Retrieval Augmented Generation (RAG) with Knowledge Graphs (KGs) is an effective way to enhance Large Language Models (LLMs). Due to the natural discrepancy between structured KGs and sequential LLMs, KGs must be linearized to text before being inputted into LLMs, leading to the problem of KG Alignment with LLMs (KGA). However, recent KG+RAG methods only consider KGA as a simple step without comprehensive and in-depth explorations, leaving three essential problems unclear: (1) What are the factors and their effects in KGA? (2) How do LLMs understand KGs? (3) How to improve KG+RAG by KGA? To fill this gap, we conduct systematic explorations on KGA, where we first define the problem of KGA and subdivide it into the graph transformation phase (graph-to-graph) and the linearization phase (graph-to-text). In the graph transformation phase, we study graph features at the node, edge, and full graph levels from low to high granularity. In the linearization phase, we study factors on formats, orders, and templates from structural to token levels. We conduct substantial experiments on 15 typical LLMs and three common datasets. Our main findings include: (1) The centrality of the KG affects the final generation; formats have the greatest impact on KGA; orders are model-dependent, without an optimal order adapting for all models; the templates with special token separators are better. (2) LLMs understand KGs by a unique mechanism, different from processing natural sentences, and separators play an important role. (3) We achieved 7.3% average performance improvements on four common LLMs on the KGQA task by combining the optimal factors to enhance KGA.

This paper illustrates the processes and results of user acceptance and satisfaction tests for the voice-activated self-monitoring (VoiS) application conducted in the field. VoiS was designed and developed for individuals with diabetes (DM) and hypertension (HTN) to support their routine and convenient self-management using a smart speaker platform. VoiS is also accessible on users' mobile devices to visualize user-generated data. A total of nine adults with DM and HTN participated and were asked to use the VoiS system at home for a week to identify its acceptability in real-world conditions. Participants completed phone interviews to report operational errors and a structured survey to assess the acceptability of VoiS. The results showed that participants agreed or strongly agreed that VoiS was easy to use and useful, and they intended to continue using it. They were highly satisfied with VoiS. They also reported operational errors and challenges including device-related technical issues and the smart reminder feature of VoiS. Overall, participants perceived VoiS as an easy and useful tool for managing their conditions and felt motivated to monitor their biomarkers routinely.

Fast response and safety operation are essential requirements for the tracking control of robotic manipulators. In this paper, a unified predefined-time self-organizing interval type-2 fuzzy neural network control (SOIT2FNNC) framework is presented for robotic manipulators subject to actuator failures and uncertainties. Such a framework operates in a parallel structure where the model-free predefined-time controller guarantees the transient performance while the proposed network controller provides appropriate torques to handle failures and uncertainties, which leads to a solution for both normal and faulty conditions. Significant features of this study are that the control design does not depend on any information about system dynamics, and theoretically, the predefined-time convergence is accomplished by means of the online parameter learning algorithm. Moreover, a hierarchical self-organizing algorithm is embedded in the proposed network controller to overcome the network structure complexity and the input partition problem. Both numerical simulation and experiment results utilizing artificial faults are implemented to demonstrate the superiority of the proposed control scheme.

This study suggests a way to utilize the existing medical ontology and natural language processing techniques to extract major medical concepts from lay vocabularies of health consumers on social media and group them based on the defined semantic types in the ontology. Diabetes-related discussions on Tumblr was used to test the efficiency of SpaCy and the Markov-Viterbi algorithm to map lay medical terms to the defined medical concepts in the UMLS. The system discussed in this paper can better analyze free texts, take care of word ambiguity and extract the lifestyle indicators from the daily life discussions of diabetic people on Tumblr. The findings of this study can contribute to developing health applications that track the health behavior of those living with chronic conditions such as diabetes. This approach can also assist researchers who are interested in processing lay languages used by health consumers to foster an understanding of their health behavior.

Partial-label learning (PLL) is an important weakly supervised learning problem, which allows each training example to have a candidate label set instead of a single ground-truth label. Identification-based methods have been widely explored to tackle label ambiguity issues in PLL, which regard the true label as a latent variable to be identified. However, identifying the true labels accurately and completely remains challenging, causing noise in pseudo labels during model training. In this paper, we propose a new method called CroSel, which leverages historical predictions from the model to identify true labels for most training examples. First, we introduce a cross selection strategy, which enables two deep models to select true labels of partially labeled data for each other. Besides, we propose a novel consistency regularization term called comix to avoid sample waste and tiny noise caused by false selection. In this way, CroSel can pick out the true labels of most examples with high precision. Extensive experiments demonstrate the superiority of CroSel, which consistently outperforms previous state-of-the-art methods on benchmark datasets. Additionally, our method achieves over 90% accuracy and quantity for selecting true labels on CIFAR-type datasets under various settings.

