Haoxiang Yu

Mr. Haoxiang Yu

orcid

Research Focus & Metrics

Academic Career Length

5Years (2022-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
6 March 2024 in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
Verified
Journal Article
Open access
EKEvan King
Evan King
HYHaoxiang Yu
Haoxiang Yu
SLSangsu Lee
Sangsu Lee
CJChristine Julien
Christine Julien
Smart home assistants function best when user commands are direct and well-specified---e.g., "turn on the kitchen light"---or when a hard-coded routine specifies the response. In more natural communication, however, human speech is unconstrained, often describing goals (e.g., "make it cozy in here" or "help me save energy") rather than indicating specific target devices and actions to take on those devices. Current systems fail to understand these under-specified commands since they cannot reason about devices and settings as they relate to human situations. We introduce large language models (LLMs) to this problem space, exploring their use for controlling devices and creating automation routines in response to under-specified user commands in smart homes. We empirically study the baseline quality and failure modes of LLM-created action plans with a survey of age-diverse users. We find that LLMs can reason creatively to achieve challenging goals, but they experience patterns of failure that diminish their usefulness. We address these gaps with Sasha, a smarter smart home assistant. Sasha responds to loosely-constrained commands like "make it cozy" or "help me sleep better" by executing plans to achieve user goals---e.g., setting a mood with available devices, or devising automation routines. We implement and evaluate Sasha in a hands-on user study, showing the capabilities and limitations of LLM-driven smart homes when faced with unconstrained user-generated scenarios.

10 February 2025 in Journal of Medical Internet Research
Verified
Journal Article
Open access
NCNamuun Clifford
Namuun Clifford
RTRachel Tunis
Rachel Tunis
AAAdetimilehin Ariyo
Adetimilehin Ariyo
HYHaoxiang Yu
Haoxiang Yu
HRHyekyun Rhee
Hyekyun Rhee
KRKavita Radhakrishnan
Kavita Radhakrishnan
Background: Hypertension (HTN) is the leading cause of cardiovascular disease morbidity and mortality worldwide. Despite effective treatments, most people with HTN do not have their blood pressure under control. Precision health strategies emphasizing predictive, preventive, and personalized care through digital tools offer notable opportunities to optimize the management of HTN. Objective: This scoping review aimed to fill a research gap in understanding the current state of precision health research using digital tools for the management of HTN in adults. Methods: This study used a scoping review framework to systematically search for articles in 5 databases published between 2013 and 2023. The included articles were thematically analyzed based on their precision health focus: personalized interventions, prediction models, and phenotyping. Data were extracted and summarized for study and sample characteristics, precision health focus, digital health technology, disciplines involved, and characteristics of personalized interventions. Results: After screening 883 articles, 46 were included; most studies had a precision health focus on personalized digital interventions (34/46, 74%), followed by prediction models (8/46, 17%) and phenotyping (4/46, 9%). Most studies (38/46, 82%) were conducted in or used data from North America or Europe, and 63% (29/46) of the studies came exclusively from the medical and health sciences, with 33% (15/46) of studies involving 2 or more disciplines. The most commonly used digital technologies were mobile phones (33/46, 72%), blood pressure monitors (18/46, 39%), and machine learning algorithms (11/46, 24%). In total, 45% (21/46) of the studies either did not report race or ethnicity data (14/46, 30%) or partially reported this information (7/46, 15%). For personalized intervention studies, nearly half (14/30, 47%) used 2 or less types of data for personalization, with only 7% (2/30) of the studies using social determinants of health data and no studies using physical environment or digital literacy data. Personalization characteristics of studies varied, with 43% (13/30) of studies using fully automated personalization approaches, 33% (10/30) using human-driven personalization, and 23% (7/30) using a hybrid approach. Conclusions: This scoping review provides a comprehensive mapping of the literature on the current trends and gaps in digital precision health research for the management of HTN in adults. Personalized digital interventions were the primary focus of most studies; however, the review highlighted the need for more precise definitions of personalization and the integration of more diverse data sources to improve the tailoring of interventions and promotion of health equity. In addition, there were significant gaps in the reporting of race and ethnicity data of participants, underuse of wearable devices for passive data collection, and the need for greater interdisciplinary collaboration to advance precision health research in digital HTN management. Trial Registration: OSF Registries osf.io/yuzf8; https://osf.io/yuzf8

14 July 2022 in IEEE Transactions on Artificial Intelligence
Verified
Journal Article
Closed access
HYHaoxiang Yu
Haoxiang Yu
VRVaskar Raychoudhury
Vaskar Raychoudhury
SSSnehanshu Saha
Snehanshu Saha
JEJanick Edinger
Janick Edinger
Roger O. SmithRoger O. Smith
Roger O. Smith
OGOsman Gani
Osman Gani
Finding barrier-free accessible path through the built environment is necessary for wheelchair users. Researchers have identified the effect of surface vibration on the health of wheelchair users and proposed various solutions to identify and annotate mobility barriers. However, the effects of surface-induced vibration on accessibility is yet to be investigated. In order to address this problem, we present the WheelShare system which uses machine learning to identify surfaces and curbs from vibration captured by accelerometer and gyroscope sensors (both smartphone and wearable device). WheelShare allows users to share their collected data via crowd-sourcing. Based on this surface-specific vibration we recognize 32 different surfaces found across the built environment (in Austria, China, France, Germany, India, and the USA for 32 human subjects) with an accuracy of 97.5%. The recognized surfaces are then grouped into different categories based on their characteristics. Our novel contributions in this paper include the following. (1) We have developed a machine learning enabled surface recognition system which uses sensor data to capture surface-induced vibration patterns. (2) We have tested the robustness of the system for devices with only accelerometer data (such as low-end smartphones) and achieved an accuracy of 91.6%. (3) We have deployed our prototype WheelShare system for real-time surface classification in new sites that are unseen in training and testing, and successfully recognized the surface characteristics.

11 September 2022
Verified
Conference Paper
Open access
MRMrittika Raychoudhury
Mrittika Raychoudhury
HYHaoxiang Yu
Haoxiang Yu
JKJames Kiper
James Kiper
Falls are the biggest threat among all events to elderly people and patients and pose a serious risk to their well-being, confidence, and mortality. Visual impairments which distort depth perceptions are often identified as major reasons for falls. Single and/or multi-focal length glasses used to correct myopia (blurriness of distance vision) and presbyopia (blurriness of near vision) cause distorted depth perceptions if not used properly. Users, especially older adults, need to change between glasses while transitioning between sedentary to mobile activities and vice versa. Forgetting to change into proper glasses may lead to fall. In this paper, we plan to develop a novel system called Activisee which uses sensor-augmented glasses to automatically detect activity transitions and alert users to change into appropriate glasses whenever necessary. Although, wireless sensor systems have long been used for fall detection and prevention through human activity recognition, using sensor-augmented glasses for activity transition detection has never been studied. Results using data collected from 23 human users performing a set of five distinct activities show that Activisee can successfully detect transitions from sedentary to mobile activities with up to 92% accuracy.

24 August 2024 in SN Computer Science
Verified
Journal Article
Closed access
PDF file
VMVihaan Misra
Vihaan Misra
SMShivashankar S. Menon
Shivashankar S. Menon
SSSnehanshu Saha
Snehanshu Saha
AMArchana Mathur
Archana Mathur
HYHaoxiang Yu
Haoxiang Yu
VRVaskar Raychoudhury
Vaskar Raychoudhury