Abstract With the rapid proliferation of wireless Internet of Things (IoT) devices, there is a growing concern about the potential misuse of wireless IoT devices for unauthorized sensing and monitoring of daily activities. Individuals are increasingly aware of the risks related to unidentified wireless IoT devices. Therefore, there is an urgent need for a comprehensive approach to improve the transparency of wireless IoT devices. In view of this, we propose a mutual detection and identification approach for autonomously detecting and identifying wireless IoT devices based on encrypted traffic analysis. We develop an adaptive device detection algorithm that uses a mutual detection mechanism among devices to facilitate distributed detection tasks and gather encrypted traffic data. We also design a two-stage device identification method to identify the devices. Through experiments conducted on two publicly available datasets, the study achieved a classification accuracy of $ \mathbf{100\%}$ in distinguishing wireless IoT devices from other device types, as well as a high accuracy of $\mathbf{99.59\%}$ in identifying specific types of wireless IoT devices. These results highlight the efficacy and potential of the proposed method in enhancing the security and transparency of wireless IoT ecosystems.

Aspect Sentiment Quad Prediction (ASQP) is a subtask in the field of sentiment analysis and has become a hot topic in academia. Since labeled data are relatively scarce, data augmentation is usually used to solve this problem. However, existing methods limit the diversity of the augmented samples and cause inconsistencies between the generated sentences and the original labels. In this paper, we propose a method, called Multi-Syntax Template Data Augmentation (MSTDA), which can better utilize syntactic dependency templates to improve the diversity of generated text. Specifically, we use the Label-to-Text (L2T) paradigm to generate augmented samples, which ensures consistency between the label and the text. Then, we take quads with different syntactic templates as input to induce the L2T generator to generate diverse texts. Furthermore, we introduce a fluency filter and a consistency filter to filter out low-quality texts, and perform a redistribution of the augmented samples. Extensive experiments show that MSTDA outperforms other data augmentation methods on two ASQP datasets.

Multimodal sentiment analysis (MSA) aims to determine sentiment polarity in video content. Existing approaches often fuse each modality’s common and specific representations by simply concatenating them, which overlooks modality interactions and may introduce redundant information. In this paper, we propose the Multi-channel Cross-modal Fusion (MCMF) method, which effectively exploits the complementary information in common and specific representations to enhance MSA performance. Specifically, we first extract common and specific features for each modality, using a reconstruction loss to preserve original information during representation learning. In the fusion stage, MCMF integrates information through four parallel channels (i.e., text, audio, visual, and common). Each channel uses cross-modal attention to combine dominant modality representations with others, and the adaptive weight assignment module dynamically adjusts each channel’s contribution. Additionally, we introduce a dynamic sentiment intensity-guided contrastive learning module to obtain high-quality common features. Experimental results on public datasets demonstrate that MCMF significantly outperforms existing methods, particularly in the seven-category classification task.

ABSTRACT Scene memory generation (SMG) refers to training AI agents to recall scene memories similarly to the human brain. This is the key work to realize the artificial memory system. The challenge is to generate scenes rich in motion and keep it realistic while ensuring temporal consistency. Inspired by the principles of memory function in brain neuroscience, this paper proposes a motion‐aware scene generation model named SMG based on motion‐aware temporal style modulation (SMG‐MATSM), which ensures temporal consistency by redesigning the temporal latent representation and constructing a motion matrix to guide the motion of intermediate latent variables. The motion matrix preserves motion consistency in the scene memory through both the cosine similarity and the Mahalanobis distance of intermediate latent variables of adjacent frames. Additionally, SMG‐MATSM uses a style‐based approach and enhances conditional features through the motion matrix during the scene memory synthesis process. Experimental results show that SMG‐MATSM has better effect of action‐enriched scene memory generation, and has varying degrees of efficiency improvement on different datasets with Frechet video distance and Frechet inception distance evaluation metrics.

In processes of refining and chemical productions, alarm systems are generally centralized alarm management systems for process parameters. However, in order to address the challenges of advanced manipulation and maintenance during emergencies, there has been limited research on timely alarming for individual critical process parameters. This paper proposes a method based on the combination of power spectral density and statistical characteristics, which can quickly and accurately diagnose large‐scale trend changes and short‐term nonstationary abnormal trends in process parameters. First, the method employs incremental data from historical records of critical process parameters for volatility analysis. Second, the historical data of critical process parameters are segmented into multiple appropriately sized datasets. We employ a combined analysis of power spectral density and statistical characteristics to extract features from multitude of incremental data. Meanwhile, we have designed a tuning scheme for critical frequencies and their threshold parameters, which can be used for testing and online diagnostics. Experimental validation is performed using actual critical process parameters data from Chinese refineries. The experimental results indicate that the method can detect large‐scale trends and short‐term nonstationary abnormal trends in process parameters, demonstrating good diagnostic performance.

High quality underwater image plays a key role in the underwater equipment inspection of nuclear power plant (NPP). However, when the light propagates in the water, attenuation and scattering occur, that lead to some problems such as low contrast, color distortion and artifact of obtained images owing the particular underwater environment of NPP. In this paper, a U-Net based end-to-end underwater image enhancement network with an inter-connection path and context-aware pyramid mechanism, called APU-NET, is proposed. Owing to poor image pair set for the APU-Net training, underwater imaging model is modified according to the particular environment of NPP, and a GAN based network is constructed to generate realistic underwater images to compose image pairs. Experiments show that the proposed APU-Net model, training with our synthetic underwater image pairs, can enhance underwater images of nuclear power plant effectively. Comparing with those of competitive algorithms, images enhanced with the proposed APU-Net are clearer with fine detail, fewer of color cast.

Data augmentation with mixup has shown to be effective on the NLP tasks. Although its great success, the mixup still has shortcomings. First, vanilla mixup randomly selects one sample to generate the mixup sample for a given sample. It remains unclear how to best choose the input samples for the mixup. Second, linear interpolation limits the space of synthetic data and its regularization effect. In this paper, we propose the dynamic nonlinear mixup with distance-based sample selection, which not only generates multiple sample pairs based on the distance between each sample but also enlarges the space of synthetic samples. Specifically, we compute the distance between each input data by cosine similarity and select multiple samples for a given sample. Then we use the dynamic nonlinear mixup to fuse sample pairs. It does not use a linear, scalar mixing strategy, but a nonlinear interpolation strategy, where the mixing strategy is adaptively updated for the input and label pairs. Experiments on the multiple public datasets demonstrate that dynamic nonlinear mixup outperforms state-of-the-art methods.

Intent detection (ID) and slot filling (SF) are two major tasks for spoken language understanding (SLU). Recent joint learning approaches consider the relationship between intent detection and slot filling, which leverage the shared knowledge across two tasks to benefit each other. However, most existing methods do not make full use of the BERT model and gate mechanisms to improve the semantic correlation between slot filling and intent detection tasks. In this paper, we propose a joint learning model based on BERT, which introduce dual encoder structure and utilizes semantic information by performing feature gate mechanisms in predicting intents and slots. Experimental results demonstrate that our proposed method provides very competitive results on CAIS and DDoST datasets.