The widespread deployment of smart meters has created significant opportunities for applying artificial intelligence technologies to power system tasks. However, the high cost of data annotation limits the effectiveness of traditional supervised learning in this domain, making self-supervised learning an attractive alternative. In this article, we propose PowerDiffuser, a novel self-supervised learning strategy tailored for power load signals. By leveraging a diffusion model framework, PowerDiffuser integrates two mainstream self-supervised paradigms, namely contrastive learning and reconstruction-based learning, which enables the model to effectively capture both periodic patterns and local features. To address the overfitting issues commonly observed in generic time-series feature extractors when applied to power load tasks, we design two modular spatiotemporal feature extractors specifically engineered to handle samples with varying complexity levels. In addition, we adapt the involution operator to better align with the unique characteristics of power load signals. Extensive experiments on the ISMCBT, ETTh and REDD datasets demonstrate that PowerDiffuser consistently outperforms both time-series general models and existing self-supervised learning strategies across diverse downstream power load tasks. Ablation studies further validate the contributions of the proposed modules and highlight the effectiveness of transforming 1-D load signals into 2-D periodicity-based representations as a preprocessing step.

Driver fatigue is a critical factor that lead to traffic accidents with a high fatality rate. Electroencephalogram (EEG) is one of the most reliable indicators to objectively assess fatigue status, but recognizing fatigue driving status from it is still an essential and challenging problem. In this paper, we propose a multiscale global prompt Transformer (MsGPT) deep learning model, which can automatically recognize driver fatigue end-to-end. First, we construct an intra-inter-scale cascade framework based on Transformer with a multiscale convolutional patch embedding (MC-PatchEmbed), and guide global-local feature interaction by adding a global prompt token throughout. Second, to efficiently integrate intra-scale and inter-scale feature information, we design a mixed token by aggregating the output from the intra-scale, which includes rich low-level feature information for multiscale. Moreover, a novel learnable query is introduced into multi-head self-attention (MSA) to reduce the computational complexity to linear level. Experiments are conducted on the SEED-VIG dataset and the SADT dataset with both intra-subject and inter-subject settings to evaluate the performance of MsGPT, and the results show that MsGPT greatly outperforms various methods in terms of the classification evaluation metrics of EEG-based fatigue driving.Note to Practitioners—This paper considers the use of raw EEG data to recognize the driver fatigue state. Existing methods mainly rely on manually extracted EEG features and convolutional neural network (CNN) based inference. However, the large intra-individual and inter-individual differences greatly limit the extraction of EEG fatigue features. This paper suggests a multiscale global prompt Transformer (MsGPT) deep learning model. This model leverages a shared weighting mechanism to construct an inter-to intra-scale multiscale framework that can capture refined fatigue features not achievable at a single scale, we incorporate a new Transformer of the global prompt mechanism, which facilitates multiscale local-to-global fusion of long-term physiological signals. Our experiments demonstrate the superiority of our method on two datasets with intra-subject and inter-subject settings. The proposed method can be readily deployed in the automatic driving assistance system to alert drivers to avoid or reduce traffic accidents caused by excessive fatigue.

Zero-shot learning casts light on lacking unseen class data by transferring knowledge from seen classes via a joint semantic space. However, the distributions of samples from seen and unseen classes are usually imbalanced. Many zero-shot learning methods fail to obtain satisfactory results in the generalized zero-shot learning task, where seen and unseen classes are all used for the test. Also, irregular structures of some classes may result in inappropriate mapping from visual features space to semantic attribute space. A novel generative mixup networks with semantic graph alignment is proposed in this article to mitigate such problems. To be specific, our model first attempts to synthesize samples conditioned with class-level semantic information as the prototype to recover the class-based feature distribution from the given semantic description. Second, the proposed model explores a mixup mechanism to augment training samples and improve the generalization ability of the model. Third, triplet gradient matching loss is developed to guarantee the class invariance to be more continuous in the latent space, and it can help the discriminator distinguish the real and fake samples. Finally, a similarity graph is constructed from semantic attributes to capture the intrinsic correlations and guides the feature generation process. Extensive experiments conducted on several zero-shot learning benchmarks from different tasks prove that the proposed model can achieve superior performance over the state-of-the-art generalized zero-shot learning.

Unsupervised domain adaptation aims to learn a classification model for the target domain without any labeled samples by transferring the knowledge from the source domain with sufficient labeled samples. The source and the target domains usually share the same label space but are with different data distributions. In this paper, we consider a more difficult but insufficient-explored problem named as few-shot domain adaptation, where a classifier should generalize well to the target domain given only a small number of examples in the source domain. In such a problem, we recast the link between the source and target samples by a mixup optimal transport model. The mixup mechanism is integrated into optimal transport to perform the few-shot adaptation by learning the cross-domain alignment matrix and domain-invariant classifier simultaneously to augment the source distribution and align the two probability distributions. Moreover, spectral shrinkage regularization is deployed to improve the transferability and discriminability of the mixup optimal transport model by utilizing all singular eigenvectors. Experiments conducted on several domain adaptation tasks demonstrate the effectiveness of our proposed model dealing with the few-shot domain adaptation problem compared with state-of-the-art methods.

