OBJECTIVE: This study evaluates the integration of electronic health records (EHRs) and natural language processing (NLP) with large language models (LLMs) to enhance healthcare data management and patient care, focusing on using advanced language models to create secure, Health Insurance Portability and Accountability Act-compliant synthetic patient notes for global biomedical research. MATERIALS AND METHODS: The study used de-identified and re-identified versions of the MIMIC III dataset with GPT-3.5, GPT-4, and Mistral 7B to generate synthetic clinical notes. Text generation employed templates and keyword extraction for contextually relevant notes, with One-shot generation for comparison. Privacy was assessed by analyzing protected health information (PHI) occurrence and co-occurrence, while utility was evaluated by training an ICD-9 coder using synthetic notes. Text quality was measured using ROUGE (Recall-Oriented Understudy for Gisting Evaluation) and cosine similarity metrics to compare synthetic notes with source notes for semantic similarity. RESULTS: The analysis of PHI occurrence and text utility via the ICD-9 coding task showed that the keyword-based method had low risk and good performance. One-shot generation exhibited the highest PHI exposure and PHI co-occurrence, particularly in geographic location and date categories. The Normalized One-shot method achieved the highest classification accuracy. Re-identified data consistently outperformed de-identified data. DISCUSSION: Privacy analysis revealed a critical balance between data utility and privacy protection, influencing future data use and sharing. CONCLUSION: This study shows that keyword-based methods can create synthetic clinical notes that protect privacy while retaining data usability, potentially improving clinical data sharing. The use of dummy PHIs to counter privacy attacks may offer better utility and privacy than traditional de-identification.

This investigation compared the emotional impact caused by a crowd of affective virtual humans (VHs) that communicated in the users' native or foreign language. We evaluated the users' affective reactions to a crowd of VHs that exhibited distinct emotional expressions. A total of four emotions were presented, which were Positive, Negative, Neutral, and Mixed. The VHs performed verbal and non-behaviors accordingly. Under the Mixed condition, the VHs were divided into three groups equally and each group was uniquely assigned one of the three emotions (i.e., positive, negative, and neutral). Users collected ten items from a virtual reality market. To complete the tasks, they interacted using natural speech with the emotional VHs. Three language conditions were investigated: one condition in USA and another two conditions in Taiwan. The group of participants in USA interacted with the VHs in English; and the two groups of participants in Taiwan interacted with the VHs using a foreign (English) language and a native (Mandarin) language respectively. We discovered that the medium of communication or language familiarity had a strong influence on participants' emotional reactions. When participants interacted in a foreign language with VHs with a positive emotional disposition, we found their positive emotional reactions were subdued and negative reactions were elevated. However, this was not the case when participants interacted with VHs in their native language, as their emotional reactions were contingent on the emotional disposition of the VHs.

In recent years, edge computing and deep learning have been successfully performed processing and classification tasks in a variety of fields including agriculture. Therefore, this research aims to use unmanned aerial vehicle (UAV) for agriculture applications with integrating edge computing and deep learning techniques. This research experiment was carried out in the NCHU Experimental Farm. The DJI Matrice 100 drone with ASUS Tinker Board S embedded system, which connects a Logitech C925e webcam to capture images are used in this study. The ASUS Tinker Board S runs a folder monitoring program and sends images over the 4G LTE network to the backend server whenever new images are captured and stored. The backend server runs a pre-trained image semantic segmentation model and provides image inference service. The image inference results with the associated segmented image will be sent to the mobile device of the drone controller, and a dynamic flight control action can be triggered. The image semantic segmentation model adopts SegNet network architecture. For a comparison purpose, another network architecture, FCN-AlexNet, was also trained and validated. The preliminary results show SegNet outperformed FCN-AlexNet in image semantic segmentation tasks in terms of the evaluation between training and validation. The average inference speed of the semantic image segmentation model is 0.7s with segmentation identification accuracy is 89%. The promising results shed light on many agriculture applications, such as crop growth condition assessment, fertilizer management, and yield prediction. Additionally, this research provides possible solutions for the labor shortage issue of agriculture which is a common challenge in an aging community like Taiwan and many countries worldwide.

In this contribution we examined the effects on users during interaction with a virtual human crowd in an immersive virtual reality environment. We developed an agent-based crowd model with rich properties including eye gaze, facial expression, body motion, and verbal and non-verbal behaviors. The scenario was a virtual market in which the users needed to gather specific items. In a betweensubjects design, users interacted with a virtual human crowd that showed opposite valenced emotional expressions. There are four conditions in the between-subjects design. These includes different affective virtual crowds namely, Positive, Negative, Neutral, and a Mix condition. The Mix group is defined by a combination of Positive, Negative and Neutral emotional expressive characters. Depending on the specific condition, the virtual humans showed specific verbal and non-verbal behaviors. During the experiment we collected objective measures such as skin electrodermal activity, total time in the simulation, the number of interactions with the agents and performance measures. The subjective measures were the differential emotional survey (DES), and a user experience survey. We reported our findings with an in-depth analysis.

Abstract This paper studies the user collaboration experience with proactive and reactive agents in transporting boxes in virtual environments. Two main characters, the avatar and the agent, are controlled by a user and a controller, respectively. The user and the agent communicate with each other by voice. The agent can be proactive or reactive. The user follows the instruction issued by the proactive agent, whereas the user instructs the reactive agent to perform actions. The goal is to transport boxes to goal positions with orientation constraints. We conducted a user study to analyze the behaviors of participants in several aspects, including task completion time, path length, control experience, and co‐presence experience. We report our findings and make suggestions for future development.