Reconstructed Graph Neural Network With Knowledge Distillation for Lightweight Anomaly Detection

The proliferation of Internet-of-Things (IoT) technologies in modern smart society enables massive data exchange for offering intelligent services. It becomes essential to ensure secure communications while exchanging highly sensitive IoT data efficiently, which leads to high demands for lightweight models or algorithms with limited computation capability provided by individual IoT devices. In this study, a graph representation learning model, which seamlessly incorporates graph neural network (GNN) and knowledge distillation (KD) techniques, named reconstructed graph with global-local distillation (RG-GLD), is designed to realize the lightweight anomaly detection across IoT communication networks. In particular, a new graph network reconstruction strategy, which treats data communications as nodes in a directed graph while edges are then connected according to two specifically defined rules, is devised and applied to facilitate the graph representation learning in secure and efficient IoT communications. Both the structural and traffic features are then extracted from the graph data and flow data respectively, based on the graph attention network (GAT) and multilayer perceptron (MLP) techniques. These can benefit the GNN-based KD process in accordance with the more effective feature fusion and representation, considering both structural and data levels across the dynamic IoT networks. Furthermore, a lightweight local subgraph preservation mechanism improved by the graph attention mechanism and downsampling scheme to better utilize the topological information, and a so-called global information alignment defined based on the self-attention mechanism to effectively preserve the global information, are developed and incorporated in a refined graph attention based KD scheme. Compared with four different baseline methods, experiments and evaluations conducted based on two public datasets demonstrate the usefulness and effectiveness of our proposed model in improving the efficiency of knowledge transfer with higher classification accuracy but lower computational load, which can be deployed for lightweight anomaly detection in sustainable IoT computing environments.

Decentralised Global Service Discovery for the Internet of Things

The Internet of Things (IoT) consists of millions of devices deployed over hundreds of thousands of different networks, providing an ever-expanding resource to improve our understanding of and interactions with the physical world. Global service discovery is key to realizing the opportunities of the IoT, spanning disparate networks and technologies to enable the sharing, discovery, and utilisation of services and data outside of the context in which they are deployed. In this paper, we present Decentralised Service Registries (DSRs), a novel trustworthy decentralised approach to global IoT service discovery and interaction, building on DSF-IoT to allow users to simply create and share public and private service registries, to register and query for relevant services, and to access both current and historical data published by the services they discover. In DSR, services are registered and discovered using signed objects that are cryptographically associated with the registry service, linked into a signature chain, and stored and queried for using a novel verifiable DHT overlay. In contrast to existing centralised and decentralised approaches, DSRs decouple registries from supporting infrastructure, provide privacy and multi-tenancy, and support the verification of registry entries and history, service information, and published data to mitigate risks of service impersonation or the alteration of data. This decentralised approach is demonstrated through the creation and use of a DSR to register and search for real-world IoT devices and their data as well as qualified using a scalable cluster-based testbench for the high-fidelity emulation of peer-to-peer applications. DSRs are evaluated against existing approaches, demonstrating the novelty and utility of DSR to address key IoT challenges and enable the sharing, discovery, and use of IoT services.

Information Theoretic Learning-Enhanced Dual-Generative Adversarial Networks With Causal Representation for Robust OOD Generalization

Recently, machine/deep learning techniques are achieving remarkable success in a variety of intelligent control and management systems, promising to change the future of artificial intelligence (AI) scenarios. However, they still suffer from some intractable difficulty or limitations for model training, such as the out-of-distribution (OOD) issue, in modern smart manufacturing or intelligent transportation systems (ITSs). In this study, we newly design and introduce a deep generative model framework, which seamlessly incorporates the information theoretic learning (ITL) and causal representation learning (CRL) in a dual-generative adversarial network (Dual-GAN) architecture, aiming to enhance the robust OOD generalization in modern machine learning (ML) paradigms. In particular, an ITL-and CRL-enhanced Dual-GAN (ITCRL-DGAN) model is presented, which includes an autoencoder with CRL (AE-CRL) structure to aid the dual-adversarial training with causality-inspired feature representations and a Dual-GAN structure to improve the data augmentation in both feature and data levels. Following a newly designed feature separation strategy, a causal graph is built and improved based on the information theory, which can enhance the causally related factors among the separated core features and further enrich the feature representation with the counterfactual features via interventions based on the refined causal relationships. The ITL is incorporated to improve the extraction of low-dimensional feature representations and learn the optimized causal representations based on the idea of "information flow." A dual-adversarial training mechanism is then developed, which not only enables the generator to expand the boundary of feature distribution in accordance with the optimized feature representation from AE-CRL, but also allows the discriminator to further verify and improve the quality of the augmented data for OOD generalization. Experiment and evaluation results based on an open-source dataset demonstrate the outstanding learning efficiency and classification performance of our proposed model for robust OOD generalization in modern smart applications compared with three baseline methods.

