With the rapid growth of IoT technology, Visual IoT (VIoT) plays a key role in areas like security surveillance and intelligent transportation, where large volumes of sensitive data are collected and transmitted. Many applications face limitations in computing and storage, such as challenges in managing traffic data. Recently, Compressed Sensing (CS) theory has been applied to improve data acquisition and processing efficiency. However, existing CS-based privacy methods focus mainly on protecting single privacy zones through key-based control and lack support for hierarchical protection across multiple sensitive regions. To address this, we propose a multi-level privacy-preserving scheme for VIoT data using CS, offering full, partial, and no access levels to meet varying security needs. We also implement watermarks to prevent key-sharing attacks among partially authorized users. Experimental results demonstrate that the scheme ensures data security while minimizing time and space overhead, making it suitable for resource-limited VIoT scenarios.

The prevalence of generative models in image and video generation has raised extensive concerns about potential harm and misuse. To identify the truthfulness of generated images, most of the existing methods typically apply Fast Fourier Transform (FFT) for frequency extraction. An existing problem is that the frequency-domain representations extracted by FFT are not comprehensive for AI-generated image detection. In this paper, we propose a Multi-perspective Frequency Domain Learning (MFDL) framework, which aims to learn both generalized and discriminative frequency representations via DWT and FFT. Specifically, we design a Frequency Representation Enhancement (FRE) module using the Discrete Wavelet Transform (DWT) and incorporating a multi-granularity enhancement strategy that amplifies all subbands across high frequency to improve discriminability. Additionally, we introduce a Frequency Representation Consistency (FRC) module, which employs complex convolution to capture and preserve forgery patterns in the real and imaginary components derived from FFT. By integrating complementary frequency representations from the DWT and FFT domains obtained through the FRE and FRC modules, MFDL achieves a comprehensive understanding of forgery traces in the frequency domain. This enhances the model’s generalization capability for detecting generated content. Extensive experiments conducted on 32 distinct datasets, covering both GAN-generated and Diffusion-based images, demonstrate the effectiveness of our proposed MFDL framework. These experiments validate the effectiveness of multi-perspective frequency domain learning and show that MFDL outperforms existing detection methods, confirming its strong generalization ability across diverse generative models.

Recent end-to-end image compressive sensing networks primarily use Convolutional Neural Networks (CNNs) and Transformers, each with distinct limitations: CNNs struggle with global feature capture, while Transformers lack local feature extraction. We propose a novel network, DCCT-Net, which combines Dynamic CNN (DCNN) and Transformer. This integration leverages DCNN’s local feature strengths and the Transformer’s global representation capabilities, resulting in superior image reconstruction quality. To further enhance the network’s performance, we propose a Feature Dynamic Augment Module (FDAM), which dynamically extracts features based on the saliency of segmented image regions, thereby amplifying the CNN’s local feature expression. Additionally, we design a Weighted Fusion Module (WFM), which optimizes the combination of local and global features extracted by the DCNN and Transformer, respectively. Extensive experiments demonstrate that our proposed DCCT-Net significantly outperforms most existing state-of-the-art methods in the field.

Multimodal sentiment analysis uses text, visual, and audio data to assess user sentiment, while both the discrimination power of modalities and sample distributions over categories remain imbalanced in practice. To address these challenges, we propose a Multi-grained Perception Model with Perceived Contrastive Focal loss, denoted MGSA1. More specifically, we design a Multi-grained Cross-modal Attention Perception (MCP) module, which employs coarse-grained and fine-grained cross-modal attention to deeply explore the complementary semantics between modalities, thereby modeling sentiment polarity and intensity by fusing text-video and text-audio data, respectively. Modeling sentiment polarity and intensity helps alleviate feature interference between modalities due to their differing discriminative power. Furthermore, the Perceived Contrastive Focal (PCF) loss is designed to address the challenge of unbalanced samples. We enhance the focal loss by incorporating inverse document frequency to dynamically weight samples within each class. Furthermore, information noise contrastive estimation is introduced to replace the class probability predictions in the enhanced focal loss, thereby more effective differentiation between positive and negative samples. Experiments on the MOSI and MOSEI datasets demonstrate that MGSA outperforms all baselines across a range of metrics.

