pressing. In this paper we investigate how world models can enhance the effectiveness of reinforcement learning in social navigation tasks. We introduce three approaches that leverage predictive world models, which are then benchmarked against state-of-the-art algorithms. For a comprehensive and reliable evaluation, we employed multiple metrics during the training and testing phases. The key novelty of our approach consists in the integration and evaluation of predictive world models within the context of social navigation, as well as in the models themselves. Based on a diverse set of performance metrics, the experimental results provide evidence that predictive world models help improve reinforcement learning techniques for social navigation.

A chatbot is one of the most innovative tools for interaction between humans and machines. It simulates human behaviour and interacts with a user in natural language to provide required information, services, thus enhancing the user experience. An increasing number of organisations are utilising chatbots to offer better services to users in real time to improve user satisfaction, streamline the user journey, and provide user-centric support. However, this requires building specific chatbots for specific services, which leads to variational designs of chatbots based on their application areas.This paper aims to demystify fundamentals of a chatbot including its working, types, benefits and limitations. It will mainly focus on two main categories of chatbots rule-based and AI-based chatbots, and explain all the aforementioned facts as well their comparative analysis to simplify the chatbot topic in general.

Machine learning has been constantly evolving and revolutionizing every aspect of our lives. There is ongoing research to enhance and modify machine learning models where scientists and researchers are finding ways to improve the effectiveness and adaptability of models with the changing technology moulding to user requirements for real life application. The main challenges in this endeavour of enhancing machine learning models are obtaining quality data, selecting an appropriate model, and ensuring the data privacy. Federated learning has been developed to address the aforementioned challenges, which is an effective way to train machine learning models in a collaborative manner by using the local data from a large number of devices without directly exchanging their raw data whilst simultaneously delivering on model performance. Federated learning is not just a type of machine learning, it is an amalgamation of several technologies and techniques. To fully understand its concepts a comprehensive study is required. This paper aims to simplify the fundamentals of federated learning in order to provide a better understanding of it. It explains federated learning in a step-by-step manner covering its comprehensive definition, detailed working, different types, benefits and limitations.

The landscape of machine learning is changing rapidly due to the ever-evolving nature of data and devices. The large centralized data is replaced by the distributed data and a central server is replaced with a large number of geographically distributed, loosely connected devices, such as smartphones, laptops, and other IoT devices. Therefore, the centralized machine learning (CML) which involves centralized data training on a central server is no longer an effective solution when the data is inherently distributed or too big to process on a central server, or data privacy is paramount; and the quest for a suitable machine learning to resolve these issues led to the evolution of distributed machine learning (DML). For large-scale learning tasks, DML has evolved to effectively handle enormous data within big data and distributed computing environment. Resolving most limitations faced by CML with the implementation of parallel learning on a large number of nodes to optimise time, learning resources and performance. However, DML may not necessarily ensure strict data privacy leading to further development and innovation of federated machine learning (FML) which is a type of DML that further decentralizes learning operations using local data on each participating node incorporating data privacy adherence. This paper analyses the transformation journey of machine learning whilst explaining its evolution from centralized, distributed to federated machine learning. Examining these three variants of machine learning exemplifies their coherent and comparative analysis. Which helps grasp a better understanding of each machine learning type as well as presenting the reason for the changing landscape. Additionally, the paper will address each type of machine learning alongside their different types, strengths and limitations.

The production of chemicals is a complex task due to the highly nonlinear behaviour of chemical processes; therefore, traditional approaches may not be very effective in developing or predicting such processes and their outcomes at optimal level. Consequently, it has always been challenging to find ways to improve efficiency and productivity while reducing the time and cost. Artificial Intelligence (AI) techniques are becoming valuable in chemistry due to several reasons such as easy to learn and use, simple implementation, easy designing, effectiveness, generality, robustness, and flexibility. AI is comprised of several techniques within it, such as artificial neural networks, evolutionary algorithms and fuzzy logic. AI techniques have been widely used in various areas of chemistry including molecule design, molecular property prediction, retrosynthesis, reaction outcome prediction and reaction conditions prediction. Therefore, this paper investigates AI applications in the aforementioned areas, wherein it explains about each aforementioned area with a suitable example, limitations of traditional techniques, and types of AI techniques which are utilised within those areas.

