arxiv: 2605.07860 · v1 · submitted 2026-05-08 · 💻 cs.LG · cs.AI

Recognition: no theorem link

On the Tradeoffs of On-Device Generative Models in Federated Predictive Maintenance Systems

Usevalad Milasheuski , Piero Baraldi , Enrico Zio , Stefano Savazzi

Authors on Pith no claims yet

Pith reviewed 2026-05-11 02:02 UTC · model grok-4.3

classification 💻 cs.LG cs.AI

keywords Federated LearningGenerative ModelsDiffusion ModelsPredictive MaintenanceAnomaly DetectionPartial FederationTime Series AnalysisIoT Systems

0 comments

The pith

In federated predictive maintenance, partial sharing of diffusion model components outperforms full model sharing under bandwidth and non-IID constraints.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper investigates the use of generative models including VAEs, GANs, and diffusion models within federated learning frameworks for unsupervised anomaly detection in industrial time series data used for predictive maintenance. It evaluates full versus partial federation strategies where only selected model parts are shared among devices, highlighting different impacts on performance, stability, and communication overhead. The analysis introduces a taxonomy that treats partial component sharing as a structured way to personalize models while respecting data privacy and resource limits. Experiments on a real-world dataset demonstrate that diffusion models with decoder sharing can achieve superior results compared to complete model federation in challenging non-IID and bandwidth-restricted conditions.

Core claim

The paper establishes that while full federation enhances stability for GAN configurations relative to independent training, diffusion models benefit more from partial federation, particularly by sharing only the decoder, leading to better utility in heterogeneous, bandwidth-constrained federated predictive maintenance systems. This is supported by a proposed taxonomy formalizing such partial sharing mechanisms.

What carries the argument

The taxonomy for federated generative models, which categorizes approaches by which model components such as the encoder or decoder are selectively shared to enable personalization and efficiency.

If this is right

GAN models gain training stability from full federation but remain less robust than VAE and DDPM options.
Diffusion models with partial decoder sharing reduce communication needs while improving performance in non-IID settings.
Partial federation provides a principled way to balance model utility and scalability in IoT predictive maintenance applications.
VAE-based approaches offer a robust alternative across different federation levels.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

If the taxonomy is extended, it could guide component selection in other federated generative tasks such as image synthesis or sensor data generation.
Dynamic selection of shared components based on measured data heterogeneity could further optimize these systems.
The findings point toward hybrid architectures where clients retain private encoders and share only task-specific decoders to enhance privacy.

Load-bearing premise

The observed superiority of partial federation for diffusion models holds for the particular non-IID characteristics and bandwidth constraints of the single real-world time series dataset examined.

What would settle it

Replicating the experiments across several additional industrial datasets with quantified heterogeneity metrics and varying communication budgets to verify whether decoder sharing consistently outperforms full sharing for diffusion models.

Figures

Figures reproduced from arXiv: 2605.07860 by Enrico Zio, Piero Baraldi, Stefano Savazzi, Usevalad Milasheuski.

**Figure 1.** Figure 1: Problem Setup for PdM TSAD. During the training stage, clients train the local model and transmit shared weight ws c to the server for multiple rounds. The validation stage individually estimates personalized anomaly threshold ϵc for each client. Finally, during testing, the clients evaluate the performance on the new time series data. transfer of large models or frequent updates [10]. These challenges mot… view at source ↗

**Figure 2.** Figure 2: A comparative overview of major generative model architectures: GANs, VAEs, and Diffusion Models. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 3.** Figure 3: Examples of the Aramis dataset for the selected use-case: a) - sensors operating under normal conditions throughout [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗

