12 October 2018 in IEEE Transactions on Neural Networks and Learning Systems
For many specific applications in data mining and machine learning, we face explicit or latent size constraint for each cluster that leads to the ``balanced clustering'' problem. Many existing clustering algorithms perform well in partitioning but fail in producing balanced clusters and preserving the naturally balanced structure of some data. In this paper, we propose a novel balanced clustering framework that flexibly utilizes local and global information of data. First, we propose the global balanced clustering (GBC), in which a global discriminative partitioning model is combined with the minimization of the distribution entropy of data. Then, we show that the proposed GBC can be further used to globally regularize some widely used local clustering models, so as to transform them into balanced clustering that simultaneously capture local and global data. We apply our global balanced regularization to spectral clustering (SC) and local learning (LL)-based clustering, respectively, and propose another two novel balanced clustering models: the local and global balanced SC (LGB-SC) and LGB-LL. Finding the optimal balanced partition is nondeterministic polynomial-time (NP)-hard in general. We adopt the method of augmented Lagrange multipliers to help optimize our model. Comprehensive experiments on several real world benchmarks demonstrate the advantage of our framework to yield balanced clusters while preserving good clustering quality. Our proposed LGB-SC and LGB-LL also outperform SC and LL as well as other classical clustering methods.
9 June 2025 in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
Persistent postoperative pain (PPSP) is a significant concern in perioperative care, affecting patient quality of life and clinical outcomes. While traditional clinical data may capture patient-reported and objective physiological data in clinical settings, Ecological Momentary Assessment (EMA) collected using smartphones offers dynamic insights into patients' mental and emotional states in their natural environments. However, the self-reported nature of EMA introduces significant variability in both data quality and patient compliance. This variability is often overlooked in studies, where EMA data is treated as contributing equally for all patients. In this paper, we propose an end-to-end approach that addresses this challenge by first quantifying the prediction uncertainty for each data source, including both EMA and clinical data. Next, we integrate the uncertainties into the final decision-making process by assigning more weight to more reliable data sources at the individual level, improving overall predictive performance. Compared to traditional models, our model provides uncertainty estimates for each decision, offering clinicians critical insight into the trustworthiness of predictions. Finally, we conduct both qualitative and quantitative analysis to demonstrate the impact of our uncertainty estimates. Validated through a prospective clinical study involving 782 patients undergoing surgery, our approach achieves superior predictive performance and calibration in PPSP prediction. This work is expected to contribute to personalized perioperative care by effectively integrating traditional clinical and mobile sensing data through an end-to-end uncertainty-aware framework.
17 March 2026 in ACM Transactions on Computing for Healthcare
Wearable devices, which continuously capture activity and physiological data, offer dynamic insights into patient health that complement traditional static clinical predictors. While integrating wearable and clinical data has shown promise for predicting clinical outcomes, the impact of cohort variability on model robustness remains underexplored. In this study, we investigate the impact of cohort variability on predictive model performance in a clinical study focused on predicting pancreatic surgery outcomes using data from Fitbit wristbands and clinical characteristics. This study, initiated before the COVID-19 pandemic and disrupted by surgery delays, highlights substantial variations in patient data before and after the pandemic. Our findings also show that the predictive utility of wearable and clinical features varies across patients. To address these challenges, we propose Adaptive Mixture of Experts (AdaMoE) , a Mixture of Experts model with a diversity regularization that adaptively adjusts the weighting of wearable and clinical features per patient. In a clinical study of 83 pancreatic surgery patients, our approach achieves improved performance compared to existing models and shows promise for handling cohort variability. This work underscores the importance of accounting for cohort variability in predictive modeling and suggests a pathway to enhance model robustness under cohort variability.