Breast diseases pose a significant threat to women’s health. Automatic lesion segmentation in breast ultrasound images (BUSI) plays a crucial role in fast diagnosis. While various enhanced U-Net-based models have achieved success in multi-scale feature analysis and handling blurred boundaries, two key challenges persist that could guide the improvement of BUSI segmentation networks: 1) significant fluctuations in pixel intensity distribution similarity between the lesion and surrounding tissues, and 2) inconsistent transmission of spatial detail due to multi-scale lesion sampling. These issues highlight the necessity of semantic elasticity understanding and consistency control. To this end, we propose ElaD-Net, an Elastic Semantic Decoupling Network for lesion segmentation in BUSI. This network uses the pre-trained EfficientNet-B2 for multi-scale encoding of BUSI. The decoding stage features two key modules: Elastic Semantic Decoupling (ESD) and Spatial Semantic Reconstruction (SSR). ESD learns and decouples multi-frequency semantics in multi-scale channels with a self-calibration mechanism, enabling dynamic adjustment of receptive depth to resist similarity fluctuations. SSR further optimizes ESD outputs via feature branching, compression, and excitation to ensure spatial semantic consistency, thereby separately reconstructing edge and body.

Malware poses a great threat to information security such as user data, privacy, and assets. Early detection before it has a real impact is the main countermeasure. However, the diversity of carriers and technologies has led to a huge gap between the training scenarios and actual scenarios of detection methods. This makes it difficult for supervision-based detection frameworks to identify new malware variants and complicates threat response. We propose a novel method that integrates frequency domain techniques with feature alignment to enhance variant malware detection, reducing distribution differences between labeled (source) and new (target) samples. By converting malware into grayscale images and applying discrete cosine transform (DCT) for improved feature extraction, followed by feature extraction via a deep residual network from both domains, our model systematically aligns features. This alignment is achieved through a tailored domain adaptation technique involving the minimization of classification and domain alignment losses, which ensures the consistent learning of features across varied domains. Such rigorous alignment not only enhances detection accuracy for both known and variant malware but also supports simultaneous detection across significant distribution differences. We conduct extensive experiments on two real-world datasets to evaluate the performance of various deep learning models under consistent and inconsistent domain distributions. Compared to existing methods, our approach improves accuracy by an average of 1.4% on the BIG2015 dataset, 3.2% on the MDA dataset, 2.75% on the Malimg dataset, and also achieves the best performance on the MaleVis dataset, with similar gains in precision, recall, and F1-score across all datasets.

Anomaly detection plays a vital role as a crucial security measure for edge devices in Artificial Intelligence and Internet of Things (AIoT). With the rapid development of IoT ( Internet of Things), changes in system configurations and the introduction of new devices can lead to significant alterations in device relationships and data flows within the IoT, thereby triggering concept drift. Previously trained anomaly detection models fail to adapt to the changed distribution of streaming data, resulting in a high number of false positive events. This paper aims to address the issue of concept drift in IoT anomaly detection by proposing a comprehensive Concept Drift Detection, Interpretation, and Adaptation framework (CDDIA). We focus on accurately capturing the concept drift of normal data in unsupervised scenarios. To interpret drift samples, we integrate a search optimization algorithm and the SHAP method, providing a comprehensive interpretation of drift samples at both the sample and feature levels. Simultaneously, by utilizing the sample-level interpretation results for filtering new and old samples, we retrain the anomaly detection model to mitigate the impact of concept drift and reduce the false positive rate. This integrated strategy demonstrates significant advantages in maintaining model stability and reliability. The experimental results indicate that our method outperforms five baseline methods in adaptability across three datasets and provides interpretability for samples experiencing concept drift.

