FedHK-MVFC: Federated Heat Kernel Multi-View Clustering

Kristina P. Sinaga · 2025

Read FedHK-MVFC: Federated Heat Kernel Multi-View Clustering by Kristina P. Sinaga (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-15844.txt.

Introduction

Kristina Sinaga, a lone author, groups medical patients using several views of their data, without pooling records. Hospitals hold rich data, but privacy rules block central study. Her frame joins a heat kernel method, borrowed from physics, with federated learning across sites. Distances between patients are turned into geometry aware similarity scores.

Two algorithms result: a central one for single site study, and a federated one for cross hospital work. The federated version adds differential privacy and secure pooling, aimed at health law compliance. Tests use fake heart patient records for ten thousand people across two hospitals. Grouping accuracy rose, messages between sites fell, and speed held up against central methods. The aim is shared patient typing with privacy kept.

Catalog record

CREATORKristina P. Sinaga
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE127 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-15844

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

FedHK-MVFC: Federated Heat Kernel Multi-View Clustering Kristina P. Sinaga Abstract In the realm of distributed AI and privacy-focused medical applications, we propose a framework for multi-view clustering that links quantum field theory with federated healthcare analytics. Our method uses heat-kernel coefficients from spectral analysis to convert Euclidean distances into geometry-aware similarity measures, capturing the structure of diverse medical data. We lay this out through the Heat Kernel Distance (HKD) transformation with convergence guarantees. Two algorithms are developed: Heat Kernel-Enhanced Multi-View Fuzzy Clustering (HK-MVFC) for central analysis, and Federated Heat Kernel Multi-View Fuzzy Clustering (FedHK-MVFC) for secure, privacy-preserving learning across hospitals using differential privacy and secure aggregation to facilitate HIPAA-compliant collaboration. Tests on synthetic datasets of cardiovascular patients show an 8-12% increase in clustering accuracy, 70% reduced communication, and 98.2% efficiency retention over centralized methods. Validated on 10,000 patient records across two hospitals, it proves useful for collaborative phenotyping involving ECG, cardiac imaging, and behavioral data. Our theoretical contributions include update rules with proven convergence, adaptive view weighting, and privacy-preserving protocols. This presents a new standard for geometry-aware federated learning in healthcare, turning advanced math into workable solutions for analyzing sensitive medical data while ensuring both rigor and clinical relevance. Keywords: Federated learning, Multi-view clustering, Heat kernel methods, Privacy-preserving healthcare, Medical data analysis, Quantum field theory, Distributed machine learning 1 Introduction The increasing prevalence of heterogeneous data sources has spurred significant interest in multi-view (MV) clustering, wherein a single entity is characterized by multiple distinct feature sets or “views.” In such contexts, traditional clustering methods often fall short, necessitating sophisticated approaches that can effectively integrate complementary information across diverse views while maintaining coherence in the resulting partition. In addition, the rapid evolution of large-scale artificial intelligence (AI) systems - from centralized language models to decentralized, agent-based architectures - has fundamentally shifted how learning is conceptualized and deployed. In these new paradigms, learning must occur not only across heterogeneous data views but across distributed agents, each operating within their own local context, resource constraints, and knowledge representation. This shift demands a theoretical and algorithmic framework that accommodates structural diversity, communication limitations, and the intrinsic geometry of information flow. In this work, we propose a novel soft clustering framework that replaces conventional Euclidean distance (CED) with kernelized geometric measures derived from H-KC – tools rooted in QFT and spectral analysis. By modeling similarity via heat diffusion over data manifolds, our approach provides a richer, geometry-aware mechanism for aggregating and aligning distributed views. Moreover, the theoretical underpinnings from QFT offer a natural interpretation of influence propagation – a notion increasingly relevant in the context of collaborative AI and agent-based systems. Consider the case of multiple hospitals collaboratively monitoring cardiovascular health through FL – one through ECG signals, another via MRI imaging, others through wearables, EHRs, or genetic data. While CED treats each view as flat and uniform, it fails to capture the curved anatomical structures in imaging or the relational dependencies in medical histories. The HKD, in contrast, models how “information” or similarity diffuses across each hospital’s local