Localization plays a crucial role in enhancing the practicality and precision of VQA systems. By enabling fine-grained identification and interaction with specific parts of an object, it significantly improves the system's ability to provide contextually relevant and spatially accurate responses, crucial for applications in dynamic environments like robotics and augmented reality. However, traditional systems face challenges in accurately mapping objects within images to generate nuanced and spatially aware responses. In this work, we introduce "Detect2Interact", which addresses these challenges by introducing an advanced approach for fine-grained object visual key field detection. First, we use the segment anything model (SAM) to generate detailed spatial maps of objects in images. Next, we use Vision Studio to extract semantic object descriptions. Third, we employ GPT-4's common sense knowledge, bridging the gap between an object's semantics and its spatial map. As a result, Detect2Interact achieves consistent qualitative results on object key field detection across extensive test cases and outperforms the existing VQA system with object detection by providing a more reasonable and finer visual representation., Comment: Accepted to IEEE Intelligent Systems

In this paper, we consider observations from a series of smart meters that are either completely or partially aggregated, and our aim is to estimate the metering hierarchy. We propose to estimate this important metadata through a novel adaptation of the Chow–Liu tree learning procedure. Our approach takes into account prior knowledge from a set of dominance conditions that are easily elicited from the consumption data. In addition to more traditional correlation-based approaches we also introduce a distance-correlation-based method for detecting edges. Synthetic experiments show the benefits of distance correlation and the dominance conditions in recovering tree structure. The paper concludes with a real-world application of the method to infer energy metering hierarchies in a library building.

Human activity recognition based on generated sensor data plays a major role in a large number of applications such as healthcare monitoring and surveillance system. Yet, accurately recognizing human activities is still challenging and active research due to people’s tendency to perform daily activities in a different and multitasking way. Existing approaches based on the recurrent setting for human activity recognition have some issues, such as the inability to process data parallelly, the requirement for more memory and high computational cost albeit they achieved reasonable results. Convolutional Neural Network processes data parallelly, but, it breaks the ordering of input data, which is significant to build an effective model for human activity recognition. To overcome these challenges, this study proposes causal convolution based on performers-attention and supervised contrastive learning to entirely forego recurrent architectures, efficiently maintain the ordering of human daily activities and focus more on important timesteps of the sensors’ data. Supervised contrastive learning is integrated to learn a discriminative representation of human activities and enhance predictive performance. The proposed network is extensively evaluated for human activities using multiple datasets including wearable sensor data and smart home environments data. The experiments on three wearable sensor datasets and five smart home public datasets of human activities reveal that our proposed network achieves better results and reduces the training time compared with the existing state-of-the-art methods and basic temporal models.

Electronic Government (e-Government) systems constantly provide greater services to people, businesses, organisations, and societies by offering more information, opportunities, and platforms with the support of advances in information and communications technologies. This usually results in increased system complexity and sensitivity, necessitating stricter security and privacy-protection measures. The majority of the existing e-Government systems are centralised, making them vulnerable to privacy and security threats, in addition to suffering from a single point of failure. This study proposes a decentralised e-Government framework with integrated threat detection features to address the aforementioned challenges. In particular, the privacy and security of the proposed e-Government system are realised by the encryption, validation, and immutable mechanisms provided by Blockchain. The insider and external threats associated with blockchain transactions are minimised by the employment of an artificial immune system, which effectively protects the integrity of the Blockchain. The proposed e-Government system was validated and evaluated by using the framework of Ethereum Visualisations of Interactive, Blockchain, Extended Simulations (i.e. eVIBES simulator) with two publicly available datasets. The experimental results show the efficacy of the proposed framework in that it can mitigate insider and external threats in e-Government systems whilst simultaneously preserving the privacy of information.