Automatic medical report generation is an essential task in applying artificial intelligence to the medical domain, which can lighten the workloads of doctors and promote clinical automation. The state-of-the-art approaches employ Transformer-based encoder-decoder architectures to generate reports for medical images. However, they do not fully explore the relationships between multi-modal medical data, and generate inaccurate and inconsistent reports. To address these issues, this paper proposes a Multi-modal Memory Transformer Network (MMTN) to cope with multi-modal medical data for generating image-report consistent medical reports. On the one hand, MMTN reduces the occurrence of image-report inconsistencies by designing a unique encoder to associate and memorize the relationship between medical images and medical terminologies. On the other hand, MMTN utilizes the cross-modal complementarity of the medical vision and language for the word prediction, which further enhances the accuracy of generating medical reports. Extensive experiments on three real datasets show that MMTN achieves significant effectiveness over state-of-the-art approaches on both automatic metrics and human evaluation.

Dynamic networks, e.g., Dynamic Convolution (DY-Conv) and the Mixture of Experts (MoE), have been extensively explored as they can considerably improve the model's representation power with acceptable computational cost.The common practice in implementing dynamic networks is to convert the given static layers into fully dynamic ones where all parameters are dynamic (at least within a single layer) and vary with the input.However, such a fully dynamic setting may cause redundant parameters and high deployment costs, limiting the applicability of dynamic networks to a broader range of tasks and models.The main contributions of our work are challenging the basic commonsense in dynamic networks and proposing a partially dynamic network, namely PAD-Net, to transform the redundant dynamic parameters into static ones.Also, we further design Iterative Mode Partition to partition dynamic and static parameters efficiently.Our method is comprehensively supported by large-scale experiments with two typical advanced dynamic architectures, i.e., DY-Conv and MoE, on both image classification and GLUE benchmarks.Encouragingly, we surpass the fully dynamic networks by +0.7% top-1 acc with only 30% dynamic parameters for ResNet-50 and +1.9% average score in language understanding with only 50% dynamic parameters for BERT.Code will be released

E-commerce live streaming as an increasingly popular sales model has generated a significant amount of gross merchandise value (GMV) for e-commerce platforms. Live streaming recommendation systems (LSRS) of e-commerce aim to recommend the most appropriate live channels for users to motivate them to buy products. Existing LSRS methods focus only on the user’s interaction behaviors on the live channel (live domain) while ignoring the user’s behaviors and intentions on the e-commerce product (product domain). As a result, the user’s consistent purchase intentions in the cross-domain are not being fully captured, especially when user present differentiated purchase intentions in the cross-domain. How to disentangle user’s consistent intentions and domain-specific intentions in the cross-domain poses a challenge to the LSRS of e-commerce platforms. In this paper, we present a live channel recommendation method, named eLiveRec, developed for Taobao, one of the largest e-commerce platform in the world. Specifically, eLiveRec employs the disentangled encoder module to learn user’s cross-domain consistent intentions and domain-specific intentions. Then, an adaptive multi-task learning framework is developed to jointly optimize the multiple objectives (e.g., stay time, click goods bag, and click products after entering channel) related to live streaming recommendation. In this way, the performance of live streaming recommendation can be further improved and con-form to standard industry RS paradigms. Extensive experiments are conducted on a large-scale industry dataset collected from Taobao Live platform have been performed. Both online and offline experimental results indicate that eLiveRec consistently outperforms existing state-of-the-art baseline methods.

Gait recognition aims to identify individual-specific walking patterns by observing the different periodic movements of each body part. However, most existing methods treat each part equally and fail to account for the data redundancy caused by the different step frequencies and sampling rates of gait sequences. In this study, we propose a multi-granularity motion representation network (GaitMM) for gait sequence learning. In GaitMM, we design a combined full-body and fine-grained sequence learning module (FFSL) to explore part-independent spatio-temporal representations. Moreover, we utilize a frame-wise compression strategy, referred to as multi-scale motion aggregation (MSMA), to capture discriminative information in the gait sequence. Experiments on two public datasets, CASIA-B and OUMVLP, show that our approach reaches state-of-the-art performances.

