High-quality X-rays are now available to diagnose lung diseases with the help of radiologists. However, the diagnostic process is time consuming and depends on specialist availability in medical institutions. Patient ...
High-quality X-rays are now available to diagnose lung diseases with the help of radiologists. However, the diagnostic process is time consuming and depends on specialist availability in medical institutions. Patient information may include chest X-rays of varying quality, medical test results, doctors’ notes and prescriptions, and medication details, among others. In this study, we present a model for classifying pulmonary diseases using multimodal data from patient clinical studies and radiographic images. Various methods were used to generate artificial samples for both images and tabular data on the laboratory study results during data preparation. We also proposed a method for establishing a correspondence between the generated modals. The late fusion architecture of the proposed multimodal model was implemented. We conducted experiments on pulmonary data-set with two modalities. Results shows that an increase in accuracy and other parameters were observed for multimodal data fusion using our model in comparison with image only modality and clinical data only modality. It strengthen the fact that multimodality provides more insight for the fusion model to learn and provide a more precise diagnosis than single modality.
Recently, because of the high-quality representations of contrastive learning methods, rehearsal-based contrastive continual learning has been proposed to explore how to continually learn transferable representation e...
详细信息
ISBN:
(纸本)1577358872
Recently, because of the high-quality representations of contrastive learning methods, rehearsal-based contrastive continual learning has been proposed to explore how to continually learn transferable representation embeddings to avoid the catastrophic forgetting issue in traditional continual settings. Based on this framework, we propose Contrastive Continual learning via Importance Sampling (CCLIS) to preserve knowledge by recovering previous data distributions with a new strategy for Replay Buffer Selection (RBS), which minimize estimated variance to save hard negative samples for representation learning with high quality. Furthermore, we present the Prototype-instance Relation Distillation (PRD) loss, a technique designed to maintain the relationship between prototypes and sample representations using a self-distillation process. Experiments on standard continual learning benchmarks reveal that our method notably outperforms existing baselines in terms of knowledge preservation and thereby effectively counteracts catastrophic forgetting in online contexts. The code is available at https://***/lijy373/CCLIS.
We present a temporary keyword search over sensitive and confidential health data in a cloud environment. The cloud constitutes a semi-trusted domain, making it necessary for data owners to secure their data before ou...
We present a temporary keyword search over sensitive and confidential health data in a cloud environment. The cloud constitutes a semi-trusted domain, making it necessary for data owners to secure their data before outsourcing it through techniques like encryption. Attribute-based keyword search techniques tend to perform a search operation using a search token generated by an authorized user. These search tokens can lead to serious privacy threats, as they can extract all ciphertexts that may have been generated along with their keyword. Therefore, restricting search tokens to extract ciphertexts generated within a time interval is a more promising solution. In this paper, we present a novel ciphertext policy fine-grained temporary keyword that prevents the misuse of these search tokens. Further, it mitigates the risk of insider threats within healthcare organizations by limiting the window of opportunity for unauthorized access to minimum. To assess the security, our proposed scheme is formally proven to be secure against Selectively Chosen Keyword Attacks in the generic bilinear group model. Additionally, we demonstrate that the encryption algorithm's complexity is linear in relation to the number of attributes. Our scheme's significance and practicality are revealed by the performance evaluation.
The concept of federated learning (FL) is becoming increasingly popular as a method for training collaborative models without loss the sensitive information. The term has become ubiquitous due to the extensive develop...
详细信息
The concept of federated learning (FL) is becoming increasingly popular as a method for training collaborative models without loss the sensitive information. The term has become ubiquitous due to the extensive development of autonomous vehicles. Vehicular Networks and the Internet of Vehicles (IoV) enable cooperative learning through federated learning. It is still necessary to address several technical challenges. In recent years, Federated learning (FL) has attracted significant interest in various sectors, including smart cities and transportation systems. FL-enabled attack detection for IoVs are still in its infancy. However, to determine the main challenges of deployment in real-world scenarios, there needs to be research efforts from various areas. Performance metrics are used to evaluate the effectiveness of the proposed FL framework. According to experiments, the proposed FL approach detected attacks in IOV networks with a maximum accuracy of 99.72%. In addition to precision, recall, and F1 scores, 99.70%, 99.20%, and 99.26% were achieved. A comparison of the proposed model with the existing model shows that the proposed model is more accurate.
