Mixup is a neural network training method that generates new samples by linear interpolation of multiple and their labels. The mixup has better generalization ability than the traditional empirical risk minimization (ERM). But there lack more intuitive understanding why will perform better. In this paper, several different sample mixing methods are used to test how networks learn infer from mixed illustrate mixups work as data augmentation it regularizes networks. Then, weighting noise...

This work investigates proactive edge caching for device-to-device (D2D)-assisted wireless networks, where user equipment (UE) can be selected as nodes to assist content delivery reduce the transmission latency. In doing so, there are two challenges: 1) how precisely get user's preference cache proper contents at UEs and 2) replace cached when new popular emerging. To address these, we develop a learning-based (UPL-PEC) strategy. strategy, first propose novel context social-aware learning...

In this paper, we find that ubiquitous time series (TS) forecasting models are prone to severe overfitting. To cope with problem, embrace a de-redundancy approach progressively reinstate the intrinsic values of TS for future intervals. Specifically, renovate vanilla Transformer by reorienting information aggregation mechanism from addition subtraction. Then, incorporate an auxiliary output branch into each block original model construct highway leading ultimate prediction. The subsequent...

The performance and parameters of neural networks have a positive correlation, there are lot parameter redundancies in the existing network architectures. By exploring channels relationship whole part network, architectures convolution with tradeoff between obtained. Two implemented by dividing kernels one layer into multiple groups, thus ensuring that has more connections fewer parameters. In these two architectures, information flows from to part, which is called whole-to-part connected...

This study explores whether neural networks can classify multiple samples simultaneously in a forward process. Therefore, multi‐input multi‐prediction network architecture has been proposed. The authors call this method multi‐sample inference (MSIN). In addition to maximising the use of shared parameters, also for training. MSIN allows be randomly combined act as data augmentation, and random combination corresponding labels regularise loss regularisation, which makes have better...

Prognostic and health management is a key issue in the field of machine tool manufacturing. As "teeth" CNC tools, their status directly affects machining efficiency quality products. Accurate monitoring wear can help to avoid product problems caused by failure improve productivity. In this paper, we investigate deep learning-based fault diagnosis approach. First, new data-driven method based on improved multiscale network feature fusion (IMSNet-F) proposed recognize classify condition. It...

Traditional regression and prediction tasks often only provide deterministic point estimates. To estimate the uncertainty or distribution information of response variable, methods such as Bayesian inference, model ensembling, MC Dropout are typically used. These either assume that posterior samples follows a Gaussian process require thousands forward passes for sample generation. We propose novel approach called DistPred forecasting tasks, which overcomes limitations existing while remaining...

In this paper, we find that existing online forecasting methods have the following issues: 1) They do not consider update frequency of streaming data and directly use labels (future signals) to model, leading information leakage. 2) Eliminating leakage can exacerbate concept drift parameter updates damage prediction accuracy. 3) Leaving out a validation set cuts off model's continued learning. 4) Existing GPU devices cannot support learning large-scale data. To address above issues, propose...

The number of categories instances in the real world is normally huge, and each instance may contain multiple labels. To distinguish these massive labels utilizing machine learning, eXtreme Label Classification (XLC) has been established. However, as increases, parameters nonlinear operations classifier also rises. This results a Classifier Computational Overload Problem (CCOP). address this, we propose Multi-Head Encoding (MHE) mechanism, which replaces vanilla with multi-head classifier....

The number of categories instances in the real world is normally huge, and each instance may contain multiple labels.To distinguish these massive labels utilizing machine learning, eXtreme Label Classification (XLC) has been established.However, as increases, parameters nonlinear operations classifier also rises.This results a Classifier Computational Overload Problem (CCOP).To address this, we propose Multi-Head Encoding (MHE) mechanism, which replaces vanilla with multi-head...

As Transformer-based models have achieved impressive performance on various time series tasks, Long-Term Series Forecasting (LTSF) tasks also received extensive attention in recent years. However, due to the inherent computational complexity and long sequences demanding of methods, its application LTSF still has two major issues that need be further investigated: 1) Whether sparse mechanism designed by these methods actually reduce running real devices; 2) extra input guarantee their...

Deep neural networks generally use the information fusion at front and back layers because traditional convolutional that stack have limited ability to extract effective or cannot be passed layer. These networks, which incorporate of previous layer, are attributed improvement accuracy achieved facilitate propagation gradients various techniques for adjusting parameters. We explored relationship between residual dense found they a great degree similarity in certain situations can combine them...

In the modern machine tool manufacturing scene, as milling of CNC tool, health directly affects processing efficiency and product quality. Effective prognostic management is critical. Precise monitoring wear helps to avoid quality problems caused by fault improve production efficiency. Therefore, this paper constructs a diagnosis method based on deep learning. order effectively fuse vibration signals features from different directions spindle, we apply variety Channel Attention (CA)...

The Internet of Things (IoT) specialty belongs to the discipline engineering application, requiring students have strong practical and innovative abilities. Therefore, experimental teaching plays a very important role in cultivation talents quality education. However, traditional system can not keep up with development IoT discipline, has shown many drawbacks. it is urgent conduct research on establish technology course improve students' comprehensive ability.

Personnel scheduling in assembly workshop is a kind of problem widely existing machine tool manufacturing industry. In this paper, according to the actual production needs, minimize total job completion time as optimization objective, mathematical model shop established, and Particle Swarm Optimization algorithm (PSO) discretized. At same time, we discretize particle swarm introduce mutation mechanism Genetic Algorithm (GA) propose Hybrid Discrete (HD-PSO) algorithm. data enterprises,...

Semantic communication aims to extract and trans-mit the semantic information of objects greatly reduce transmission redundant information. This means that termi-nals need deploy large deep models features, which exacerbates computational overhead on terminal. prevents many time-sensitive scenarios, such as industrial IoT vehicular networking, from achieving satisfactory QoS. To alleviate this issue, we propose a Time-Sensitive Communication paradigm (TSSC) fulfills latency requirements...

The mixup training method has achieved a better generalization performance than the traditional method. But there is no interpretation to why such good generalization. In this paper, series of ablation experiments were first done prove that equivalent regularization and data augmentation. Then, we propose several different multi-sample methods as variations mixup, which can also achieve comparable with mixup. This shows mixing same effect original Next, network architecture classify multiple...

Wireless traffic prediction has drawn increasing research interests as it can provide guidance to the network optimization. With predicted information, one preassign resources on demand and perform congestion control adaptively. The efficiency is therefore enhanced. However, wireless in context of mobile scenario, such Internet Vehicles (IoVs), still a challenge issue. nature devices, which dynamically changes topology network, would brings difficulties prediction. This paper focuses deep...