Rehabilitation and disability researchers are increasingly considering utlizing machine learning (ML) algorithms to enhance accessibility for people with disabilities (PwD) as they interact with their environments. PwD often experience environmental barriers in the community and private buildings due to a lack of accessible infrastructure design and prior accessibility information. Such environmental barriers may inhibit PwD from full participation and impede in one’s overall independence and quality of life. The availability of healthcare services and information are essential for increased participation in occupations and optimal independence. In the current era of connected health, information has the ability to be accessed anywhere and providing accessibility content for PwD to use can enhance occupational performance factors. The purpose of this study was to 1) identify existing accessibility measurement challenges and barriers and 2) propose an intelligent accessibility evaluation and assessment for buildings, and 3) leverage mobile applications to address major challenges of accessibility measurement. This research aims to improve the accuracy and reliability of the accessibility measurement using a smarter system.

Conversational agents have gained their ground in our daily life and various domains including healthcare. Chronic condition self-management is one of the promising healthcare areas in which conversational agents demonstrate significant potential to contribute to alleviating healthcare burdens from chronic conditions. This survey paper introduces and outlines types of conversational agents, their generic architecture and workflow, the implemented technologies, and their application to chronic condition self-management.

Heart Rate (HR) and Heart Rate Variability (HRV) is an essential measurement to know the heart’s cardiovascular condition. Many works have been done for measuring HR-HRV based on the facial video non-invasively. In this paper, based on our previous work experience of measuring HR-HRV by Remote photoplethysmography signals (rPPG) analysis, we have built a prediction model from the 10-second time series data extracted from a facial video. In this work, we have used the instantaneous public dataset with several data models to predict the HR-HRV, and stress levels exclusively from the dataset. We have used here some of the popular algorithms appropriate for this task. We have also analyzed the stress level classification on the gender of a subject using the same facial videos with 16 different classifiers resulting in almost perfect accuracy for several classifiers.

Accurate and valid health information is crucial for effective medical management. Failure to collect adequate information from physical and mental health examinations can be a barrier to Virtual medical platforms and telemedicine. In this paper, we propose the non-invasive “Dream” project prototype to monitor and record heart rate (HR), heart rate-variation (HRV) (for physical health), and stress (for mental health) using only a smart-phone. This non-invasive mobile application, “Dream” uses the front camera to capture video to calculate HR-HRV and stress. The full “Dream” project encompasses our previous facial video HR-HRV and stress work. We have also compared our proposed “Dream” project with 39 works in this area. We found a significant positive difference between our proposed “Dream” project compared to other projects in respect to user accessibility, application, cost-effectiveness, hospitalization monitoring, and human health status. Furthermore, we can apply “Dream” in remote human health monitoring, driver monitoring, and creating vital sign records non-invasively without a health care assistant. Especially during a pandemic, this virtual health monitoring system can be useful for scaled-up telemedicine to serve the remote population.

The integration of motivational strategies and self-management theory with mHealth tools is a promising approach to changing the behavior of patients with chronic disease. In this manuscript, we describe the development and current architecture of a prototype voice-activated self-monitoring application (VoiS) which is based on these theories. Unlike prior mHealth applications which require textual input, VoiS app relies on the more convenient and adaptable approach of asking users to verbally input markers of diabetes and hypertension control through a smart speaker. The VoiS app can provide real-time feedback based on these markers; thus, it has the potential to serve as a remote, regular, source of feedback to support behavior change. To enhance the usability and acceptability of the VoiS application, we will ask a diverse group of patients to use it in real-world settings and provide feedback on their experience. We will use this feedback to optimize tool performance, so that it can provide patients with an improved understanding of their chronic conditions. The VoiS app can also facilitate remote sharing of chronic disease control with healthcare providers, which can improve clinical efficacy and reduce the urgency and frequency of clinical care encounters. Because the VoiS app will be configured for use with multiple platforms, it will be more robust than existing systems with respect to user accessibility and acceptability.

Retinal eye disease is the most common reason for visual deterioration. Long-term management and follow-up are critical to detect the changes in symptoms. Optical Coherence Tomography (OCT) is a non-invasive diagnostic tool for diagnosing and managing various retinal eye diseases. With the increasing desire for OCT image, the clinicians are suffering from the burden of time on the diagnostic and the treatment. In this study, an auto-grading diagnostic tool is proposed to divide the OCT image for the retinal disease classification. In this tool, the classification model implements convolutional neural networks (CNNs), and the model training is based on denoised OCT images. The tool can detect the uploaded OCT image and automatically generate a result of classification in the categories of Choroidal neovascularization (CNV), Diabetic macular edema (DME), multiple drusen, and Normal. The system will definitely improve the performance of retinal eye disease diagnosis and alleviate the burden on the medical system.