In this work, we consider the task of classifying binary positive-unlabeled (PU) data. The existing discriminative learning based PU models attempt to seek an optimal reweighting strategy for U data, so that a decent decision boundary can be found. However, given limited P conventional tend suffer from overfitting when adapted very flexible deep neural networks. contrast, are first innovate totally new paradigm attack task, perspective generative by leveraging powerful adversarial networks...

Partial domain adaptation (PDA) for fault identification has been widely researched to help construct self-monitoring systems in the era of Industrial Internet Things (IIoT). However, existing PDA methods neglect influence uncertainty target on performance. To solve this problem, work developed a prototype-guided method with momentum weight diagnosis. Specifically, reduce risk ruling out outlier by output classifier or discriminator, classwise selectively source weighting strategy that...

Recent studies aim to establish contrastive self-supervised learning (CSL) algorithms specialized for the family of Vision Transformers (ViTs) make them function normally as ordinary convolutional-based backbones in training progress. Despite obtaining promising performance on related downstream tasks, one compelling property ViTs is ignored those approaches. As previous have demonstrated, vision transformers benefit from early stage global attention mechanics, feature representations that...

Grammatical Error Correction (GEC) aims to automatically detect and correct grammatical errors. In this aspect, dominant models are trained by one-iteration learning while performing multiple iterations of corrections during inference. Previous studies mainly focus on the data augmentation approach combat exposure bias, which suffers from two drawbacks.First, they simply mix additionally-constructed training instances original ones train models, fails help be explicitly aware procedure...

In this paper, the weight distributions of two classes linear codes based on all known explicit perfect nonlinear functions from <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Fqm</i> to itself are determined using a unified approach. All minimal codewords these characterized according their weights, which suggests that covering structures determined. Finally, access sets secret sharing schemes dual obtained.

Abstract The classification performance of support vector machine (SVM) algorithm is highly dependent on the careful tuning hyper-parameters and penalty coefficient. This paper introduces a novel SVM parameter optimization method by using advanced whale (AWOA) that an improved (WOA) with external archiving strategy. A new framework for based AWOA built. To demonstrate our proposed method, six typical data sets are chosen to evaluate effect problem. Experimental results show higher accuracy...

Tensor network structure search (TN-SS), aiming at searching for suitable tensor (TN) structures in representing high-dimensional problems, largely promotes the efficacy of TN various machine learning applications. Nonetheless, finding a satisfactory using existing algorithms remains challenging. To develop more effective and avoid human labor-intensive development process, we explore knowledge embedded large language models (LLMs) automatic design TN-SS algorithms. Our approach, dubbed...

<title>Abstract</title> Accurate short-term load forecasts play an important role in guiding and regulating the operations of electric utilities.Using long- memory neural networks with improved whale optimization technique, this study suggests a combination strategy for network prediction.The long-short-term mitigates issue gradient vanishing explosion, caused by cumulative multiplication activation function RNN when handling lengthy sequences.In order to address model parameter randomness,...

Semi-supervised learning (SSL) has been proven to be a powerful method for leveraging unlabelled data alleviate models' dependence on large labelled datasets. The common framework among recent approaches is train the model amount of with consistency regularization constrain predictions invariant input perturbation. However, existing SSL frameworks still have room improvement in method. Instead regularizing category label space as frameworks, this paper proposes feature renormalization (FSR)...

Contrastive Self-supervised Learning (CSL) is a practical solution that learns meaningful visual representations from massive data in an unsupervised approach. The ordinary CSL embeds the features extracted neural networks onto specific topological structures. During training progress, contrastive loss draws different views of same input together while pushing embeddings inputs apart. One drawbacks term requires large number negative samples to provide better mutual information bound...

Under variable working conditions, the tool status signal is affected by changing machine processing parameters, resulting in a decreased prediction accuracy of remaining useful life (RUL). Aiming at this problem, method based on multi-sensor fusion for RUL was proposed. Firstly, factorization (FM) used to extract nonlinear features low-frequency condition signal, and one-dimensional separable convolution applied state from multi-channel high-frequency sensor signals. Secondly, residual...

Contrastive self-supervised learning (CSL) with a prototypical regularization has been introduced in meaningful representations for downstream tasks that require strong semantic information. However, to optimize CSL loss performs the aggressively, e.g., ProtoNCE loss, might cause "coagulation" of examples embedding space. That is, intra-prototype diversity samples collapses trivial solutions their prototype being well-separated from others. Motivated by previous works, we propose mitigate...

Grammatical Error Correction (GEC) aims to automatically detect and correct grammatical errors. In this aspect, dominant models are trained by one-iteration learning while performing multiple iterations of corrections during inference. Previous studies mainly focus on the data augmentation approach combat exposure bias, which suffers from two drawbacks. First, they simply mix additionally-constructed training instances original ones train models, fails help be explicitly aware procedure...

Aiming at the problem of low recall and precision threshold-based disk failure prediction methods, this paper studies through machine learning methods. By labeling examples in data set balancing number with different labels, after selecting features that have a greater impact on results, method random forest is used to train model, performance model verified. When predicts disk, are both higher than those method.