**Figure 4.** Figure 4: Time offset for different experimental setups: a) - Centralized Learning; b) - Independent Learning; c) Federated [PITH_FULL_IMAGE:figures/full_fig_p013_4.png] view at source ↗

read the original abstract

Federated Learning (FL) has emerged as a promising paradigm for preserving client data ownership and control over distributed Internet of Things (IoT) environments. While discriminative models dominate most FL use cases, recent advances in generative models -- such as Variational Autoencoders (VAE), Generative Adversarial Networks (GAN), and Diffusion Models (DM) -- offer new opportunities for unsupervised anomaly detection in time series analysis, with relevant applications in predictive maintenance (PdM) in critical industrial infrastructures. In this work, we present a comprehensive analysis of VAEs, GANs, and DMs in the context of federated PdM. We analyze their performance and communication overhead under both full and partial federation setups, where only subsets of model components are shared. Building on this analysis, the paper proposes a novel taxonomy for federated generative models that formalizes partial component sharing as a principled mechanism for model personalization. Our experiments over a real-world time series dataset reveal distinct trade-offs in model utility, stability, and scalability, especially in heterogeneous and bandwidth-constrained FL settings. For the evaluated GAN-based configurations, full federation improves training stability relative to independent local training, although the model remains less robust than the VAE- and DDPM-based alternatives. For DMs, however, partial federation -- especially decoder sharing -- can outperform full federation in bandwidth-constrained, non-IID settings.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The taxonomy formalizing partial component sharing is the actual new piece, but the DM decoder-sharing advantage rests on one dataset with no reported heterogeneity measures.

read the letter

The main takeaway is a taxonomy that treats partial federation—sharing only certain layers or components—as a deliberate design lever for generative models in federated settings. That framing is new relative to the usual full-model or local-only choices, and it is applied here to VAEs, GANs, and diffusion models for predictive maintenance on time series data. The paper walks through how these choices affect utility, stability, and bandwidth use, which is directly relevant for on-device IoT deployments where clients cannot afford full model exchange every round. The experiments compare full federation, partial setups, and independent local training on a real-world dataset, and they surface different trade-offs: GANs gain stability from full sharing while diffusion models sometimes do better with decoder-only sharing under constraints. That comparison is useful even if the numbers are not yet definitive. The soft spot is the experimental grounding. Everything sits on a single dataset, and the abstract supplies no quantitative handle on client heterogeneity—no Wasserstein distances, no Dirichlet parameters, no per-client distribution shift statistics. Without those controls it is hard to know whether the reported decoder-sharing win for diffusion models is a general property of bandwidth-limited non-IID regimes or an artifact of this particular data and implementation. Error bars and statistical tests are also missing from the summary, so the performance gaps are difficult to weigh. This is for researchers already working on federated anomaly detection or maintenance in industrial IoT who need practical pointers on model architecture and sharing strategy. It gives a workable taxonomy and some initial trade-off maps, but the central empirical claim needs tighter controls before it can be treated as reliable guidance. I would send it to peer review. The taxonomy is worth referee discussion and the experimental gaps are straightforward to address with additional datasets and reporting.

Referee Report

2 major / 2 minor

Summary. The paper analyzes VAEs, GANs, and diffusion models (DMs) for unsupervised anomaly detection in federated predictive maintenance on time-series IoT data. It compares full federation against partial federation (sharing only subsets of model components such as the decoder), introduces a taxonomy formalizing partial component sharing for personalization, and reports experimental trade-offs in utility, stability, and communication overhead on a single real-world dataset, with the key finding that decoder-only partial federation can outperform full federation for DMs under bandwidth-constrained non-IID conditions.

Significance. If the empirical claims are substantiated with proper controls, the taxonomy and partial-federation analysis could provide a useful framework for balancing model performance and communication costs when deploying generative models on-device in heterogeneous FL settings for industrial PdM.

major comments (2)

[Abstract and Experiments] Abstract and Experiments section: the headline result that 'partial federation -- especially decoder sharing -- can outperform full federation' for DMs in bandwidth-constrained non-IID settings lacks any quantitative support in the provided text. No heterogeneity metric (Wasserstein distance, Dirichlet parameter, per-client shift), no per-round bit budget, no error bars, no statistical significance tests, and no data-split details are reported, so the observed advantage cannot be attributed to the claimed regime rather than dataset idiosyncrasies.
[Experimental Setup] Experimental Setup: reliance on a single real-world time-series dataset without characterizing its non-IID degree or explicit bandwidth constraints makes the cross-model trade-off claims (full vs. partial federation for GANs, VAEs, and DMs) difficult to generalize or reproduce.

