We study the problem of generating inferential texts of events for a variety of commonsense like if-else relations. Existing approaches typically use limited evidence from training examples and learn for each relation...
详细信息
The information interaction when the brain processes emotional activities is intricate. Therefore, it is very necessary for us to explore the mechanisms of the functional coordination of various brain regions. Recent ...
详细信息
ISBN:
(数字)9781728189543
ISBN:
(纸本)9781728189550
The information interaction when the brain processes emotional activities is intricate. Therefore, it is very necessary for us to explore the mechanisms of the functional coordination of various brain regions. Recent studies have demonstrated that there are hub regions in the brain that are highly connected and highly central, and are responsible for coordinating dynamic interactions between brain regions. This study aims to construct the dynamic brain networks related to emotions and examine the structure of the hubs of these networks that reflect functional connectivity. In this paper, we demonstrate that a combination of network hubs set a closely connected collective called “rich-club”,which can be used as the core network architecture for identifying different emotions. According to the phase synchronization relationship of the Electroencephalogram (EEG) in three time periods to construct the corresponding brain networks in different sub-states, we found that the dynamic connection patterns produced by different emotions are different, which is manifested in the different rich-club organization. The discovery of rich-club structure related to emotions in this study is of great significance for studying the brain mechanisms behind human emotional processes.
Hashing techniques have been seen huge adoption in large-scale retrieval owing to the computational and storage efficiencies of binary codes. However, the binary codes might not preserve the semantic similarity powerf...
详细信息
The identification of active binding drugs for target proteins (termed as drug-target interaction prediction) is the key challenge in virtual screening, which plays an essential role in drug discovery. Although recent...
详细信息
Federated collaborative filtering (Fed-CF) is a variant of federated learning (FL) models, which can protect user privacy in recommender systems. In Fed-CF, the recommendation model is collectively trained across mult...
详细信息
ISBN:
(数字)9781728176260
ISBN:
(纸本)9781728176277
Federated collaborative filtering (Fed-CF) is a variant of federated learning (FL) models, which can protect user privacy in recommender systems. In Fed-CF, the recommendation model is collectively trained across multiple decentralized clients by exchanging gradients only. However, the decentralized nature of Fed-CF makes it vulnerable to shilling attacks, which can be realized by inserting fake ratings of target items to distort recommendation results. Unfortunately, previous detection algorithms cannot work well in the FL framework, as all original data samples are not disclosed at all. In this paper, we are the first to systematically study the problem of shilling attacks in the context of federated learning, and propose an effective detection method called Federated Shilling Attack Detector (FSAD) to detect shilling attackers in Fed-CF. We first show the feasibility of shilling attacks in Fed-CF. Next, we dedicatedly design four novel features based on exchanged gradients among clients. By incorporating these gradient-based features, we train a semi-supervised Bayes classifier to identify shilling attackers effectively. Finally, we conduct extensive experiments based on real-world datasets to evaluate the performance of our proposed FSAD method. The experimental results show that FSAD can detect shilling attackers in Fed-CF with high accuracy, with the F 1 value as high as 0.90 on the Netflix dataset, which approaches the performance of the optimal detector that utilizes complete private user information for detection.
In spite that Federated Learning (FL) is well known for its privacy protection when training machine learning models among distributed clients collaboratively, recent studies have pointed out that the naive FL is susc...
详细信息
Although the Convolutional Neural Network (CNN) has been shown to be effective in solving image classification tasks, its architecture is often difficult to design (i.e., much tuning is required for optimization) due ...
详细信息
It is always a challenging problem to deliver a huge volume of videos over the Internet. To meet the high bandwidth and stringent playback demand, one feasible solution is to cache video contents on edge servers based...
详细信息
Background and Objective: Accurate lung tumor segmentation is crucial for early clinical diagnosis and subsequent treatment planning for patients. Conducting centralized training to enhance model performance is imprac...
详细信息
Background and Objective: Accurate lung tumor segmentation is crucial for early clinical diagnosis and subsequent treatment planning for patients. Conducting centralized training to enhance model performance is impractical in real-world scenarios. Although federated learning offers a promising solution to this issue, effectively and efficiently utilizing unlabeled data remains a challenge. Methods: In this work, we propose a novel federated semi-supervised learning framework (FSSL) for complex and realistic scenario, which can efficiently leverage the limited labeled data and abundant unlabeled data from multiple clinical sites. Specifically, we initially conduct federated self-supervised pre-training at completely unlabeled sites using masked image modeling strategy. Subsequently, in the fine-tuning stage, we propose a pseudo supervision refinement strategy to leverage the unlabeled data from partially labeled sites, which not only reduces pseudo label noise but also helps stabilize the training process. Moreover, we propose a dynamic model aggregation strategy to assist the server in dynamically combining all local models during communication round. Results: We conducted extensive comparative and ablation experiments to validate the effectiveness of our proposed method. When only a limited amount of labeled data is available in the partially labeled sites, even without loading the pre-trained model, our approach consistently achieves the highest average Dice scores of 0.7383, 0.4768, and 0.6629 on the internal test sets, external unseen test sets, and pre-training test sets, respectively. Upon loading the pre-trained model, the scores on the three test sets further improve to 0.8205, 0.6185, and 0.7334, respectively. Conclusion: Our proposed method is capable of collaboratively training a global model with robust segmentation performance and generalizability by efficiently utilizing unlabeled data from different sites, even when only a limited amount of labeled da
Accurate lung tumor segmentation is crucial for early clinical diagnosis and subsequent treatment planning for patients. Centralized model training faces privacy constraints, although federated learning offers a promi...
详细信息
暂无评论