Abstract
We present a robust personalized federated learning framework that leveragesheat-kernel enhanced tensorized multi-view fuzzy c-means clustering withadvanced tensor decomposition techniques. Our approach integrates heat-kernelcoefficients adapted from quantum field theory with Tucker decomposition andcanonical polyadic decomposition (CANDECOMP/PARAFAC) to transform conventionaldistance metrics and efficiently represent high-dimensional multi-viewstructures. The framework employs matriculation and vectorization techniques tofacilitate the discovery of hidden structures and multilinear relationships viaN-way generalized tensors. The proposed method introduces a dual-leveloptimization scheme: local heat-kernel enhanced fuzzy clustering with tensordecomposition operating on order-N input tensors, and federated aggregation oftensor factors with privacy-preserving personalization mechanisms. The localstage employs tensorized kernel Euclidean distance transformations and Tuckerdecomposition to discover client-specific patterns in multi-view tensor data,while the global aggregation process coordinates tensor factors (core tensorsand factor matrices) across clients through differential privacy-preservingprotocols. This tensorized approach enables efficient handling ofhigh-dimensional multi-view data with significant communication savings throughlow-rank tensor approximations.