minor comments (2)

The abstract states that experiments 'reveal distinct trade-offs' yet supplies no numerical values, tables, or figures; moving at least one key metric (e.g., F1 or reconstruction error under each regime) into the abstract would improve clarity.
The proposed taxonomy for partial component sharing is mentioned but not formally defined with notation or a diagram in the abstract; a concise definition or table would help readers understand the 'decoder sharing' configuration.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for their insightful comments on our work. We address each of the major comments below and outline the revisions we will make to improve the clarity and rigor of the experimental analysis.

read point-by-point responses

Referee: [Abstract and Experiments] Abstract and Experiments section: the headline result that 'partial federation -- especially decoder sharing -- can outperform full federation' for DMs in bandwidth-constrained non-IID settings lacks any quantitative support in the provided text. No heterogeneity metric (Wasserstein distance, Dirichlet parameter, per-client shift), no per-round bit budget, no error bars, no statistical significance tests, and no data-split details are reported, so the observed advantage cannot be attributed to the claimed regime rather than dataset idiosyncrasies.

Authors: We agree that the abstract and main text would benefit from more explicit quantitative details to support the headline finding. The Experiments section contains comparative tables showing performance differences between full and partial federation for the DMs, but we will enhance the manuscript by adding: a heterogeneity metric (e.g., Wasserstein distance between client data distributions), per-round bit budgets for the different sharing strategies, error bars from multiple independent runs, and detailed data-split information. These revisions will help attribute the observed advantages specifically to the bandwidth-constrained non-IID conditions. revision: yes
Referee: [Experimental Setup] Experimental Setup: reliance on a single real-world time-series dataset without characterizing its non-IID degree or explicit bandwidth constraints makes the cross-model trade-off claims (full vs. partial federation for GANs, VAEs, and DMs) difficult to generalize or reproduce.

Authors: We recognize that a single dataset limits the generalizability of the trade-off claims. The dataset used is a real industrial time-series collection with inherent heterogeneity from different equipment. In the revision, we will characterize the non-IID degree by reporting metrics such as the Wasserstein distance or an approximated Dirichlet parameter across clients, and we will explicitly define the bandwidth constraints through the communication volumes for partial vs. full model sharing. This will improve reproducibility. However, incorporating additional datasets is not feasible in this revision cycle. revision: partial

standing simulated objections not resolved

The generalizability of the cross-model trade-offs to other datasets or settings, since the study is based on one real-world dataset and expanding to multiple datasets would require new experiments.

Circularity Check

0 steps flagged

No circularity: empirical claims rest on dataset experiments without self-referential derivations

full rationale

The manuscript presents no equations, derivations, or fitted parameters that reduce to their own inputs. Central claims (e.g., partial decoder sharing outperforming full federation for diffusion models) are supported solely by experimental comparisons on one real-world time-series dataset. The proposed taxonomy formalizes partial component sharing from the same analysis without invoking self-citations as load-bearing uniqueness theorems or smuggling ansatzes. No self-definitional loops, fitted-input predictions, or renaming of known results appear. This is the expected non-finding for an experimental FL paper whose results are externally falsifiable via replication on other datasets.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The paper is an empirical comparison of existing generative models under federated protocols; it introduces no new mathematical free parameters, axioms, or postulated entities beyond standard assumptions of federated learning and generative modeling.

pith-pipeline@v0.9.0 · 5558 in / 1100 out tokens · 41309 ms · 2026-05-11T02:02:47.012032+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

60 extracted references · 60 canonical work pages · 7 internal anchors

[1]