data manifold—honoring intrinsic structures. This enables the clustering of patients who may appear dissimilar numerically but share underlying physiological patterns across modalities. Our method enables flexible, privacy-preserving learning across multiple views and agents, with theoretical guarantees and practical scalability. It bridges foundational mathematics with contemporary machine learning (ML) architecture, contributing a unified lens to interpret and design decentralized intelligent systems (DIS). In summary, the contributions of this study include: 1. In the era of modular AI, where not only data views but distributed agents must harmonize across diverse sources, we propose a unified theoretical framework bridging MV learning, kernel geometry, and quantum field theory (QFT) to support scalable, interpretable federated learning (FL). 2. Just as large language models (LLMs) or “LLM-agents” encode semantic geometry in high-dimensional manifolds, this work proposes heat kernel-based soft clustering to capture local geometric structures across decentralized views. In other words, this paper revisits the principles of information diffusion, not only across data views but across distributed agents-drawing connections from classical field theory to the most recent breakthroughs in neural language model. 3. We introduce a novel centralized clustering algorithms that bridge the Heat Kernel-Enhanced Multi-View Fuzzy Clustering (HK-MVFC). 4. The proposed HK-MVFC algorithm employs H-KC to transform the CED metric into an HKD, enabling more effective capturing of complex relationship between data points and cluster centers across multiple views. Here, HK-MVFC incorporates view-specific weighting factors to balance the contribution of each view and treats feature-specific weights across views are equally important. 5. We implement the mathematical formulation for federated HK-MVFC, abbreviated as FedHK-MVFC. The federated variant of HK-MVFC employs a parameter synchronization protocol that preserves data privacy. 2 Literature Review In this section, we review the relevant literature on federated learning, clustering, and their intersections. Federated learning (FL) has emerged as a paradigm for collaborative model training across decentralized data sources while preserving privacy [ 1 , 2 ] . Early works focused on federated averaging (FedAvg) and its variants, addressing communication efficiency and robustness to data heterogeneity. Recent advances include personalization strategies [ 3 , 4 ] , secure aggregation [ 5 ] , and applications in healthcare, finance, and IoT. Clustering algorithms, such as k-means, hierarchical clustering, and fuzzy c-means (FCM) [ 6 ] , are foundational in unsupervised learning. Multi-view clustering extends these methods to data with multiple feature representations, leveraging complementary information for improved accuracy [ 7 , 8 ] . Kernel-based approaches, including spectral clustering and kernel k-means, address non-linear relationships and manifold structures [ 9 , 10 ] . Distributed clustering integrates clustering algorithms with distributed computing frameworks to scale to large datasets and decentralized environments [ 11 , 12 ] . Federated clustering combines FL principles with clustering, enabling privacy-preserving unsupervised learning across clients [ 13 , 14 ] . Existing federated clustering methods often rely on communication-efficient protocols and local model updates, but face challenges in handling heterogeneous data distributions and aligning cluster assignments across clients. Recent works have explored kernel methods and manifold learning in federated settings [ 15 , 16 ] , highlighting the importance of geometry-aware similarity measures. However, most approaches use conventional distance metrics and lack theoretical guarantees for convergence and consistency in multi-view, distributed contexts. Our work builds on these foundations by introducing heat-kernel coefficients and kernelized distances for multi-view federated clustering, bridging quantum field theory concepts with practical distributed learning algorithms. 2.1 Federated Learning Federated learning (FL) enables decentralized model training by aggregating updates from multiple clients, each operating on local data. The core mechanism involves iterative communication between clients and a central server, where clients compute local model updates and the server aggregates these updates to refine the global model. This approach mitigates privacy risks associated with centralized data storage and is particularly suited for sensitive domains such as healthcare, finance, and mobile applications. Key challenges in FL include handling data heterogeneity, ensuring communication efficiency, and maintaining model convergence. Techniques such as federated averaging (FedAvg) [ 1 ] , secure aggregation [ 5 ] , and personalization strategies [ 3 , 17 ] have been developed to address these issues. Recent research also explores robust aggregation methods, differential privacy, and federated optimization algorithms to further enhance

Read the complete text