Numerous electronic health records (EHRs) offer valuable opportunities for understanding patients' health status at different stages, namely health progression. Extracting the health progression patterns allows researchers to perform accurate predictive analysis of patient outcomes. However, most existing works on this task suffer from the following two limitations: 1) the diverse dependencies among heterogeneous medical entities are overlooked, which leads to the one-sided modeling of patients' status and 2) the extraction granularity of patient's health progression patterns is coarse, limiting the model's ability to accurately infer the patient's future status. To address these challenges, a pretrained Health progression network via heterogeneous medical information fusion, HealthNet, is proposed in this article. Specifically, a global heterogeneous graph in HealthNet is built to integrate heterogeneous medical entities and the dependencies among them. In addition, the proposed health progression network is designed to model hierarchical medical event sequences. By this method, the fine-grained health progression patterns of patients' health can be captured. The experimental results on real disease datasets demonstrate that HealthNet outperforms the state-of-the-art models for both diagnosis prediction task and mortality prediction task.

Automatic diagnosis for medical images is a significant research problem of computer-aided diagnosis to reduce the workload of doctors in recent years. However, existing deep learning approaches for diagnoses are usually black-box models with implicit decision-making processes that make them inexplainable. To alleviate the issue, in this paper, we propose a Knowledge-driven Interpretable Network (KdINet) for interpretable disease classification of medical images. KdINet first exploits the pretrained CNN module and the hierarchical representation module to learn two different disease feature representations (i.e., visual disease features and hierarchical disease features). Subsequently, the joint training of two disease features by KdINet’s disease classifier learns a hierarchical classification criterion to infer the diagnosis and generate corresponding interpretable justifications (i.e., ancestral disease paths of the diagnosis). Extensive experiments on three datasets demonstrate that our KdINet achieves significantly higher effectiveness than state-of-the-art approaches on disease classification metrics.

Massive EHR (electronic health records) data contains rich health-related information and offers valuable opportunities for healthcare prediction tasks. Most existing works have achieved great success in improving the accuracy of prediction by leveraging deep learning to model sequential EHR data. However, entangled patient modeling results in inability of quantifying the importance of each type of feature. In this paper, we propose a feature-guided logical perception network FeLON for health risk prediction. FeLON is built on model-level explanation and consists of two main steps. The core part of the first step is a logical perception network to extract logical rules from patient features. In the second step, FeLON builds a fully-connected layer to integrate logical rules information to calculate health risk. Extensive experiments are conducted on two real disease datasets. The experimental results demonstrate that FeLON is able to maintain high prediction accuracy while bringing transparent reasoning process.

Clinical pathway recommendation aims to recommend a set of treatment modalities or treatment procedures for a patient. In order to achieve personalized clinical pathway recommendation, more and more researchers try to mine similar clinical paths from the statistical features of patient-related clinical data (such as electronic health records), while ignoring the high-order interactions between patient-related medical entities, and the time-series change pattern of this high-order interaction relationship, resulting in the inability of existing methods to fully describe patient characteristics and to accurately recommend personalized clinical pathways. To solve the above problems, we propose a new personalized clinical pathway recommendation model TempHRec. To model the complex high-order relationship in clinical pathway, hypergraph technology is introduced to solve the problem that clinical events are correlated at the same time window. On this basis, we propose a temporal hypergraph, to construct a hypergraph for each timestamp with the help of a sliding time window to capture the timing information at the clinical pathway. Extensive experimental results on real-world datasets show that the proposed model achieves the best results compared to baseline methods.

As considerable amounts of POI check-in data have been accumulated, successive point-of-interest (POI) recommendation is increasingly popular. Existing successive POI recommendation methods only predict where user will go next, ignoring when this behavior will occur. In this work, we focus on predicting POIs that will be visited by users in the next 24 hours. As check-in data is very sparse, it is challenging to accurately capture user preferences in temporal patterns. To this end, we propose a category-aware deep model CatDM that incorporates POI category and geographical influence to reduce search space to overcome data sparsity. We design two deep encoders based on LSTM to model the time series data. The first encoder captures user preferences in POI categories, whereas the second exploits user preferences in POIs. Considering clock influence in the second encoder, we divide each user’s check-in history into several different time windows and develop a personalized attention mechanism for each window to facilitate CatDM to exploit temporal patterns. Moreover, to sort the candidate set, we consider four specific dependencies: user-POI, user-category, POI-time and POI-user current preferences. Extensive experiments are conducted on two large real datasets. The experimental results demonstrate that our CatDM outperforms the state-of-the-art models for successive POI recommendation on sparse check-in data.