As more business transactions and information services have been implemented via communication networks, both personal and organization assets encounter a higher risk of attacks. To safeguard these, a perimeter defence like NIDS (network-based intrusion detection system) can be effective for known intrusions. There has been a great deal of attention within the joint community of security and data science to improve machine-learning based NIDS such that it becomes more accurate for adversarial attacks, where obfuscation techniques are applied to disguise patterns of intrusive traffics. The current research focuses on non-payload connections at the TCP (transmission control protocol) stack level that is applicable to different network applications. In contrary to the wrapper method introduced with the benchmark dataset, three new filter models are proposed to transform the feature space without knowledge of class labels. These ECT (ensemble clustering based transformation) techniques, i.e., ECT-Subspace, ECT-Noise and ECT-Combined, are developed using the concept of ensemble clustering and three different ensemble generation strategies, i.e., random feature subspace, feature noise injection and their combinations. Based on the empirical study with published dataset and four classification algorithms, new models usually outperform that original wrapper and other filter alternatives found in the literature. This is similarly summarized from the first experiment with basic classification of legitimate and direct attacks, and the second that focuses on recognizing obfuscated intrusions. In addition, analysis of algorithmic parameters, i.e., ensemble size and level of noise, is provided as a guideline for a practical use.

Human activity recognition (HAR) is an essential research field that has been used in different applications including home and workplace automation, security and surveillance as well as healthcare. Starting from conventional machine learning methods to the recently developing deep learning techniques and the Internet of things, significant contributions have been shown in the HAR area in the last decade. Even though several review and survey studies have been published, there is a lack of sensor-based HAR overview studies focusing on summarising the usage of wearable sensors and smart home sensors data as well as applications of HAR and deep learning techniques. Hence, we overview sensor-based HAR, discuss several important applications that rely on HAR, and highlight the most common machine learning methods that have been used for HAR. Finally, several challenges of HAR are explored that should be addressed to further improve the robustness of HAR.

Human activity recognition based on generated sensor data plays a major role in a large number of applications such as healthcare monitoring and surveillance system. Yet, accurately recognizing human activities is still challenging and active research due to people’s tendency to perform daily activities in a different and multitasking way. Existing approaches based on the recurrent setting for human activity recognition have some issues, such as the inability to process data parallelly, the requirement for more memory and high computational cost albeit they achieved reasonable results. Convolutional Neural Network processes data parallelly, but, it breaks the ordering of input data, which is significant to build an effective model for human activity recognition. To overcome these challenges, this study proposes causal convolution based on performers-attention and supervised contrastive learning to entirely forego recurrent architectures, efficiently maintain the ordering of human daily activities and focus more on important timesteps of the sensors’ data. Supervised contrastive learning is integrated to learn a discriminative representation of human activities and enhance predictive performance. The proposed network is extensively evaluated for human activities using multiple datasets including wearable sensor data and smart home environments data. The experiments on three wearable sensor datasets and five smart home public datasets of human activities reveal that our proposed network achieves better results and reduces the training time compared with the existing state-of-the-art methods and basic temporal models. © 2022, The Author(s).

Existing models based on sensor data for human activity recognition are reporting state-of-the-art performances. Most of these models are conducted based on single-domain learning in which for each domain a model is required to be trained. However, the generation of adequate labelled data and a learning model for each domain separately is often time-consuming and computationally expensive. Moreover, the deployment of multiple domain-wise models is not scalable as it obscures domain distinctions, introduces extra computational costs, and limits the usefulness of training data. To mitigate this, we propose a multi-domain learning network to transfer knowledge across different but related domains and alleviate isolated learning paradigms using a shared representation. The proposed network consists of two identical causal convolutional sub-networks that are projected to a shared representation followed by a linear attention mechanism. The proposed network can be trained using the full training dataset of the source domain and a dataset of restricted size of the target training domain to reduce the need of large labelled training datasets. The network processes the source and target domains jointly to learn powerful and mutually complementary features to boost the performance in both domains. The proposed multi-domain learning network on six real-world sensor activity datasets outperforms the existing methods by applying only 50% of the labelled data. This confirms the efficacy of the proposed approach as a generic model to learn human activities from different but related domains in a joint effort, to reduce the number of required models and thus improve system efficiency.

