The field is witnessing a significant shift towards the development of more sophisticated tensor-based methods and interpretable models, particularly in healthcare and data analysis. Researchers are exploring new ways to capture complex relationships in multivariate time-series data, such as electronic health records (EHRs) and traffic data, using techniques like dynamic graph learning, contrastive augmentation, and tensor nuclear norm. These advancements have led to improved performance in tasks like patient readmission prediction, traffic data imputation, and low-rank tensor completion. Furthermore, there is a growing emphasis on developing models that provide interpretability and transparency, enabling clinicians and practitioners to make more informed decisions. Notable papers in this area include DynaGraph, which proposes an end-to-end interpretable contrastive graph model for multivariate time-series EHRs, and MNT-TNN, which introduces a novel spatiotemporal traffic imputation method using a compact multimode nonlinear transform-based tensor nuclear norm. Other noteworthy papers are TRACE, which presents a Transformer-based model for clinical event nowcasting, and Geometric Median Matching, which proposes a robust k-subset selection strategy for noisy data.
Advances in Tensor-based Methods and Interpretable Models for Healthcare and Data Analysis
Sources
DynaGraph: Interpretable Multi-Label Prediction from EHRs via Dynamic Graph Learning and Contrastive Augmentation
MNT-TNN: Spatiotemporal Traffic Data Imputation via Compact Multimode Nonlinear Transform-based Tensor Nuclear Norm
Prediction of 30-day hospital readmission with clinical notes and EHR information
A Novel Transformed Fibered Rank Approximation with Total Variation Regularization for Tensor Completion
TRACE: Intra-visit Clinical Event Nowcasting via Effective Patient Trajectory Encoding
An extrapolated and provably convergent algorithm for nonlinear matrix decomposition with the ReLU function
Tensor Generalized Approximate Message Passing
Geometric Median Matching for Robust k-Subset Selection from Noisy Data
SplineSketch: Even More Accurate Quantiles with Error Guarantees
Temporal Gaussian Copula For Clinical Multivariate Time Series Data Imputation
Robust Randomized Low-Rank Approximation with Row-Wise Outlier Detection
Adaptive Bivariate Quarklet Tree Approximation via Anisotropic Tensor Quarklets
MENA: Multimodal Epistemic Network Analysis for Visualizing Competencies and Emotions