Predictive maintenance algorithms, artificial intelligence digital twin technologies, and internet of robotic things in big data-driven industry 4.0 manufacturing systems,

M. Nagy, M. Figura, K. Valaskova, and G. L ˘az˘aroiu, “Predictive maintenance algorithms, artificial intelligence digital twin technologies, and internet of robotic things in big data-driven industry 4.0 manufacturing systems,”Mathematics, vol. 13, no. 6, 2025. [Online]. Available: https://www.mdpi.com/2227-7390/13/6/981

work page 2025
[2]

Prognostics and health management (phm): Where are we and where do we (need to) go in theory and practice,

E. Zio, “Prognostics and health management (phm): Where are we and where do we (need to) go in theory and practice,”Reliability Engineering & System Safety, vol. 218, p. 108119, 2022. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0951832021006153

work page 2022
[3]

Advanced anomaly detection in manufacturing processes: Leveraging feature value analysis for normalizing anomalous data,

S. Kim, H. Seo, and E. C. Lee, “Advanced anomaly detection in manufacturing processes: Leveraging feature value analysis for normalizing anomalous data,”Electronics, vol. 13, no. 7, 2024. [Online]. Available: https://www.mdpi.com/2079-9292/13/7/1384

work page 2024
[4]

An anomaly de- tection framework for cyber-security data,

M. Evangelou and N. M. Adams, “An anomaly de- tection framework for cyber-security data,”Computers & Security, vol. 97, p. 101941, 2020. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0167404820302170

work page 2020
[5]

Detection and explanation of anomalies in healthcare data,

D. Samariya, J. Ma, S. Aryal, and X. Zhao, “Detection and explanation of anomalies in healthcare data,”Health Information Science and Systems, vol. 11, no. 1, p. 20, Apr 2023

work page 2023
[6]

Representing data quality in sensor data streaming environments,

A. Klein and W. Lehner, “Representing data quality in sensor data streaming environments,”ACM J. Data Inf. Qual., vol. 1, pp. 10:1–10:28, 2009. [Online]. Available: https://api.semanticscholar.org/CorpusID:17086637

work page 2009
[7]

Deep generative modelling: A comparative review of vaes, gans, normalizing flows, energy-based and autoregressive models,

S. Bond-Taylor, A. Leach, Y . Long, and C. G. Willcocks, “Deep generative modelling: A comparative review of vaes, gans, normalizing flows, energy-based and autoregressive models,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 44, no. 11, pp. 7327– 7347, 2022

work page 2022
[8]

Regulation (eu) 2016/679 of the european parliament and of the council (general data protection regulation),

European Parliament and Council of the European Union, “Regulation (eu) 2016/679 of the european parliament and of the council (general data protection regulation),” Official Journal of the European Union, L 119, 4 May 2016, pp. 1–88, May 2016, oJ L 119, 4.5.2016, pp. 1–88. [Online]. Available: https://eur-lex.europa.eu/eli/reg/2016/679/oj

work page 2016
[9]

Federated learning on non-iid data silos: An experimental study,

Q. Li, Y . Diao, Q. Chen, and B. He, “Federated learning on non-iid data silos: An experimental study,” in2022 IEEE 38th International Conference on Data Engineering (ICDE), 2022, pp. 965–978

work page 2022
[10]

Energy inefficiency in iot networks: Causes, impact, and a strategic framework for sustainable optimisation,

Z. Almudayni, B. Soh, H. Samra, and A. Li, “Energy inefficiency in iot networks: Causes, impact, and a strategic framework for sustainable optimisation,”Electronics, vol. 14, no. 1, p. 159, 2025. [Online]. Available: https://doi.org/10.3390/electronics14010159

work page doi:10.3390/electronics14010159 2025
[11]

Communication-efficient learning of deep networks from decentralized data,

H. B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas, “Communication-efficient learning of deep networks from decentralized data,” 2023. [Online]. Available: https://arxiv.org/abs/1602.05629

work page arXiv 2023
[12]

