Search

Ubiquitous computing has potentials to harness the flexibility of distributed computing systems including cloud, edge, and internet of things devices. Mobile edge computing (MEC) benefits time-critical applications by providing low latency connections. However, most of the resource-constrained edge devices are not computationally feasible to host deep learning (DL) solutions. Further, these edge devices if deployed under denser deployments may result in topological dependencies which if not taken into consideration may adversely affect the MEC performance. To bring more intelligence to the edge under topological dependencies, compared to optimization heuristics, this work proposes a novel collaborative distributed DL approach. The proposed approach exploits topological dependencies of the edge using a resource-optimized graph neural network (GNN) version with an accelerated inference. By exploiting edge collaborative learning using stochastic gradient (SGD), the proposed approach called CGNN-edge ensures fast convergence and high accuracy. Collaborative learning of the deployed CGNN-edge incurs extra communication overhead and latency. To cope, this work proposes compressed collaborative learning based on momentum correction called cCGNN-edge with better scalability while preserving accuracy. Performance evaluation under IEEE 802.11ax-high-density wireless local area networks deployment demonstrates that both the schemes outperform cloud-based GNN inference in response time, satisfaction of latency requirements, and communication overhead.

Mammography is a popular diagnostic imaging procedure for detecting breast cancer at an early stage. Various deep-learning approaches to breast cancer detection incur high costs and are erroneous. Therefore, they are not reliable to be used by medical practitioners. Specifically, these approaches do not exploit complex texture patterns and interactions. These approaches warrant the need for labelled data to enable learning, limiting the scalability of these methods with insufficient labelled datasets. Further, these models lack generalisation capability to new-synthesised patterns/textures. To address these problems, in the first instance, we design a graph model to transform the mammogram images into a highly correlated multigraph that encodes rich structural relations and high-level texture features. Next, we integrate a pre-training self-supervised learning multigraph encoder (SSL-MG) to improve feature presentations, especially under limited labelled data constraints. Then, we design a semi-supervised mammogram multigraph convolution neural network downstream model (MMGCN) to perform multi-classifications of mammogram segments encoded in the multigraph nodes. Our proposed frameworks, SSL-MGCN and MMGCN, reduce the need for annotated data to 40% and 60%, respectively, in contrast to the conventional methods that require more than 80% of data to be labelled. Finally, we evaluate the classification performance of MMGCN independently and with integration with SSL-MG in a model called SSL-MMGCN over multi-training settings. Our evaluation results on DSSM, one of the recent public datasets, demonstrate the efficient learning performance of SSL-MNGCN and MMGCN with 0.97 and 0.98 AUC classification accuracy in contrast to the multitask deep graph (GCN) method Hao Du et al. (2021) with 0.81 AUC accuracy.

The IEEE 802.11ax standard is designed to provide high-efficiency WLAN operating in dense deployment, with a focus on increasing robustness and uplink transmission. IEEE 802.11ax aims to deliver self-configuration and self-adaptation capabilities for interference management techniques in dense deployments, to improve overall network conditions. The IEEE 802.11ax medium strives to intelligently use network allocation vectors (NAV), which block simultaneous transmission of neighboring co-existing access points from transmitting for a set period of time. Given an acceptable collision probability, this collision avoidance strategy tends to degrade network throughput and increase transmission delays since the medium is never completely exploited. In this paper, we propose affinityGNN, a graph neural network approach to actively optimize network co-existence transmission in a dense scenario. We introduce direct-affinityGNN which considers direct communications between IEEE802.11ax-compliant wireless systems and in contrast, skip-affinityGNN, which considers communication between IEEE802.11ax and Legacy devises by aggregating neural messages from direct and skip (two-hop) communicating neighbors in the dense wireless network deployment. Extensive experimental results not only show the superior performance of our proposed model over the state-of-the-art, but also demonstrate its potentially good interpretability and robustness for wireless network structure.