Most defence mechanisms such as a network-based intrusion detection system (NIDS) are often sub-optimal for the detection of an unseen malicious pattern. In response, a number of studies attempt to empower a machine-learning-based NIDS to improve the ability to recognize adversarial attacks. Along this line of research, the present work focuses on non-payload connections at the TCP stack level, which is generalized and applicable to different network applications. As a compliment to the recently published investigation that searches for the most informative feature space for classifying obfuscated connections, the problem of class imbalance is examined herein. In particular, a multiple-clustering-based undersampling framework is proposed to determine the set of cluster centroids that best represent the majority class, whose size is reduced to be on par with that of the minority. Initially, a pool of centroids is created using the concept of ensemble clustering that aims to obtain a collection of accurate and diverse clusterings. From that, the final set of representatives is selected from this pool. Three different objective functions are formed for this optimization driven process, thus leading to three variants of FF-Majority, FF-Minority and FF-Overall. Based on the thorough evaluation of a published dataset, four classification models and different settings, these new methods often exhibit better predictive performance than its baseline, the single-clustering undersampling counterpart and state-of-the-art techniques. Parameter analysis and implication for analyzing an extreme case are also provided as a guideline for future applications.

The advent of new malware types and their attack vectors poses serious challenges for security experts in discovering effective malware detection and analysis techniques. The preliminary step in malware analysis is filtering out samples of counterfeit malware from the suspicious samples by classifying them into most likely and unlikely malware categories. This will enable effective utilisation of resources and expertise for the most likely category of samples in subsequent stages and avoid nugatory effort. This process requires a very fast and resource-optimised method as it is applied on a large sample size. Fuzzy hashing and import hashing methods satisfy these requirements of malware analysis, though, with some limitations. Therefore, the proper integration of these methods, may overcome some of the limitations and improve the detection accuracy without affecting the overall performance of analysis. Hence, this paper proposes a fuzzy-import hashing technique, which is the integration of two methods, namely, fuzzy hashing and import hashing. This integration can offer several benefits such as an improved detection rate by complementing each other when one method cannot detect malware, then the other method can; and the generation of fuzzfied results for subsequent clustering or classification, as the import hashing result can be easily merged with the fuzzy hashing result. The success of this proposed fuzzy-import hashing method is demonstrated through several experiments namely: on the collected malware and goodware corpus; a comparative evaluation against the established YARA rules and application in fuzzy c-means clustering.

A honeypot is a concealed security system that functions as a decoy to entice cyberattackers to reveal their information. Therefore, it is essential to disguise its identity to ensure its successful operation. Nonetheless, cyberattackers frequently attempt to uncover these honeypots; one of the most effective techniques for revealing their identity is a fingerprinting attack. Once identified, a honeypot can be exploited as a zombie by an attacker to attack others. Several effective techniques are available to prevent a fingerprinting attack, however, that would be contrary to the purpose of a honeypot, which is designed to interact with attackers to attempt to discover information relating to them. A technique to discover any attempted fingerprinting attack is highly desirable, for honeypots, while interacting with cyberattackers. Unfortunately, no specific method is available to detect and predict an attempted fingerprinting attack in real-time due to the difficulty of isolating it from other attacks. This paper presents a computational intelligence enabled honeypot that is capable of discovering and predicting an attempted fingerprinting attack by using a Principal components analysis and Fuzzy inference system. This proposed system is successfully tested against the five popular fingerprinting tools Nmap, Xprobe2, NetScanTools Pro, SinFP3 and Nessus.