Federated learning for iot: A survey of techniques, challenges, and applications,

E. Dritsas and M. Trigka, “Federated learning for iot: A survey of techniques, challenges, and applications,”Journal of Sensor and Actuator Networks, vol. 14, no. 1, 2025. [Online]. Available: https://www.mdpi.com/2224-2708/14/1/9

work page 2025
[13]

Fedsw- tsad: SWGAN-based federated time series anomaly detection,

X. Zhang, H. Zhao, W. Zhang, S. Cao, H. Sun, and B. Zhang, “Fedsw- tsad: SWGAN-based federated time series anomaly detection,”Sensors, vol. 25, no. 13, p. 4014, 2025

work page 2025
[14]

Federated generative models for predictive maintenance in industrial environments,

U. Milasheuski, P. Baraldi, E. Zio, and S. Savazzi, “Federated generative models for predictive maintenance in industrial environments,” in2024 8th International Conference on System Reliability and Safety (ICSRS), 2024, pp. 156–161

work page 2024
[15]

A comprehensive survey of anomaly detection techniques for high dimensional big data,

S. Thudumu, P. Branch, J. Jin, and J. J. Singh, “A comprehensive survey of anomaly detection techniques for high dimensional big data,” Journal of Big Data, vol. 7, no. 1, p. 42, July 2020. [Online]. Available: https://doi.org/10.1186/s40537-020-00320-x

work page doi:10.1186/s40537-020-00320-x 2020
[16]

Deep learning advancements in anomaly detection: A comprehensive survey,

H. Huang, P. Wang, J. Pei, J. Wang, S. Alexanian, and D. Niyato, “Deep learning advancements in anomaly detection: A comprehensive survey,” IEEE Internet of Things Journal, vol. 12, no. 21, pp. 44 318–44 342, 2025

work page 2025
[17]

Comparison of deep learning models for predictive maintenance in industrial manufacturing systems using sensor data,

W. Li and T. Li, “Comparison of deep learning models for predictive maintenance in industrial manufacturing systems using sensor data,” Scientific Reports, vol. 15, no. 1, p. 23545, Jul 2025. [Online]. Available: https://doi.org/10.1038/s41598-025-08515-z

work page doi:10.1038/s41598-025-08515-z 2025
[18]

A survey on intelligent predictive maintenance (ipdm) in the era of fully connected intelligence,

T. Zhu, Y . Ran, X. Zhou, and Y . Wen, “A survey on intelligent predictive maintenance (ipdm) in the era of fully connected intelligence,” IEEE Communications Surveys &; Tutorials, p. 1–1, 2025. [Online]. Available: http://dx.doi.org/10.1109/COMST.2025.3567802

work page doi:10.1109/comst.2025.3567802 2025
[19]

Auto-encoding variational bayes,

D. P. Kingma and M. Welling, “Auto-encoding variational bayes,”

work page
[20]

Auto-Encoding Variational Bayes

[Online]. Available: https://arxiv.org/abs/1312.6114

work page internal anchor Pith review Pith/arXiv arXiv
[21]

Generative Adversarial Networks

I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y . Bengio, “Generative adversarial networks,” 2014. [Online]. Available: https://arxiv.org/abs/1406.2661

work page internal anchor Pith review arXiv 2014
[22]

Denoising Diffusion Probabilistic Models

J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,”ArXiv, vol. abs/2006.11239, 2020. [Online]. Available: https://api.semanticscholar.org/CorpusID:219955663

work page internal anchor Pith review arXiv 2006
[23]

A multimodal anomaly detector for robot-assisted feeding using an lstm-based variational autoencoder,

D. Park, Y . Hoshi, and C. C. Kemp, “A multimodal anomaly detector for robot-assisted feeding using an lstm-based variational autoencoder,” IEEE Robotics and Automation Letters, vol. 3, no. 3, pp. 1544–1551, 2018

work page 2018
[24]

Anomaly detection of time series with smoothness-inducing sequential variational auto-encoder,

