Medical image translation plays a crucial role in assisting clinical diagnosis by enabling cross-modal synthesis (e.g., Computed Tomography to Magnetic Resonance Imaging) and super-resolution, effectively addressing clinical challenges such as radiation exposure, prolonged scan times, and allergic reactions to contrast agents. However, existing approaches primarily focus on developing specialized models for specific tasks, limiting their adaptability across different applications. Developing a general model capable of handling arbitrary medical image translation tasks not only enhances cross-domain generalization but also aligns with the broader trend of artificial intelligence evolving from specialized to general-purpose solutions. Achieving such one-for-all model, however, presents three key challenges: (1) varying task complexity, (2) modality discrepancies, and (3) structural variations across anatomical regions within the same modality. To tackle the first challenge, we utilize an advanced diffusion-based training paradigm to endow the denoising model with extensive pattern coverage capabilities, thereby handling tasks of varying difficulty levels. Subsequently, a General Diffusion Transformer incorporating a Fuzzy Mixture-of-Experts (FMoE) module and an Entropy-guided Attention Soft Prompt (EASP) module is proposed. The FMoE module, equipped with nonlinear modeling capabilities, is designed to address modality discrepancies, while the EASP module is employed to enhance the model's perception of structural variations in images. Extensive qualitative and quantitative experiments demonstrate the effectiveness of the proposed model in the arbitrary medical image translation task.

Zero-shot Sketch-based Image Retrieval (ZS-SBIR) is a challenging yet rewarding task, as it demands models to possess both brain- like zero-shot learning and cross-view alignment capabilities. Recent advances suggest that powerful pre-trained vision encoders, such as CLIP, offer a promising alternative for addressing the ZS-SBIR task. However, the problem of simultaneously evoking the zero-shot learning capability and cross-view alignment capability of pre-trained vision encoders has barely been discussed. To this end, we propose the CrOss-view Attention Prompt (COP) framework, which is composed of an Attention Prompt module and a Cross-view Query module. Specifically, we formulate prompt construction as a retrieval problem by introducing a prompt pool and attention mechanism, thereby constructing attention prompts with fine granularity to enhance the zero-shot learning capability. Furthermore, to endow COP with cross-view alignment capabilities, we replace single-view queries with carefully designed cross-view queries, which can be smoothly inserted into the Attention Prompt module. The proposed COP is scenario-agnostic and supports vision encoders with diverse pre-training schemes. Comprehensive experiments show that COP achieves competitive performance in ZS-SBIR, Generalized ZS-SBIR, and Cross-data ZS-SBIR scenarios, regardless of whether it is based on the ImageNet pre-trained vision encoder or the CLIP pre-trained vision encoder.

The Internet of Underwater Things (IoUT) connects underwater devices for applications like environmental monitoring, marine exploration, and disaster response. Accurate localization of acoustic sources is vital to IoUT, enabling tasks such as sensor deployment and vehicle navigation. The Time Difference of Arrival (TDOA) method, which uses differences in signal arrival times at multiple receivers, is widely employed for this purpose. However, traditional TDOA approaches assume a constant sound speed, overlooking depth-dependent variations caused by changes in salinity, temperature, and pressure. Additionally, multipath propagation complicates localization, as the first received signal may include reflections rather than the direct path. This paper presents a novel synchronization-free localization method, Synchronized-Transmission TDOA (ST-TDOA), tailored for IoUT environments. The proposed method eliminates the need for clock synchronization among receivers, while addressing sound speed variability and mitigating multipath effects. A dynamic model is developed to adaptively adjust for sound speed changes, and a synchronized transmission algorithm ensures accurate TDOA measurements, reducing errors caused by clock discrepancies. Simulation results demonstrate significant improvements in localization accuracy and reliability, highlighting the effectiveness of ST-TDOA in addressing critical challenges in underwater localization for IoUT systems.

Detecting abnormal nodes from attributed networks plays an important role in various applications, including cybersecurity, finance, and social networks. Most existing methods focus on learning different scales of graphs or using augmented data to improve the quality of feature representation. However, the performance is limited due to two critical problems. First, the high sensitivity of attributed networks makes it uncontrollable and uncertain to use conventional methods for data augmentation, leading to limited improvement in representation and generalization capabilities. Second, under the unsupervised paradigm, anomalous nodes mixed in the training data may interfere with the learning of normal patterns and weaken the discrimination ability. In this work, we propose a novel multiview and multiscale contrastive learning framework to address these two issues. Specifically, a network augmentation method based on parameter perturbation is introduced to generate augmented views for both node–node and node–subgraph level contrast branches. Then, cross-view graph contrastive learning is employed to improve the representation without the need for augmented data. We also provide a cycle training strategy where normal samples detected in the former step are collected for an additional training step. In this way, the ability to learn normal patterns is enhanced. Extensive experiments on six benchmark datasets demonstrate that our method outperforms the existing state-of-the-art baselines.

Predicting stock trends is a highly rewarding but high-risk endeavor due to the complex interplay of market dynamics, irrational behaviors, and diverse sentiments. Previous studies have used time-series analysis on historical prices or sentiment analysis on textual information. However, these methods often fail to capture the dynamic interactions between text and time-series modalities and overlook the different perspectives embedded in textual data. To address these limitations, we propose the pleno-alignment framework (PAFrame) that enhances multimodal stock information through intermodal and intramodal alignment to capture market dynamics. Our framework first integrates textual and time-series data in a shared representation space to learn modal-invariant information. To tackle divergent sentiments in textual data, we employ a contrastive learning approach to extract abstract semantic meanings from objective and subjective perspectives, thereby improving the robustness of language representations. Finally, we use a hybrid approach that explicitly combines cross-attention mechanisms to create a unified representation and utilizes prompts to implicitly guide language models with numerical financial indicators for final prediction. Our comprehensive experiments on five real-world datasets show that PAFrame outperforms existing methods in predicting stock trends.

Unsupervised anomaly detection for multivariate time series is challenging due to the diversity of anomalies and the increased scale of dimensions. Recently, reconstruction-based methods have played an important role and made impressive progress. However, these methods can easily suffer from overfitting or lack of discrimination between normal and abnormal samples. In this work, we propose a memory-augmented U-Transformer framework to address these issues. Specifically, we insert down-sampling and up-sampling layers into the Transformer encoder and decoder separately to improve feature representation. The encoder and decoder are then connected by specially designed memory modules at multiple levels. We exploit the power of the memory modules to record prototypical patterns of normal samples and alleviate unexpected generalization to abnormal ones. Extensive experiments on various public benchmarks demonstrate that our method has achieved state-of-the-art performance.

Internet of Things (IoT) systems typically generate large amounts of sensory signals that get involved to represent the states of the systems. Most existing methods focus on learning the temporal patterns of the signals to detect anomalies. However, the performance is limited due to two critical problems. First, the relationships between different signals are rarely considered, resulting in missing important information for representation. Second, the high sensitivity of time series makes it difficult to use conventional methods for data augmentation, which limits the improvement of representation and generalization capabilities. In this work, we propose a novel reconstruction-based framework with contrastive learning from multiple views to address these two issues. Specifically, intrasignal and intersignal graph structures are formed and learned in parallel to model the temporal context and capture the dependency relationships between signals, respectively. Multiview graph contrastive learning strategy is designed to improve the graph representation. We also provide an adaptive data augmentation method to generate graph views for contrastive learning, which helps to accurately capture valuable intrinsic patterns from two different perspectives. Finally, the contrastive learning task and the reconstruction task are jointly trained. Extensive experiments on four real-world data sets demonstrate that our method outperforms the existing state-of-the-art baselines.