Federated learning (FL) as a new decentralized learning/computing technique has potential advantages (e.g., accelerating computation task processing and protecting user privacy) for edge intelligence. However, due to ...
详细信息
Federated learning (FL) as a new decentralized learning/computing technique has potential advantages (e.g., accelerating computation task processing and protecting user privacy) for edge intelligence. However, due to limited computing/ caching capacities of network edges and dynamic arrivals of computation tasks, edge intelligence with FL cannot appropriately offload and effectively process computation tasks, which will degrade multi-user quality of experience (QoE). To address these challenges, it is critical to enhance the cooperation of network edges and quantify the multi-user QoE. In this article, we investigate the issue of cooperative edge intelligence by considering federated multi-agent reinforcement learning to enhance the multi-user QoE. Particularly, we present a cooperative edge intelligence architecture with vertical-horizontal cooperation supporting computation offloading. We model a comprehensive system cost to quantify the multi-user QoE and formulate the optimization problem as minimizing the expected long-term system cost. We further propose a decentralized intelligent offloading framework based on soft actor-critic and FL with an attention mechanism. Evaluation results demonstrate that the proposed scheme outperforms existing offloading schemes in terms of convergence and multi-user QoE. Finally, we discuss several open issues and opportunities of edge intelligence with FL.
6D pose estimation has garnered significant attention and research. RGB images and point clouds converted from RGB-D images provide complementary color and geometry information, making them the mainstream data sources...
详细信息
6D pose estimation has garnered significant attention and research. RGB images and point clouds converted from RGB-D images provide complementary color and geometry information, making them the mainstream data sources for object 6D pose estimation. However, due to the fact that RGB image and point cloud belong to different dimensional spaces and have different distribution characteristics, the fusion of these two complementary data sources remains a key technical challenge for 6D pose estimation. In contrast to prior approaches that simply concatenate separately processed RGB images and point clouds, this work introduces a Transformer -based multi -modal fusion network to address this challenge. More precisely, We build a Transformer architecture based pixel -wise feature extraction to optimize feature extraction from RGB images and point clouds. Subsequently, we investigate various multi -modal feature fusion methods to process these features, enabling deeper fusion of complementary data. Additionally, during the experimental phase, we design a 6D pose estimation network based on depth prediction to assess the impact of point cloud accuracy on the multi -modal fusion module. Finally, the proposed method is verified on four datasets: LineMOD, Occlusion Linemod, MP6D and YCB-Video. Experimental results show that the proposed method outperforms similar methods on these datasets.
Warts are benign tumors, caused due to the infection of human papillomavirus (HPV). The identification of wart-specific treatment methods is pertaining to major challenges such as class imbalance, prediction accuracy,...
详细信息
Warts are benign tumors, caused due to the infection of human papillomavirus (HPV). The identification of wart-specific treatment methods is pertaining to major challenges such as class imbalance, prediction accuracy, and biased nature of learning algorithm. In this article, a bagged ensemble of cost-sensitive extra tree classifier (BECSETC) is developed toward the selection of wart-specific treatment methods. BECSETC outperforms the state-of-the-art techniques (SOTA) by a margin of (0-45, 0-31.60), (0-12, 0-2.6) in terms of sensitivity and specificity which overcome the imbalanced distribution on both immunotherapy and cryotherapy datasets. However, on merged dataset, BECSETC algorithm gave an improvement of 6.04-10.57%, and 4.63% in terms of sensitivity and specificity, as compared to SOTA techniques.
Decentralized Stochastic Gradient Descent (SGD) is an emerging neural network training approach that enables multiple agents to train a model collaboratively and simultaneously. Rather than using a central parameter s...
