Evolving Federated Density Peak Clustering with Shuffler-Based Anonymization

Miha Ožbot; Seiichi Ozawa; Igor Škrjanc

cs.LGOct 22, 2026

Evolving Federated Density Peak Clustering with Shuffler-Based Anonymization

Authors: Miha Ožbot, Seiichi Ozawa, Igor Škrjanc

Organizations: Faculty of Electrical Engineering, University of Ljubljana, Tržaška cesta 25, Ljubljana, 1000, Slovenia. · Center for Mathematical and Data Sciences, Kobe University, 1-1 Rokkodai-cho, Nada-ku, Kobe, 657-8501, Japan.

Abstract

In this paper, we present a novel one-shot federated clustering approach based on evolving fuzzy clustering and shuffler-based anonymization. In federated clustering, each round reveals additional model information to the server and can increase the opportunity to infer properties of local datasets, even when raw samples are never shared. Consequently, fewer client-server exchanges and unlinkability are important, so transmitted information cannot be attributed to any single client. Data-stream clustering algorithms, which learn in a single pass over the data, align well with the one-shot federated learning paradigm, where a single communication round can be sufficient. The proposed Evolving Federated Density Peak Clustering (eFDPC) method combines evolving fuzzy data-stream clustering and density peak clustering in a federated setting. Each client independently evolves micro-clusters from private data mini-batches through cluster addition and merging. To limit communication, density peak clustering is applied to the learned micro-clusters and only the selected peaks are transmitted, rather than the full local models. A shuffler-based layer is introduced between the clients and the server to anonymize contributions prior to reconstruction and global peak selection. We consider a peak-only variant, eFDPC-P, as well as eFDPC-G with server-side label propagation. Experiments on synthetic and real-world benchmark datasets show that eFDPC outperforms recent federated clustering baselines in one-shot settings and scales to higher-dimensional, non-IID data.

Figures & tables

Figure 1: Overview of the proposed eFDPC pipeline. (1) Each client learns evolving micro-clusters from its private data using mini-batch updates. (2) Each client applies a local DPC step on its micro-clusters to select a small set of local peaks. (3) The selected peak statistics are split into additive shares and sent to multiple shufflers. Each shuffler permutes the received messages and forwards them to the server. (4) The server reconstructs the peak statistics by matching share identifiers and summing the shares, obtaining an unlinkable pool of candidate peaks. (5) The server performs a second DPC step on the reconstructed candidates to select the global peak set, and (6) broadcasts the global peaks back to clients for the next communication round.

Stage	Operation	Time	Memory
Client	Micro-cluster update (per batch)	$\mathcal{O}(N_{b}Cd)$	$\mathcal{O}((C{+}N_{b})d)$
	Micro-cluster update (full pass)	$\mathcal{O}(NCd)$
	Local DPC peak selection	$\mathcal{O}(C^{2}d)$
	Share generation	$\mathcal{O}(SPd)$
Communication	Per client, single round	$\mathcal{O}(Pd)$	$\mathcal{O}(Pd)$
Server	Peak reconstruction	$\mathcal{O}(SKPd)$	$\mathcal{O}(KPd)$

Table 1: Computational and memory complexity of eFDPC.

Panel of Figure 2 — Figure 2: Comparison of eFDPC- $\mathcal{G}$ behavior on Skewed ( $K=10$ , left column) and MNIST ( $K=3$ , right column). From top to bottom, rows depict the decision graph (a,b), cluster grouping (c,d), and clustering accuracy (e,f) at the server-side. The evaluation is performed using the propagated labels on the shared peaks.

Dataset	Metric	eFDPC- $\mathcal{P}$	eFDPC- $\mathcal{G}$	FKDC	$k$ -FED	MUFC	FedFCM	LLF-FCM	LLF-CM
Aggregation $N=788$ $d=2$	ACC [ $10^{-2}$ ]	92 $\pm$ 02	92 $\pm$ 02	85 $\pm$ 03	89 $\pm$ 03	88 $\pm$ 03	82 $\pm$ 05	79 $\pm$ 04	80 $\pm$ 05
	ARI [ $10^{-2}$ ]	86 $\pm$ 05	86 $\pm$ 04	75 $\pm$ 06	81 $\pm$ 05	82 $\pm$ 05	70 $\pm$ 06	67 $\pm$ 07	68 $\pm$ 07
	NMI [ $10^{-2}$ ]	89 $\pm$ 02	90 $\pm$ 02	81 $\pm$ 04	86 $\pm$ 04	85 $\pm$ 03	76 $\pm$ 06	75 $\pm$ 05	76 $\pm$ 05
	Time [ $10^{-3}{\mathrm{s}}$ ]	98 $\pm$ 07	98 $\pm$ 07	94 $\pm$ 13	29 $\pm$ 04	88 $\pm$ 06	27 $\pm$ 10	09 $\pm$ 00	02 $\pm$ 00
Asymmetric $N=1000$ $d=2$	ACC [ $10^{-2}$ ]	96 $\pm$ 02	95 $\pm$ 03	80 $\pm$ 09	87 $\pm$ 10	94 $\pm$ 05	89 $\pm$ 07	77 $\pm$ 09	76 $\pm$ 10
	ARI [ $10^{-2}$ ]	90 $\pm$ 04	90 $\pm$ 05	71 $\pm$ 11	81 $\pm$ 12	88 $\pm$ 06	81 $\pm$ 09	66 $\pm$ 12	66 $\pm$ 11