Convolutional neural networks (CNNs) are widely used in the field of remote sensing images. However, the applications of CNNs and related techniques often ignore the properties of remote sensing data. In our study, we focus on the hyperspectral image (HSI) classification problem, and address the issue of including the very rich spectral information present in HSIs in CNN-based models to produce highly accurate classification results. We propose a two-step classification technique, ClusterCNN. The first step divides HSI pixels into different clusters, to form a material map which can be considered as a compressed expression of the original spectral features. The second step trains a CNN that can extract spatial features from the material map, and then exploits these spatial features to classify HSI pixels. The proposed approach follows a strict hierarchy to exploit both the spectral and spatial features in HSIs. Experimental results show the effectiveness of ClusterCNN as compared to the much more complicated state-of-the-art approaches.

Zero-shot learning has received great interest in visual recognition community. It aims to classify new unobserved classes based on the model learned from observed classes. Most zero-shot learning methods require pre-provided semantic attributes as the mid-level information to discover the intrinsic relationship between observed and unobserved categories. However, it is impractical to annotate the enriched label information of the observed objects in real-world applications, which would extremely hurt the performance of zero-shot learning with limited labeled seen data. To overcome this obstacle, we develop a Low-rank Semantics Grouping (LSG) method for zero-shot learning in a semi-supervised fashion, which attempts to jointly uncover the intrinsic relationship across visual and semantic information and recover the missing label information from seen classes. Specifically, the visual-semantic encoder is utilized as projection model, low-rank semantic grouping scheme is explored to capture the intrinsic attributes correlations and a Laplacian graph is constructed from the visual features to guide the label propagation from labeled instances to unlabeled ones. Experiments have been conducted on several standard zero-shot learning benchmarks, which demonstrate the efficiency of the proposed method by comparing with state-of-the-art methods. Our model is robust to different levels of missing label settings. Also visualized results prove that the LSG can distinguish the test unseen classes more discriminative.

In this paper, we propose a new approach to establish a landslide displacement forecasting model based on artificial neural networks (ANNs) with random hidden weights. To quantify the uncertainty associated with the predictions, a framework for probabilistic forecasting of landslide displacement is developed. The aim of this paper is to construct prediction intervals (PIs) instead of deterministic forecasting. A lower-upper bound estimation (LUBE) method is adopted to construct ANN-based PIs, while a new single hidden layer feedforward ANN with random hidden weights for LUBE is proposed. Unlike the original implementation of LUBE, the input weights and hidden biases of the ANN are randomly chosen, and only the output weights need to be adjusted. Combining particle swarm optimization (PSO) and gravitational search algorithm (GSA), a hybrid evolutionary algorithm, PSOGSA, is utilized to optimize the output weights. Furthermore, a new ANN objective function, which combines a modified combinational coverage width-based criterion with one-norm regularization, is proposed. Two benchmark data sets and two real-world landslide data sets are presented to illustrate the capability and merit of our method. Experimental results reveal that the proposed method can construct high-quality PIs.

Evaluation of uncertainties associated with landslide displacement prediction is essential for improving the reliability of landslide early warning systems. An efficient probabilistic forecasting method for the construction of prediction intervals (PIs) using bootstrap and kernel-based extreme learning machine (ELM) is proposed. To overcome the drawbacks of artificial neural networks (ANNs) in predicting mutational displacement points with time lags, this paper proposes an ANNs switched prediction scheme to construct PIs with a three-stage formulation. In the first stage, K-means clustering is applied to divide the whole training dataset into two sub-training sets: the stationary points and the mutational points. In the second stage, a weighted ELM classifier is applied to construct the switched rules. In the third stage, bootstrap- and kernel-based ELMs are applied to construct candidate PIs for each sub-training set. The final PIs are constructed by switching between these two candidate PIs. The effectiveness of the proposed ANNs switched prediction method has been validated through comprehensive tests using three real-world landslide datasets from the Three Gorges region of China.

An efficient and accurate method for landslide displacement prediction is very important to reduce the casualties and property losses caused by this type of natural hazard. In recent years, many kinds of artificial neural networks (ANNs) have been widely applied to landslide displacement prediction. But we can't know which type of ANN is the best until we have calculated the prediction error. An improper choice of ANN may result in bad prediction results. In this paper, we use a neural networks combination prediction method based on the discounted MSFE (mean squared forecast error) to reduce the risk of selecting the types of ANNs. Four popular ANNs, radial basis function neural network (RBFNN), support vector regression (SVR), least squares support vector machine (LSSVM) and extreme learning machine (ELM), are selected as candidate neural networks. The performance of our model is verified through two case studies in Baishuihe landslide and Bazimen landslide. Experimental results reveal that the combining neural networks can improve the generalization abilities of ANNs.