Cybercriminals are becoming more sophisticated wearing a mask of anonymity and unleashing more destructive malware on a daily basis. The biggest challenge is coping with the abundance of malware created and filtering targeted samples of destructive malware for further investigation and analysis whilst discarding any inert samples, thus optimising the analysis by saving time, effort and resources. The most common technique is malware triaging to separate likely malware and unlikely malware samples. One such triaging technique is YARA rules, commonly used to detect and classify malware based on string and pattern matching, rules are triggered and alerted when their condition is satisfied. This pattern matching technique used by YARA rules and its detection rate can be improved in several ways, however, it can lead to bulky and complex rules that affect the performance of YARA rules. This paper proposes a fuzzy hashing aided enhanced YARA rules to improve the detection rate of YARA rules without significantly increasing the complexity and overheads inherent in the process. This proposed approach only uses an additional fuzzy hashing alongside basic YARA rules to complement each other, so that when one method cannot detect a match, then the other technique can. This work employs three triaging methods fuzzy hashing, import hashing and YARA rules to perform extensive experiments on the collected malware samples. The detection rate of enhanced YARA rules is compared against the detection rate of the employed triaging methods to demonstrate the improvement in the overall triaging results.

The YARA rules technique is used in cybersecurity to scan for malware, often in its default form, where rules are created either manually or automatically. Creating YARA rules that enable analysts to label files as suspected malware is a highly technical skill, requiring expertise in cybersecurity. Therefore, in cases where rules are either created manually or automatically, it is desirable to improve both the performance and detection outcomes of the process. In this paper, two methods are proposed utilising the techniques of fuzzy hashing and fuzzy rules, to increase the effectiveness of YARA rules without escalating the complexity and overheads associated with YARA rules. The first proposed method utilises fuzzy hashing referred to as enhanced YARA rules in this paper, where if existing YARA rules fails to detect the inspected file as malware, then it is subjected to fuzzy hashing to assess whether this technique would identify it as malware. The second proposed technique called embedded YARA rules utilises fuzzy hashing and fuzzy rules to improve the outcomes further. Fuzzy rules countenance circumstances where data are imprecise or uncertain, generating a probabilistic outcome indicating the likelihood of whether a file is malware or not. The paper discusses the success of the proposed enhanced YARA rules and embedded YARA rules through several experiments on the collected malware and goodware corpus and their comparative evaluation against YARA rules.

Systems of sensor human activity recognition are becoming increasingly popular in diverse fields such as healthcare and security. Yet, developing such systems poses inherent challenges due to the variations and complexity of human behaviors during the performance of physical activities. Recurrent neural networks, particularly long short-term memory have achieved promising results on numerous sequential learning problems, including sensor human activity recognition. However, parallelization is inhibited in recurrent networks due to sequential operation and computation that lead to slow training, occupying more memory and hard convergence. One-dimensional convolutional neural network processes input temporal sequential batches independently that lead to effectively executed operations in parallel. Despite that, a one-dimensional Convolutional Neural Network is not sensitive to the order of the time steps which is crucial for accurate and robust systems of sensor human activity recognition. To address this problem, we propose a network architecture based on dilated causal convolution and multi-head self-attention mechanisms that entirely dispense recurrent architectures to make efficient computation and maintain the ordering of the time steps. The proposed method is evaluated for human activities using smart home binary sensors data and wearable sensor data. Results of conducted extensive experiments on eight public and benchmark HAR data sets show that the proposed network outperforms the state-of-the-art models based on recurrent settings and temporal models.

On error of value function inevitably causes an overestimation phenomenon and has a negative impact on the convergence of the algorithms. To mitigate the negative effects of the approximation error, we propose Error Controlled Actor-critic which ensures confining the approximation error in value function. We present an analysis of how the approximation error can hinder the optimization process of actor-critic methods.Then, we derive an upper boundary of the approximation error of Q function approximator and find that the error can be lowered by restricting on the KL-divergence between every two consecutive policies when training the policy. The results of experiments on a range of continuous control tasks demonstrate that the proposed actor-critic algorithm apparently reduces the approximation error and significantly outperforms other model-free RL algorithms.