Table 2: Clustering comparison on synthetic datasets (

K=3

\alpha=0.9

Dataset	Metric	eFDPC- $\mathcal{P}$	eFDPC- $\mathcal{G}$	FKDC	$k$ -FED	MUFC	FedFCM	LLF-FCM	LLF-CM
Dermatology $N=358$ $d=34$	ACC [ $10^{-2}$ ]	88 $\pm$ 04	88 $\pm$ 03	68 $\pm$ 04	79 $\pm$ 08	80 $\pm$ 04	74 $\pm$ 05	71 $\pm$ 06	66 $\pm$ 08
	ARI [ $10^{-2}$ ]	87 $\pm$ 03	86 $\pm$ 04	57 $\pm$ 07	75 $\pm$ 11	72 $\pm$ 07	67 $\pm$ 07	61 $\pm$ 11	51 $\pm$ 17
	NMI [ $10^{-2}$ ]	91 $\pm$ 01	91 $\pm$ 02	73 $\pm$ 04	85 $\pm$ 07	78 $\pm$ 04	76 $\pm$ 05	67 $\pm$ 10	65 $\pm$ 10
	Time [ $10^{-3}{\mathrm{s}}$ ]	48 $\pm$ 03	48 $\pm$ 03	76 $\pm$ 08	29 $\pm$ 03	59 $\pm$ 07	19 $\pm$ 07	04 $\pm$ 01	01 $\pm$ 00
Heart $N=270$ $d=13$	ACC [ $10^{-2}$ ]	78 $\pm$ 05	79 $\pm$ 04	76 $\pm$ 05	75 $\pm$ 07	71 $\pm$ 09	76 $\pm$ 05	73 $\pm$ 07	70 $\pm$ 07
	ARI [ $10^{-2}$ ]	32 $\pm$ 10	34 $\pm$ 08	29 $\pm$ 09	27 $\pm$ 13	21 $\pm$ 14	28 $\pm$ 09	24 $\pm$ 11	18 $\pm$ 11

Table 3: Clustering comparison on real-world datasets (

K=3

\alpha=0.9

Dataset	Metric	eFDPC- $\mathcal{P}$	eFDPC- $\mathcal{G}$	FKDC	$k$ -FED	MUFC	FedFCM	LLF-FCM	LLF-CM
MNIST $K=5$	ACC [ $10^{-2}$ ]	52 $\pm$ 02	59 $\pm$ 04	55 $\pm$ 02	55 $\pm$ 03	45 $\pm$ 03	31 $\pm$ 04	38 $\pm$ 03	35 $\pm$ 04
	ARI [ $10^{-2}$ ]	28 $\pm$ 03	39 $\pm$ 05	36 $\pm$ 02	38 $\pm$ 02	24 $\pm$ 04	13 $\pm$ 04	19 $\pm$ 03	14 $\pm$ 04
	NMI [ $10^{-2}$ ]	42 $\pm$ 01	55 $\pm$ 03	46 $\pm$ 02	50 $\pm$ 02	35 $\pm$ 03	23 $\pm$ 05	28 $\pm$ 02	26 $\pm$ 04
	Time [ $10^{-1}{\mathrm{s}}$ ]	16 $\pm$ 00	16 $\pm$ 00	43 $\pm$ 01	07 $\pm$ 00	55 $\pm$ 00	50 $\pm$ 27	15 $\pm$ 00	05 $\pm$ 01
MNIST $K=10$	ACC [ $10^{-2}$ ]	51 $\pm$ 03	60 $\pm$ 04	54 $\pm$ 02	55 $\pm$ 05	47 $\pm$ 02	33 $\pm$ 03	36 $\pm$ 03	33 $\pm$ 06
	ARI [ $10^{-2}$ ]	29 $\pm$ 03	43 $\pm$ 06	32 $\pm$ 02	39 $\pm$ 04	27 $\pm$ 03	13 $\pm$ 03	17 $\pm$ 02	11 $\pm$ 04

Table 4: Client-count

K

sweep for Dirichlet

\alpha=0.9

Panel of Figure 3 — Figure 3: Mini-batch size sensitivity of eFDPC- $\mathcal{G}$ on MNIST for $K=3,10,30$ clients (Dirichlet $\alpha=0.9$ ). ACC, ARI, and NMI are stable across the mini-batch size $B$ , with a mild optimum at moderate sizes, and improve with the number of clients; runtime falls as $B$ grows and grows with $K$ . Bands show $\pm 1$ standard deviation over the repetitions.