L. Li, J. Yan, H. Wang, and Y . Jin, “Anomaly detection of time series with smoothness-inducing sequential variational auto-encoder,” 2021. [Online]. Available: https://arxiv.org/abs/2102.01331

work page arXiv 2021
[25]

Anomaly detection for time series using vae-lstm hybrid model,

S. Lin, R. Clark, R. Birke, S. Sch ¨onborn, N. Trigoni, and S. Roberts, “Anomaly detection for time series using vae-lstm hybrid model,” in ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2020, pp. 4322–4326

work page 2020
[26]

Tanogan: Time series anomaly detection with generative adversarial networks,

M. A. Bashar and R. Nayak, “Tanogan: Time series anomaly detection with generative adversarial networks,” in2020 IEEE Symposium Series on Computational Intelligence (SSCI), 2020, pp. 1778–1785. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 15

work page 2020
[27]

Mad-gan: Multivariate anomaly detection for time series data with generative ad- versarial networks,

D. Li, D. Chen, B. Jin, L. Shi, J. Goh, and S.-K. Ng, “Mad-gan: Multivariate anomaly detection for time series data with generative ad- versarial networks,” inArtificial Neural Networks and Machine Learning – ICANN 2019: Text and Time Series, I. V . Tetko, V . K˚urkov´a, P. Karpov, and F. Theis, Eds. Cham: Springer International Publishing, 2019, pp. 703–716

work page 2019
[28]

Tadgan: Time series anomaly detection using generative adversarial networks,

A. Geiger, D. Liu, S. Alnegheimish, A. Cuesta-Infante, and K. Veera- machaneni, “Tadgan: Time series anomaly detection using generative adversarial networks,” in2020 IEEE International Conference on Big Data (Big Data), 2020, pp. 33–43

work page 2020
[29]

Unpaired Image- to-Image Translation using Cycle-Consistent Adversarial Networks,

J.-Y . Zhu, T. Park, P. Isola, and A. A. Efros, “Unpaired image-to-image translation using cycle-consistent adversarial networks,” 2020. [Online]. Available: https://arxiv.org/abs/1703.10593

work page arXiv 2020
[30]

Diffusion Models Beat GANs on Image Synthesis

P. Dhariwal and A. Nichol, “Diffusion models beat gans on image synthesis,”arXiv preprint arXiv:2105.05233, 2021. [Online]. Available: https://api.semanticscholar.org/CorpusID:234357997

work page internal anchor Pith review arXiv 2021
[31]

Anoddpm: Anomaly detection with denoising diffusion probabilistic models using simplex noise,

J. Wyatt, A. Leach, S. M. Schmon, and C. G. Willcocks, “Anoddpm: Anomaly detection with denoising diffusion probabilistic models using simplex noise,” in2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2022, pp. 649–655

work page 2022
[32]

Unsupervised anomaly detection for multivariate time series using diffusion model,

R. Hu, X. Yuan, Y . Qiao, B. Zhang, and P. Zhao, “Unsupervised anomaly detection for multivariate time series using diffusion model,” inICASSP 2024 - 2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2024, pp. 9606–9610

work page 2024
[33]

Federated generative adversarial learning,

C. Fan and P. Liu, “Federated generative adversarial learning,” 2020. [Online]. Available: https://arxiv.org/abs/2005.03793

work page arXiv 2020
[34]

Federated variational learning for anomaly detection in multivariate time series,

K. Zhang, Y . Jiang, L. Seversky, C. Xu, D. Liu, and H. Song, “Federated variational learning for anomaly detection in multivariate time series,” in 2021 IEEE International Performance, Computing, and Communications Conference (IPCCC), 2021, pp. 1–9

work page 2021
[35]

Training diffusion models with federated learning,

M. de Goede, B. Cox, and J. Decouchant, “Training diffusion models with federated learning,” 2024. [Online]. Available: https://arxiv.org/abs/2406.12575

work page arXiv 2024
[36]