Malware has remained a consistent threat since its emergence, growing into a plethora of types and in large numbers. In recent years, numerous new malware variants have enabled the identification of new attack surfaces and vectors, and have become a major challenge to security experts, driving the enhancement and development of new malware analysis techniques to contain the contagion. One of the preliminary steps of malware analysis is to remove the abundance of counterfeit malware samples from the large collection of suspicious samples. This process assists in the management of man and machine resources effectively in the analysis of both unknown and likely malware samples. Hashing techniques are one of the fastest and efficient techniques for performing this preliminary analysis such as fuzzy hashing and import hashing. However, both hashing methods have their limitations and they may not be effective on their own, instead the combination of two distinctive methods may assist in improving the detection accuracy and overall performance of the analysis. This paper proposes a Fuzzy-Import hashing technique which is the combination of fuzzy hashing and import hashing to improve the detection accuracy and overall performance of malware analysis. This proposed Fuzzy-Import hashing offers several benefits which are demonstrated through the experimentation performed on the collected malware samples and compared against stand-alone techniques of fuzzy hashing and import hashing.

YARA rules utilises string or pattern matching to perform malware analysis and is one of the most effective methods in use today. However, its effectiveness is dependent on the quality and quantity of YARA rules employed in the analysis. This can be managed through the rule optimisation process, although, this may not necessarily guarantee effective utilisation of YARA rules and its generated findings during its execution phase, as the main focus of YARA rules is in determining whether to trigger a rule or not, for a suspect sample after examining its rule condition. YARA rule conditions are Boolean expressions, mostly focused on the binary outcome of the malware analysis, which may limit the optimised use of YARA rules and its findings despite generating significant information during the execution phase. Therefore, this paper proposes embedding fuzzy rules with YARA rules to optimise its performance during the execution phase. Fuzzy rules can manage imprecise and incomplete data and encompass a broad range of conditions, which may not be possible in Boolean logic. This embedding may be more advantageous when the YARA rules become more complex, resulting in multiple complex conditions, which may not be processed efficiently utilising Boolean expressions alone, thus compromising effective decision-making. This proposed embedded approach is applied on a collected malware corpus and is tested against the standard and enhanced YARA rules to demonstrate its success.

The YARA rules technique is used in cybersecurity to scan for malware, often in its default form, where rules are created either manually or automatically. Creating YARA rules that enable analysts to label files as suspected malware is a highly technical skill, requiring expertise in cybersecurity. Therefore, in cases where rules are either created manually or automatically, it is desirable to improve both the performance and detection outcomes of the process. In this paper, two methods are proposed utilising the techniques of fuzzy hashing and fuzzy rules, to increase the effectiveness of YARA rules without escalating the complexity and overheads associated with YARA rules. The first proposed method utilises fuzzy hashing referred to as enhanced YARA rules in this paper, where if existing YARA rules fails to detect the inspected file as malware, then it is subjected to fuzzy hashing to assess whether this technique would identify it as malware. The second proposed technique called embedded YARA rules utilises fuzzy hashing and fuzzy rules to improve the outcomes further. Fuzzy rules countenance circumstances where data are imprecise or uncertain, generating a probabilistic outcome indicating the likelihood of whether a file is malware or not. The paper discusses the success of the proposed enhanced YARA rules and embedded YARA rules through several experiments on the collected malware and goodware corpus and their comparative evaluation against YARA rules.

Triaging is an initial stage of malware analysis to assess whether a sample is malware or not and the degree of similarity it holds with known malware. It can be applied to any malware category such as ransomware, which is a type of malware that blocks access to a system or data, usually by encrypting it. It has become the main modus operandi for cybercriminals to extort monies from victims due to the growth of cryptocurrencies. Consequently, it severely affects all types of users whether they be from corporates or ordinary home users. Ransomware can be prevented in several different ways, however, the simple and initial step in prevention is its triaging without execution. Several triaging methods are in use such as fuzzy hashing, import hashing and YARA rules, amongst all, YARA rules are one of the most popular and widely used methods. Nonetheless, its success or failure is dependent on the quality of rules employed for malware triaging. This paper performs ransomware triaging using fuzzy hashing, import hashing and YARA rules and demonstrates how YARA rules can be improved using fuzzy hashing to obtain relatively better triaging results. Subsequently, it proposes the augmented YARA rules fused with fuzzy hashing to obtain improved triaging results and performance efficiency in comparison to all three triaging methods individually. Finally, the paper demonstrates how the use of the fused YARA rules can improve triaging results irrespective of the type of malware.

