Federated learning (FL) allows multiple clients to collectively train a high-performance global model without sharing their private data. However, the key challenge in federated is that have significant statistical heterogeneity among local data distributions, which would cause inconsistent optimized models on clientside. To address this fundamental dilemma, we propose novel algorithm with drift decoupling and correction (FedDC). Our FedDC only introduces lightweight modifications training...

Trained with the standard cross entropy loss, deep neural networks can achieve great performance on correctly labeled data. However, if training data is corrupted label noise, models tend to overfit noisy labels, thereby achieving poor generation performance. To remedy this issue, several loss functions have been proposed and demonstrated be robust noise. Although most of stem from Categorical Cross Entropy (CCE) they fail embody intrinsic relationships between CCE other functions. In paper,...

Partial label learning aims to induce a multi-class classifier from training examples where each of them is associated with set candidate labels, among which only one the ground-truth label. The common strategy train predictive model disambiguation, i.e. differentiating modeling outputs individual labels so as recover labeling information. Recently, feature-aware disambiguation was proposed generate different confidences over by utilizing graph structure feature space. However, existence...

In partial label learning, a multi-class classifier is learned from the ambiguous supervision where each training example associated with set of candidate labels among which only one valid. An intuitive way to deal this problem <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">label disambiguation</i> , i.e., differentiating labeling confidences different so as try recover ground-truth information. Recently, feature-aware disambiguation has...

In this paper, we propose a novel approach to classify short texts by combining both their lexical and semantic features. We present an improved measurement method for feature selection furthermore obtain the features with background knowledge repository which covers target category domains. The combination of is achieved mapping words topics different weights. way, dimensionality space reduced number topics. here use Wikipedia as employ Support Vector Machine (SVM) classifier. experiment...

In this paper, a novel quantum encryption algorithm for color image is proposed based on multiple discrete chaotic systems.The utilize the controlled-NOT generated by logistic map, asymmetric tent map and Chebyshev to control XOR operation in process.Experiment results analysis show that has high efficiency security against differential statistical attacks.

Noise exposure is one of the most common causes hearing loss and hyperacusis. Studies have shown that noise can induce a cortical gain to compensate for reduced input cochlea, which may contribute increased sound sensitivity. However, many people with hyperacusis no measurable cochlear lesion after being exposed loud sound. In this experiment, we studied neurological alterations in subcortical areas following prolonged moderate level (84 dB SPL, 8 h/day 4 weeks) laboratory mice. The function...

Graph neural networks (GNNs) integrate deep architectures and topological structure modeling in an effective way. However, the performance of existing GNNs would decrease significantly when they stack many layers, because over-smoothing issue. Node embeddings tend to converge similar vectors keep recursively aggregating representations neighbors. To enable GNNs, several methods have been explored recently. But are developed from either techniques convolutional or heuristic strategies. There...

The paper presents an adaptive wear-leveling scheme based on several wear-thresholds in different periods. basic idea behind this is that blocks can have wear-out speeds and the mechanism does not conduct data migration until erasure counts of some hot hit a threshold. Through series emulation experiments realistic disk traces, we show proposed reduce total yield uniform among all at late lifetime storage devices. As result, only performance systems be advanced, lifespan flash-based memory...

This paper presents an initiative data prefetching scheme on the storage servers in distributed file systems for cloud computing. In this technique, client machines are not substantially involved process of prefetching, but can directly prefetch after analyzing history disk I/O access events, and then send prefetched to relevant proactively. To put technique work, information about nodes is piggybacked onto real requests, forwarded server. Next, two prediction algorithms have been proposed...

Small object detection is widely used in the real world. Detecting small objects complex scenes extremely difficult as they appear with low resolution. At present, many studies have made significant progress improving accuracy of objects. However, some them cannot balance speed and well. To solve above problems, a lightweight multi-scale network (LMSN) was proposed to exploit information this article. Firstly, it explicitly modeled semantic interactions at every scale via feature fusion...