Phase sensitive Optical Time Domain Reflectometer (Φ-OTDR), usually addressed as Distributed Acoustic Sensor (DAS), is an emerging IoT-based sensing technology that generates large volumes of data during real-time acquisition. Artificial intelligence-driven models and methods have made accurate and real-time analysis of its disturbance events possible in recent years. However, identifying novel disturbance events still poses a key challenge. This challenge is further exacerbated under semi-supervised conditions due to the scarcity of labeled data, and becomes particularly severe in open-world scenarios, where the model must not only recognize known categories but also actively discover novel ones in the complete absence of supervision for unknown classes. This paper proposes an effective semi-supervised method—ODAS (Open-world Semi-Supervised DAS)—to adaptively recognize new categories of sensing events. The method jointly trains on class probability and embeddings to mitigate confirmation bias, leveraging graph-based contrastive learning with graph neural network-based regularization to explicitly enforce structural consistency among sample embeddings, thereby effectively alleviating representation degeneration and enhancing model robustness under sparse-label conditions. An adaptive synchronized threshold is applied to balance the learning rates of known and novel classes, accompanied by an uncertainty-guided temperature scaling to generate more reliable pseudo-labels and feature embeddings. Experimental results demonstrate that our proposed method achieves over 98% accuracy for known event classes and over 85% for novel event classes, with an overall classification accuracy exceeding 93%. This approach introduces open-world semi-supervised learning to the Φ-OTDR open application environment, overcoming the limitations of traditional closed-world scenarios. It significantly reduces data labeling costs, providing the system with enhanced recognition capabilities when encountering new types of disturbances. Particularly in unlabeled or unknown disturbance environments, this method provides superior robustness and generalization ability to sensing event classification of Φ-OTDR.

In recent years, the positive impact of virtual reality restorative environments on mental health has been confirmed by numerous studies. This study employed OpenAI's Sora model to create three types of virtual videos with restorative effects: natural, animal, and human environments. Using a facial emotion recognition system built with Keras, OpenCV, and PyQt5, along with the fer2013 facial expression database, and the psychological indicator detection technology of the PAD emotion scale, we conducted an experiment with 24 college students to assess their emotional responses and intensity to these three types of virtual videos in immersive virtual reality scenarios. We also collected subjective and objective physiological data from participants after they experienced different restorative virtual videos and analyzed the data using statistical methods.The results indicate that the three types of virtual videos generated by the Sora model had a positive impact on the participants' emotions. Most participants experienced a significant increase in positive emotions such as gentleness and surprise after watching the virtual natural environment videos, while negative emotions like anxiety and unease were effectively alleviated. Further analysis showed that the animal and natural environments were more effective in emotional regulation than the human environment.This study innovatively investigates the application of AI technology, particularly the Sora model, in virtual reality restorative environments, paving new directions for the exploration and application of AI technology in the emotional regulation of college students and providing new insights for future research.

Recent advancements in Large Language Models (LLMs) and Large Multi-modal Models (LMMs) have shown potential in various medical applications, such as Intelligent Medical Diagnosis. Although impressive results have been achieved, we find that existing benchmarks do not reflect the complexity of real medical reports and specialized in-depth reasoning capabilities. In this work, we establish a comprehensive benchmark in the field of medical specialization and introduced RJUA-MedDQA, which contains 2000 real-world Chinese medical report images poses several challenges: comprehensively interpreting imgage content across a wide variety of challenging layouts, possessing the numerical reasoning ability to identify abnormal indicators and demonstrating robust clinical reasoning ability to provide the statement of disease diagnosis, status and advice based on a collection of medical contexts. We carefully design the data generation pipeline and proposed the Efficient Structural Restoration Annotation (ESRA) Method, aimed at restoring textual and tabular content in medical report images. This method substantially enhances annotation efficiency, doubling the productivity of each annotator, and yields a 26.8% improvement in accuracy. We conduct extensive evaluations, including few-shot assessments of 5 LMMs which are capable of solving Chinese medical QA tasks. To further investigate the limitations and potential of current LMMs, we conduct comparative experiments on a set of strong LLMs by using image-text generated by ESRA method. We report the performance of baselines and offer several observations: (1) The overall performance of existing LMMs is still limited; however LMMs more robust to low-quality and diverse-structured images compared to LLMs. (3) Reasoning across context and image content present significant challenges. We hope this benchmark helps the community make progress on these challenging tasks in multi-modal medical document understanding and facilitate its application in healthcare. Our dataset will be publicly available for noncommercial use at https://github.com/Alipay-Med/medDQA_benchmark.git

Rapid growth of modern technologies is bringing dramatically increased e-commerce payments, as well as the explosion in transaction fraud. Many data mining methods have been proposed for fraud detection. Nevertheless, there is always a contradiction that most methods are irrelevant to transaction sequence, yet sequence-related methods usually cannot learn information at single-transaction level well. In this paper, a new “within→between→within” sandwich-structured sequence learning architecture has been proposed by stacking an ensemble model, a deep sequential learning model and another top-layer ensemble classifier in proper order. Moreover, attention mechanism has also been introduced in to further improve performance. Models in this structure have been manifested to be very efficient in scenarios like fraud detection, where the information sequence is made up of vectors with complex interconnected features.