Feature extraction plays a vital role in visual action recognition. Many existing gradient-based feature extractors, including histogram of oriented gradients (HOG), histogram of optical flow (HOF), motion boundary histograms (MBH), and histogram of motion gradients (HMG), build histograms for representing different actions over the spatio-temporal domain in a video. However, these methods require to set the number of bins for information aggregation in advance. Varying numbers of bins usually lead to inherent uncertainty within the process of pixel voting with regard to the bins in the histogram. This paper proposes a novel method to handle such uncertainty by fuzzifying these feature extractors. The proposed approach has two advantages: i) it better represents the ambiguous boundaries between the bins and thus the fuzziness of the spatio-temporal visual information entailed in videos, and ii) the contribution of each pixel is flexibly controlled by a fuzziness parameter for various scenarios. The proposed family of fuzzy descriptors and a combination of them were evaluated on two publicly available datasets, demonstrating that the proposed approach outperforms the original counterparts and other state-of-the-art methods.

Human activity recognition has become essential to a wide range of applications, such as smart home monitoring, health-care, surveillance. However, it is challenging to deliver a sufficiently robust human activity recognition system from raw sensor data with noise in a smart environment setting. Moreover, imbalanced human activity datasets with less frequent activities create extra challenges for accurate activity recognition. Deep learning algorithms have achieved promising results on balanced datasets, but their performance on imbalanced datasets without explicit algorithm design cannot be promised. Therefore, we aim to realise an activity recognition system using multi-modal sensors to address the issue of class imbalance in deep learning and improve recognition accuracy. This paper proposes a joint diverse temporal learning framework using Long Short Term Memory and one-dimensional Convolutional Neural Network models to improve human activity recognition, especially for less represented activities. We extensively evaluate the proposed method for Activities of Daily Living recognition using binary sensors dataset. A comparative study on five smart home datasets demonstrate that our proposed approach outperforms the existing individual temporal models and their hybridization. Furthermore, this is particularly the case for minority classes in addition to reasonable improvement on the majority classes of human activities.

Since its inception as a solution for secure cryptocurrencies sharing in 2008, the blockchain technology has now become one of the core technologies for secure data sharing and storage over trustless and decentralised peer-to-peer systems. E-government is amongst the systems that stores sensitive information about citizens, businesses and other affiliates, and therefore becomes the target of cyber attackers. The existing e-government systems are centralised and thus subject to single point of failure. This paper proposes a secure and decentralised e-government system based on the consortium blockchain technology, which is a semi-public and decentralised blockchain system consisting of a group of pre-selected entities or organisations in charge of consensus and decisions making for the benefit of the whole network of peers. In addition, a number of e-government nodes are pre-selected to perform the tasks of user and transaction validation before being added to the blockchain network. Accordingly, e-government users of the consortium blockchain network are given the rights to create, submit, access, and review transactions. Performance evaluation on single transaction time and transactions processed per second demonstrate the practicability of the proposed consortium blockchain-based e-government system for secure information sharing amongst all stakeholders., Comment: 9 pages