Panel of Figure 4 — Figure 4: Non-IID label partition of MNIST across $K=10$ clients with the Dirichlet split used in the experiments. Small $\alpha$ results in each true cluster being split between a few clients (a), while large $\alpha$ approaches an IID split where every client holds every cluster in similar proportions (d).

Dataset	Metric	eFDPC- $\mathcal{P}$	eFDPC- $\mathcal{G}$	FKDC	$k$ -FED	MUFC	FedFCM	LLF-FCM	LLF-CM
MNIST $\alpha=0.1$	ACC [ $10^{-2}$ ]	48 $\pm$ 05	54 $\pm$ 06	53 $\pm$ 04	48 $\pm$ 06	48 $\pm$ 03	30 $\pm$ 06	38 $\pm$ 03	35 $\pm$ 04
	ARI [ $10^{-2}$ ]	29 $\pm$ 04	37 $\pm$ 07	33 $\pm$ 04	31 $\pm$ 06	26 $\pm$ 04	11 $\pm$ 05	19 $\pm$ 03	14 $\pm$ 04
	NMI [ $10^{-2}$ ]	38 $\pm$ 04	52 $\pm$ 05	43 $\pm$ 04	44 $\pm$ 05	38 $\pm$ 03	20 $\pm$ 06	28 $\pm$ 02	26 $\pm$ 04
	Time [ $10^{0}{\mathrm{s}}$ ]	03 $\pm$ 03	03 $\pm$ 03	04 $\pm$ 00	00 $\pm$ 00	06 $\pm$ 00	12 $\pm$ 14	01 $\pm$ 00	00 $\pm$ 00
MNIST $\alpha=0.3$	ACC [ $10^{-2}$ ]	49 $\pm$ 04	57 $\pm$ 05	54 $\pm$ 03	52 $\pm$ 04	48 $\pm$ 02	33 $\pm$ 03	38 $\pm$ 03	35 $\pm$ 04
	ARI [ $10^{-2}$ ]	28 $\pm$ 04	41 $\pm$ 06	33 $\pm$ 03	35 $\pm$ 03	27 $\pm$ 02	14 $\pm$ 03	19 $\pm$ 03	14 $\pm$ 04

Table 5: Non-IID Dirichlet

\alpha

sweep for

K=10

Clustering

Clustering groups data points by shared structure, developing methods to discover, validate, and interpret groups across diverse data and constraints.

See all 342 papers

Federated Deep Clustering Networks for High-Dimensional and Heterogeneous Data
Sep 18, 2026Morris Stallmann, Charalampos S. Kouzinopoulos, Marcin Pietrasik +1Deep Clustering Federated Learning
Department of Advanced Computing Sciences (DACS) Maastricht University Maastricht, Netherlands
Fed-BRDECS: Privacy-Preserving and Heterogeneity-Aware Federated Deep Embedded Clustering
Oct 5, 2026Haemin Park, Diego Klabjan, Martin W. Braun +2Clustering Deep Clustering
Northwestern University · Intel Corporation
Differentially Private Clustered Federated Learning with Privacy-Preserving Initialization and Normality-Driven Aggregation
Apr 22, 2026Jie Xu, Haaris Mehmood, Rogier Van Dalen +2Federated Learning Aggregation Non-IID Federated Learning
Samsung R&D Institute UK (SRUK) · Samsung AI Centre Cambridge

Sep 18, 2026J/K move · Enter open · S save

Federated Deep Clustering Networks for High-Dimensional and Heterogeneous Data

Morris Stallmann, Charalampos S. Kouzinopoulos, Marcin Pietrasik, Anna Wilbik

Department of Advanced Computing Sciences (DACS) Maastricht University Maastricht, Netherlands

Deep Clustering Federated Learning Non-IID Federated Learning Heterogeneous Federated Learning

Open paper AskPDF arXiv

Clustering high-dimensional data is a fundamental task in unsupervised machine learning with applications to a variety of domains. In the centralized data scenario, this task is commonly solved using deep clustering methods that utilize deep neural network architectures to learn clustering-friendly latent space representations. In Federated Learning, where data is distributed between clients and is private, deep clustering methods are less explored. In particular, recently introduced federated deep clustering methods, despite showing very promising performance, still fall short in reliably providing good performance if data across clients are non-identically-independently distributed. In this work, we introduce a generalization of Deep Clustering Networks to the federated scenario, named FedDCN, that simultaneously optimizes a reconstruction loss and a clustering loss. To ensure robustness and latent space alignment in non-identically-independently distributed data scenarios, FedDCN generates synthetic data augmentations, and its learning objective includes a geometric regularization for latent space alignment. Through experimental evaluation, the effectiveness of the approach under IID and non-IID assumptions is demonstrated, and future research directions are identified.

Abstract

Clustering

Explore similar work

Federated Deep Clustering Networks for High-Dimensional and Heterogeneous Data

Fed-BRDECS: Privacy-Preserving and Heterogeneity-Aware Federated Deep Embedded Clustering

Differentially Private Clustered Federated Learning with Privacy-Preserving Initialization and Normality-Driven Aggregation