Federated Learning with Personalization Layers

M. G. Arivazhagan, V . Aggarwal, A. K. Singh, and S. Choudhary, “Federated learning with personalization layers,” 2019. [Online]. Available: https://arxiv.org/abs/1912.00818

work page internal anchor Pith review arXiv 2019
[37]

Think locally, act globally: Federated learning with local and global representations,

P. P. Liang, T. Liu, L. Ziyin, R. Salakhutdinov, and L. Morency, “Think locally, act globally: Federated learning with local and global representations,”CoRR, vol. abs/2001.01523, 2020. [Online]. Available: http://arxiv.org/abs/2001.01523

work page arXiv 2001
[38]

Fl-plas: Federated learning with partial layer aggregation for backdoor defense against high-ratio malicious clients,

J. Zhang, Z. Zhou, Y . Li, and Q. Jin, “Fl-plas: Federated learning with partial layer aggregation for backdoor defense against high-ratio malicious clients,” 2025. [Online]. Available: https://arxiv.org/abs/2505.12019

work page arXiv 2025
[39]

Straggler-resilient personalized federated learning,

I. Tziotis, Z. Shen, R. Pedarsani, H. Hassani, and A. Mokhtari, “Straggler-resilient personalized federated learning,”Transactions on Machine Learning Research, 2023. [Online]. Available: https://openreview.net/forum?id=gxEpUFxIgz

work page 2023
[40]

Effectively detecting and diagnos- ing distributed multivariate time series anomalies via unsupervised fed- erated hypernetwork,

J. Hao, P. Chen, J. Chen, and X. Li, “Effectively detecting and diagnos- ing distributed multivariate time series anomalies via unsupervised fed- erated hypernetwork,”Information Processing & Management, vol. 62, no. 4, p. 104107, 2025

work page 2025
[41]

Privtsad-fedwgan: A novel federated learning and wgan framework for privacy-preserving multivariate time series anomaly detection,

H. Jiang, X. Chen, D. Miao, H. Zhang, X. Qin, S. Du, and P. Lu, “Privtsad-fedwgan: A novel federated learning and wgan framework for privacy-preserving multivariate time series anomaly detection,”Expert Systems with Applications, vol. 307, p. 131049, 01 2026

work page 2026
[42]

Z. Soi, C. Fan, A. Shankar, A. Malan, and L. Chen,Federated Time Series Generation on Feature and Temporally Misaligned Data, 09 2025, pp. 384–399

work page 2025
[43]

Federated foundation models on heterogeneous time series,

S. Chen, G. Long, J. Jiang, and C. Zhang, “Federated foundation models on heterogeneous time series,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 15, 2025, pp. 15 839–15 847

work page 2025
[44]

Fedal: Federated dataset learning for general time series foundation models,

S. Chen, G. Long, M. Blumenstein, and J. Jiang, “Fedal: Federated dataset learning for general time series foundation models,”arXiv preprint arXiv:2508.04045, 2025, accepted at ICLR 2026

work page arXiv 2025
[45]

Exploiting shared representations for personalized federated learning,

L. Collins, H. Hassani, A. Mokhtari, and S. Shakkottai, “Exploiting shared representations for personalized federated learning,” in Proceedings of the 38th International Conference on Machine Learning, ser. Proceedings of Machine Learning Research, M. Meila and T. Zhang, Eds., vol. 139. PMLR, 18–24 Jul 2021, pp. 2089–2099. [Online]. Available: https://pro...

work page 2021
[46]

Lstm-autoencoder-based anomaly detection for indoor air quality time- series data,

Y . Wei, J. Jang-Jaccard, W. Xu, F. Sabrina, S. Camtepe, and M. Boulic, “Lstm-autoencoder-based anomaly detection for indoor air quality time- series data,”IEEE Sensors Journal, vol. 23, no. 4, pp. 3787–3800, 2023

work page 2023
[47]