详细信息
Decentralized Stochastic Gradient Descent (SGD) is an emerging neural network training approach that enables multiple agents to train a model collaboratively and simultaneously. Rather than using a central parameter server to collect gradients from all the agents, each agent keeps a copy of the model parameters and communicates with a small number of other agents to exchange model updates. Their communication, governed by the communication topology and gossip weight matrices, facilitates the exchange of model updates. The state-of-the-art approach uses the dynamic one-peer exponential-2 topology, achieving faster training times and improved scalability than the ring, grid, torus, and hypercube topologies. However, this approach requires a power-of-2 number of agents, which is impractical at scale. In this paper, we remove this restriction and propose Decentralized SGD with Communication-optimal Exact Consensus Algorithm (DSGD-CECA), which works for any number of agents while still achieving state-of-the-art properties. In particular, DSGD-CECA incurs a unit per-iteration communication overhead and an (O) over tilde (n(3)) transient iteration complexity. Our proof is based on newly discovered properties of gossip weight matrices and a novel approach to combine them with DSGD's convergence analysis. Numerical experiments show the efficiency of DSGD-CECA.
Video Individual Counting (VIC) aims to predict the number of unique individuals in a single video. Existing methods learn representations based on trajectory labels for individuals, which are annotation-expensive. To...
详细信息
ISBN:
(数字)9798350353006
ISBN:
(纸本)9798350353013
Video Individual Counting (VIC) aims to predict the number of unique individuals in a single video. Existing methods learn representations based on trajectory labels for individuals, which are annotation-expensive. To provide a more realistic reflection of the underlying practical challenge, we introduce a weakly supervised VIC task, wherein trajectory labels are not provided. Instead, two types of labels are provided to indicate traffic entering the field of view (inflow) and leaving the field view (outflow). We also propose the first solution as a baseline that formulates the task as a weakly supervised contrastive learning problem under group-level matching. In doing so, we devise an end-to-end trainable soft contrastive loss to drive the network to distin-guish inflow, outflow, and the remaining. To facilitate future study in this direction, we generate annotations from the existing VIC datasets Sense Crowd and CroHD and also build a new dataset, UAVVIC. Extensive results show that our baseline weakly supervised method outperforms supervised methods, and thus, little information is lost in the transition to the more practically relevant weakly supervised task. The code and trained model can be found at CGNet.
Recent developments in the Internet of Things (IoT) and various communication technologies have reshaped numerous application areas. Nowadays, IoT is assimilated into various medical devices and equipment, leading to ...
详细信息
Recent developments in the Internet of Things (IoT) and various communication technologies have reshaped numerous application areas. Nowadays, IoT is assimilated into various medical devices and equipment, leading to the progression of the Internet of Medical Things (IoMT). Therefore, various IoMT-based healthcare applications are deployed and used in the day-to-day scenario. Traditionally, machinelearning (ML) models use centralized data compilation and learning that is impractical in pragmatic healthcare frameworks due to rising privacy and data security issues. Federated learning (FL) has been observed as a developing distributed collective paradigm, the most appropriate for modern healthcare framework, that manages various stakeholders (e.g., patients, hospitals, laboratories, etc.) to carry out training of the models without the actual exchange of sensitive medical data. Consequently, in this work, the authors present an exhaustive survey on the security of FL-based IoMT applications in smart healthcare frameworks. First, the authors introduced IoMT devices, their types, applications, datasets, and the IoMT security framework in detail. Subsequently, the concept of FL, its application domains, and various tools used to develop FL applications are discussed. The significant contribution of FL in deploying secure IoMT systems is presented by focusing on FL-based IoMT applications, patents, real-world FL-based healthcare projects, and datasets. A comparison of FL-based security techniques with other schemes in the smart healthcare ecosystem is also presented. Finally, the authors discussed the challenges faced and potential future research recommendations to deploy secure FL-based IoMT applications in smart healthcare frameworks.& COPY;2023 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license (http://***/licenses/by/4.0/).
暂无评论