Human Behavior Recognition via Hierarchical Patches Descriptor and Approximate Locality-Constrained Linear Coding

Human behavior recognition technology is widely adopted in intelligent surveillance, human-machine interaction, video retrieval, and ambient intelligence applications. To achieve efficient and accurate human behavior recognition, a unique approach based on the hierarchical patches descriptor (HPD) and approximate locality-constrained linear coding (ALLC) algorithm is proposed. The HPD is a detailed local feature description, and ALLC is a fast coding method, which makes it more computationally efficient than some competitive feature-coding methods. Firstly, energy image species were calculated to describe human behavior in a global manner. Secondly, an HPD was constructed to describe human behaviors in detail through the spatial pyramid matching method. Finally, ALLC was employed to encode the patches of each level, and a feature coding with good structural characteristics and local sparsity smoothness was obtained for recognition. The recognition experimental results on both Weizmann and DHA datasets demonstrated that the accuracy of five energy image species combined with HPD and ALLC was relatively high, scoring 100% in motion history image (MHI), 98.77% in motion energy image (MEI), 93.28% in average motion energy image (AMEI), 94.68% in enhanced motion energy image (EMEI), and 95.62% in motion entropy image (MEnI).

Adaptive Wireless Network Management with Multi-Agent Reinforcement Learning

Wireless networks are trending towards large scale systems, containing thousands of nodes, with multiple co-existing applications. Congestion is an inevitable consequence of this scale and complexity, which leads to inefficient use of the network capacity. This paper proposes an autonomous and adaptive wireless network management framework, utilising multi-agent deep reinforcement learning, to achieve efficient use of the network. Its novel reward function incorporates application awareness and fairness to address both node and network level objectives. Our experimental results demonstrate the proposed approach's ability to be optimised for application-specific requirements, while optimising the fairness of the network. The results reveal significant performance benefits in terms of adaptive data rate and an increase in responsiveness compared to a single-agent approach. Some significant qualitative benefits of the multi-agent approach-network size independence, node-led priorities, variable iteration length, and reduced search space-are also presented and discussed.

Housing Risk Factors Associated with Respiratory Disease: A Systematic Review

Cold, damp and mouldy housing arises from the degradation of the housing stock over time due to weathering and a lack of maintenance. Living in such houses is associated with many adverse impacts on human health, especially for those with existing health issues. This paper presents a systematic review, using the PRISMA protocol, consisting of an exploratory analysis of housing-related risk factors associated with respiratory disease. The review consisted of 360 studies investigating 19 risk factors associated with respiratory conditions. Each fall into one of four categories, namely, (1) outdoor environment-related factors; (2) indoor air pollution-related factors; (3) housing non-structure-related factors; or (4) housing structure-related factors. The results show that effects of poor housing conditions on occupants’ respiratory health is a growing research field, where poor indoor air quality, mainly due to a lack of adequate ventilation, was found to be the most influential risk factor. Usage of solid fuel and living in an urban area without a pollutant-free air filtration system are the main risk factors related to inadequate ventilation. Therefore, an adequate and reliable ventilation system with air-infiltration was considered to be the main mitigation solution to improve indoor air quality. It is suggested that government organisations and health practitioners could use the identified risk factors to measure the healthiness of existing dwellings and take measures to improve existing conditions and develop regulations for new housing construction to promote the healthy home concept. Further research is needed for risk mitigation strategies to reduce the respiratory health burden attributed to housing.

Fusing Appearance and Spatio-Temporal Models for Person Re-Identification and Tracking

Knowing who is where is a common task for many computer vision applications. Most of the literature focuses on one of two approaches: determining who a detected person is (appearance-based re-identification) and collating positions into a list, or determining the motion of a person (spatio-temporal-based tracking) and assigning identity labels based on tracks formed. This paper presents a model fusion approach, aiming towards combining both sources of information together in order to increase the accuracy of determining identity classes for detected people using re-ranking. First, a Sequential k-Means re-identification approach is presented, followed by a Kalman filter-based spatio-temporal tracking approach. A linear weighting approach is used to fuse the outputs from these models together, with modification of the weights using a decay function and a rule-based system to reflect the strengths and weaknesses of the models under different conditions. Preliminary experimental results with two different person detection algorithms on an indoor person tracking dataset show that fusing the appearance and spatio-temporal models significantly increases the overall accuracy of the classification operation.