Improved Techniques for Training GANs

T. Salimans, I. Goodfellow, W. Zaremba, V . Cheung, A. Radford, and X. Chen, “Improved techniques for training gans,” 2016. [Online]. Available: https://arxiv.org/abs/1606.03498

work page Pith review arXiv 2016
[48]

Mode collapse in generative adversarial networks: An overview,

Y . Kossale, M. Airaj, and A. Darouichi, “Mode collapse in generative adversarial networks: An overview,” in2022 8th International Confer- ence on Optimization and Applications (ICOA), 2022, pp. 1–6

work page 2022
[49]

Wasserstein GAN

M. Arjovsky, S. Chintala, and L. Bottou, “Wasserstein gan,” 2017. [Online]. Available: https://arxiv.org/abs/1701.07875

work page Pith review arXiv 2017
[50]

Improved Training of Wasserstein GANs

I. Gulrajani, F. Ahmed, M. Arjovsky, V . Dumoulin, and A. Courville, “Improved training of wasserstein gans,” 2017. [Online]. Available: https://arxiv.org/abs/1704.00028

work page Pith review arXiv 2017
[51]

U-net: Convolutional networks for biomedical image segmentation,

O. Ronneberger, P. Fischer, and T. Brox, “U-net: Convolutional networks for biomedical image segmentation,” inMedical Image Computing and Computer-Assisted Intervention – MICCAI 2015, N. Navab, J. Horneg- ger, W. M. Wells, and A. F. Frangi, Eds. Cham: Springer International Publishing, 2015, pp. 234–241

work page 2015
[52]

Perslev, M

M. Perslev, M. H. Jensen, S. Darkner, P. J. Jennum, and C. Igel,U-Time: a fully convolutional network for time series segmentation applied to sleep staging. Red Hook, NY , USA: Curran Associates Inc., 2019

work page 2019
[53]

A Tutorial on Bayesian Optimization

P. I. Frazier, “A tutorial on bayesian optimization,” 2018. [Online]. Available: https://arxiv.org/abs/1807.02811

work page internal anchor Pith review arXiv 2018
[54]

The aramis data challenge to prognostics and health management methods for application in evolving environments,

P. Baraldi, M. Compare, E. Zio, F. Cannarile, and Z. Yang, “The aramis data challenge to prognostics and health management methods for application in evolving environments,”Journal of Risk and Reliability, vol. 237, no. 5, pp. 958–965, October 2023. [Online]. Available: https://ideas.repec.org/a/sae/risrel/v237y2023i5p958-965.html

work page 2023
[55]

Swat: a water treatment testbed for research and training on ics security,

A. P. Mathur and N. O. Tippenhauer, “Swat: a water treatment testbed for research and training on ics security,” in2016 International Workshop on Cyber-physical Systems for Smart Water Networks (CySWater), 2016, pp. 31–36

work page 2016
[56]

Adam: A method for stochastic optimization,

D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,”

work page
[57]

Adam: A Method for Stochastic Optimization

[Online]. Available: https://arxiv.org/abs/1412.6980

work page internal anchor Pith review Pith/arXiv arXiv
[58]

C-mapss dataset,

L. Ouyang and Y . Gao, “C-mapss dataset,” 2025. [Online]. Available: https://dx.doi.org/10.21227/q7dr-1b93

work page doi:10.21227/q7dr-1b93 2025
[59]

Turbofan engine degradation simulation data set 2 (n-cmapss),

M. Chao, C. Arias, K. Kulkarni, K. Goebel, and O. Fink, “Turbofan engine degradation simulation data set 2 (n-cmapss),” N-CMAPSS DS02-006.h5, 2017. [Online]. Available: https://ti.arc.nasa.gov/tech/dash/groups/pcoe/prognostic-data- repository/#turbofan

work page 2017
[60]

Edd datasets,

D. Wei, “Edd datasets,” May 2024. [Online]. Available: https://doi.org/10.6084/m9.figshare.25844203.v2

work page doi:10.6084/m9.figshare.25844203.v2 2024