Electronic government (e-government) uses information and communication technologies to deliver public services to individuals and organisations effectively, efficiently and transparently. E-government is one of the most complex systems which needs to be distributed, secured and privacy-preserved, and the failure of these can be very costly both economically and socially. Most of the existing e-government systems such as websites and electronic identity management systems (eIDs) are centralized at duplicated servers and databases. A centralized management and validation system may suffer from a single point of failure and make the system a target to cyber attacks such as malware, denial of service attacks (DoS), and distributed denial of service attacks (DDoS). The blockchain technology enables the implementation of highly secure and privacy-preserving decentralized systems where transactions are not under the control of any third party organizations. Using the blockchain technology, exiting data and new data are stored in a sealed compartment of blocks (i.e., ledger) distributed across the network in a verifiable and immutable way. Information security and privacy are enhanced by the blockchain technology in which data are encrypted and distributed across the entire network. This paper proposes a framework of a decentralized e-government peer-to-peer (p2p) system using the blockchain technology, which can ensure both information security and privacy while simultaneously increasing the trust of the public sectors. In addition, a prototype of the proposed system is presented, with the support of a theoretical and qualitative analysis of the security and privacy implications of such system., Comment: 11 pages

Egocentric video analysis is an important tool in healthcare that serves a variety of purposes, such as memory aid systems and physical rehabilitation, and feature extraction is an indispensable process for such analysis. Local feature descriptors have been widely applied due to their simple implementation and reasonable efficiency and performance in applications. This paper proposes an enhanced spatial and temporal local feature descriptor extraction method to boost the performance of action classification. The approach allows local feature descriptors to take advantage of saliency maps, which provide insights into visual attention. The effectiveness of the proposed method was validated and evaluated by a comparative study, whose results demonstrated an improved accuracy of around 2%

This paper designs an accurate and low-cost phishing detection sensor by exploring deep learning techniques. Phishing is a very common social engineering technique. The attackers try to deceive online users by mimicking a uniform resource locator (URL) and a webpage. Traditionally, phishing detection is largely based on manual reports from users. Machine learning techniques have recently been introduced for phishing detection. With the recent rapid development of deep learning techniques, many deep-learning-based recognition methods have also been explored to improve classification performance. This paper proposes a light-weight deep learning algorithm to detect the malicious URLs and enable a real-time and energy-saving phishing detection sensor. Experimental tests and comparisons have been conducted to verify the efficacy of the proposed method. According to the experiments, the true detection rate has been improved. This paper has also verified that the proposed method can run in an energy-saving embedded single board computer in real-time.

Meta-learning has been widely used for implementing few-shot learning and fast model adaptation. One kind of meta-learning methods attempt to learn how to control the gradient descent process in order to make the gradient-based learning have high speed and generalization. This work proposes a method that controls the gradient descent process of the model parameters of a neural network by limiting the model parameters in a low-dimensional latent space. The main challenge of this idea is that a decoder with too many parameters is required. This work designs a decoder with typical structure and shares a part of weights in the decoder to reduce the number of the required parameters. Besides, this work has introduced ensemble learning to work with the proposed approach for improving performance. The results show that the proposed approach is witnessed by the superior performance over the Omniglot classification and the miniImageNet classification tasks., Comment: 13papers

Deep learning and other big data technologies have over time become very powerful and accurate. There are algorithms and models developed that have near human accuracy in their task. In health care, the amount of data available is massive and hence, these technologies have a great scope in health care. This paper reviews a few interesting contributions to the field specifically to medical imaging, genomics and patient health records., Comment: Accepted at the International Conference on Intelligent Technologies and Applications(INTAP) 2019

Egocentric action recognition has been intensively studied in the fields of computer vision and clinical science with applications in pervasive health-care. The majority of the existing egocentric action recognition techniques utilize the features extracted from either the entire contents or the regions of interest (ROI) in video frames as the inputs of action classifiers. The former might suffer from moving backgrounds or irrelevant foregrounds usually associated with egocentric action videos, while the latter may be impaired by the mismatch between the calculated and the ground truth ROI. This paper proposes a new gaze-informed feature extraction approach, by which the features are extracted from the regions around the gaze points and thus representing the genuine ROI from a first person of view. The activity of daily life can then be classified based only on the identified regions using the extracted gaze-informed features. The proposed approach has been further applied to a memory support system for people with poor memory, such as those with amnesia or dementia, and their carers. The experimental results demonstrate the efficacy of the proposed approach in egocentric action recognition, and thus the potential of the memory support tool in health care.