A Novel Accelerometer-Based Technique for Robust Detection of Walking Direction

OBJECTIVE

Distance estimation in pedestrian dead reckoning is acquired using vector norm of accelerations, which results in positive values. However, anteroposterior acceleration is negative when a step is taken backward, which must be detected for accurate localization. This paper proposes a novel approach for the detection of walking direction, which uses a dominant trend duration.

METHODS

The approach evaluates anteroposterior acceleration out of a foot-worn accelerometer for temporal dominance of acceleration trends during swing phase of the walk. The approach is tested for forward and backward walks with speed variations on a straight path as well as for forward walk at normal speed on a turning path. To validate the detection accuracy, success rates per participant per walk trial are calculated and then overall success rate for all the trials are reported. Moreover, metrics precision, recall and F1 scores are calculated for detection reliability in both directions.

RESULTS

Overall 98 ± 2% detection accuracy is achieved on linear path considering both directions and all speed variations, whereas 93 ± 7% on turning path including left and right turns. In comparison with the state-of-the-art bidirectional detection approach, the proposed approach delivers accurate detection with speed variations without requiring prior training and relies on a single sensory feature.

CONCLUSION

Dominant trend duration is a novel and reliable feature to detect directional changes during communal walk with speed variation.

SIGNIFICANCE

The approach can be employed in different contexts, such as enabling pedestrian localization approaches to accommodate back stepping or any application that requires knowledge of changing directions while walking.

Bipedal gait model for precise gait recognition and optimal triggering in foot drop stimulator: a proof of concept

Electrical stimulators are often prescribed to correct foot drop walking. However, commercial foot drop stimulators trigger inappropriately under certain non-gait scenarios. Past researches addressed this limitation by defining stimulation control based on automaton of a gait cycle executed by foot drop of affected limb/foot only. Since gait is a collaborative activity of both feet, this research highlights the role of normal foot for robust gait detection and stimulation triggering. A novel bipedal gait model is proposed where gait cycle is realized as an automaton based on concurrent gait sub-phases (states) from each foot. The input for state transition is fused information from feet-worn pressure and inertial sensors. Thereafter, a bipedal gait model-based stimulation control algorithm is developed. As a feasibility study, bipedal gait model and stimulation control are evaluated in real-time simulation manner on normal and simulated foot drop gait measurements from 16 able-bodied participants with three speed variations, under inappropriate triggering scenarios and with foot drop rehabilitation exercises. Also, the stimulation control employed in commercial foot drop stimulators and single foot gait-based foot drop stimulators are compared alongside. Gait detection accuracy (98.9%) and precise triggering under all investigations prove bipedal gait model reliability. This infers that gait detection leveraging bipedal periodicity is a promising strategy to rectify prevalent stimulation triggering deficiencies in commercial foot drop stimulators. Graphical abstract Bipedal information-based gait recognition and stimulation triggering.

A Single RF Emitter-Based Indoor Navigation Method for Autonomous Service Robots

Location-aware services are one of the key elements of modern intelligent applications. Numerous real-world applications such as factory automation, indoor delivery, and even search and rescue scenarios require autonomous robots to have the ability to navigate in an unknown environment and reach mobile targets with minimal or no prior infrastructure deployment. This research investigates and proposes a novel approach of dynamic target localisation using a single RF emitter, which will be used as the basis of allowing autonomous robots to navigate towards and reach a target. Through the use of multiple directional antennae, Received Signal Strength (RSS) is compared to determine the most probable direction of the targeted emitter, which is combined with the distance estimates to improve the localisation performance. The accuracy of the position estimate is further improved using a particle filter to mitigate the fluctuating nature of real-time RSS data. Based on the direction information, a motion control algorithm is proposed, using Simultaneous Localisation and Mapping (SLAM) and A* path planning to enable navigation through unknown complex environments. A number of navigation scenarios were developed in the context of factory automation applications to demonstrate and evaluate the functionality and performance of the proposed system.

A Hybrid Indoor Localization and Navigation System with Map Matching for Pedestrians Using Smartphones

Pedestrian dead reckoning is a common technique applied in indoor inertial navigation systems that is able to provide accurate tracking performance within short distances. Sensor drift is the main bottleneck in extending the system to long-distance and long-term tracking. In this paper, a hybrid system integrating traditional pedestrian dead reckoning based on the use of inertial measurement units, short-range radio frequency systems and particle filter map matching is proposed. The system is a drift-free pedestrian navigation system where position error and sensor drift is regularly corrected and is able to provide long-term accurate and reliable tracking. Moreover, the whole system is implemented on a commercial off-the-shelf smartphone and achieves real-time positioning and tracking performance with satisfactory accuracy.