Hybrid FSO/RF system requires an efficient FSO and RF link switching mechanism to improve the system capacity by realizing the complementary benefits of both the links. The dynamics of network conditions, such as fog, dust, and sand storms compound the link switching problem and control complexity. To address this problem, we initiate the study of deep reinforcement learning (DRL) for link switching of hybrid FSO/RF systems. Specifically, in this work, we focus on actor-critic called Actor/Critic-FSO/RF and Deep-Q network (DQN) called DQN-FSO/RF for FSO/RF link switching under atmospheric turbulences. To formulate the problem, we define the state, action, and reward function of a hybrid FSO/RF system. DQN-FSO/RF frequently updates the deployed policy that interacts with the environment in a hybrid FSO/RF system, resulting in high switching costs. To overcome this, we lift this problem to ensemble consensus-based representation learning for deep reinforcement called DQNEnsemble-FSO/RF. The proposed novel DQNEnsemble-FSO/RF DRL approach uses consensus learned features representations based on an ensemble of asynchronous threads to update the deployed policy. Experimental results corroborate that the proposed DQNEnsemble-FSO/RF’s consensus learned features switching achieves better performance than Actor/Critic-FSO/RF, DQN-FSO/RF, and MyOpic for FSO/RF link switching while keeping the switching cost significantly low.

Mammography is a popular diagnostic imaging procedure for detecting breast cancer at an early stage. Various deep learning (DL) approaches to breast cancer detection incur high costs and are prone to classify incorrectly. Therefore, they are not sufficiently reliable to replace existing techniques used by medical practitioners. Specifically, these DL approaches do not exploit the complex texture patterns and interactions in mammograms. These DL approaches need labelled data to enable learning, limiting the scalability of these methods because of sufficient labelled datasets. Further, these DL models lack generalisation capability to newly-synthesised patterns/textures. To address these problems, in the first instance, we design a graph model to transform the mammogram images into highly correlated multi-graphs, encoding rich structural relations and high-level texture features. Then, we consider a self-supervised learning multi-graph encoder (SSL-MG) to improve the features presentation, especially when limited labelled data is available. Finally, we design a semi-supervised mammogram multi graph convolution neural network downstream model (MMGCN) to perform multi classification of mammogram segments encoded in the multi-graph nodes. We evaluate the classification performance of MMGCN independently and with integration with SSL-MG in a model called SSL-MMGCN over several training settings. Our results reveal the efficient learning performance of SSL-MNGCN and MMGCN with 0.97 and 0.98 AUC classification accuracy in contrast to the multitask deep graph (GCN) method of Hao Du et al. (2021) with 0.81 AUC accuracy.

Mammography is a popular diagnostic imaging procedure for detecting breast cancer at an early stage. Variousdeep learning approaches to breast cancer detection incur high costs and are erroneous. Therefore, they are notreliable to be used by medical practitioners. Specifically, these approaches do not exploit the complex texture patternsand interactions. These approaches warrant the need for labelled data to enable learning, limiting the scalabilityof these methods under insufficient labelled datasets. Further, these models lack generalisation capability to newsynthesizedpatterns/textures. To address these problems, in the first instance, we design a graph model to transform themammograms images into highly correlated multigraph encode rich structural relations and high-level texture features.Then, we consider a self-supervised learning multigraph encoder (SSL-MG) to improve the features presentation,especially under limited labelled data constraints. Finally, we design a semi-supervised mammogram multigraphconvolution neural network downstream model (MMGCN) to perform multi classifications of mammogram segmentsencoded in the multigraph nodes. We evaluate the classification performance of MMGCN independently and withintegration with SSL-MG in a model called SSL-MMGCN over multi training settings. Our results reveal the efficientlearning performance of SSL-MNGCN and MMGCN with 0.97 and 0.98 AUC classification accuracy in contrast tothe multitask deep graph (GCN) method Hao Du et al. (2021) with 0.81 AUC accuracy.