Aggregation on Learnable Manifolds for Asynchronous Federated Optimisation

Archie Licudi, Anshul Thakur, Soheila Molaei, Danielle Belgrave, David A. Clifton
Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, PMLR 300:4942-4950, 2026.

Abstract

Asynchronous federated learning (FL) with heterogeneous clients faces two key issues: curvature-induced loss barriers encountered by standard linear parameter interpolation techniques (e.g. FedAvg) and interference from stale updates misaligned with the server’s current optimisation state. To alleviate these issues, we introduce a geometric framework that casts aggregation as curve learning in a Riemannian model space and decouples choice of update direction from staleness conflict resolution. Within this, we propose $\textbf{AsyncBezier}$, which replaces linear aggregation with low-degree polynomial (B{é}zier) trajectories to bypass loss barriers, and $\textbf{OrthoDC}$, which orthogonally projects delayed updates to reduce interference. We establish framework-level convergence guarantees covering each variant given simple assumptions on their components. On three datasets spanning general-purpose and healthcare domains, including LEAF Shakespeare and FEMNIST, our approach consistently improves accuracy and client fairness over strong asynchronous baselines; finally, we show that these gains are preserved even when other methods are allocated a higher local compute budget.

Cite this Paper

BibTeX
@InProceedings{pmlr-v300-licudi26a, title = { Aggregation on Learnable Manifolds for Asynchronous Federated Optimisation }, author = {Licudi, Archie and Thakur, Anshul and Molaei, Soheila and Belgrave, Danielle and Clifton, David A.}, booktitle = {Proceedings of The 29th International Conference on Artificial Intelligence and Statistics}, pages = {4942--4950}, year = {2026}, editor = {Khan, Emtiyaz and Li, Yingzhen and Solin, Arno and Ramdas, Aaditya}, volume = {300}, series = {Proceedings of Machine Learning Research}, month = {02--05 May}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v300/main/assets/licudi26a/licudi26a.pdf}, url = {https://proceedings.mlr.press/v300/licudi26a.html}, abstract = { Asynchronous federated learning (FL) with heterogeneous clients faces two key issues: curvature-induced loss barriers encountered by standard linear parameter interpolation techniques (e.g. FedAvg) and interference from stale updates misaligned with the server’s current optimisation state. To alleviate these issues, we introduce a geometric framework that casts aggregation as curve learning in a Riemannian model space and decouples choice of update direction from staleness conflict resolution. Within this, we propose $\textbf{AsyncBezier}$, which replaces linear aggregation with low-degree polynomial (B{é}zier) trajectories to bypass loss barriers, and $\textbf{OrthoDC}$, which orthogonally projects delayed updates to reduce interference. We establish framework-level convergence guarantees covering each variant given simple assumptions on their components. On three datasets spanning general-purpose and healthcare domains, including LEAF Shakespeare and FEMNIST, our approach consistently improves accuracy and client fairness over strong asynchronous baselines; finally, we show that these gains are preserved even when other methods are allocated a higher local compute budget. } }
Endnote
%0 Conference Paper %T Aggregation on Learnable Manifolds for Asynchronous Federated Optimisation %A Archie Licudi %A Anshul Thakur %A Soheila Molaei %A Danielle Belgrave %A David A. Clifton %B Proceedings of The 29th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2026 %E Emtiyaz Khan %E Yingzhen Li %E Arno Solin %E Aaditya Ramdas %F pmlr-v300-licudi26a %I PMLR %P 4942--4950 %U https://proceedings.mlr.press/v300/licudi26a.html %V 300 %X Asynchronous federated learning (FL) with heterogeneous clients faces two key issues: curvature-induced loss barriers encountered by standard linear parameter interpolation techniques (e.g. FedAvg) and interference from stale updates misaligned with the server’s current optimisation state. To alleviate these issues, we introduce a geometric framework that casts aggregation as curve learning in a Riemannian model space and decouples choice of update direction from staleness conflict resolution. Within this, we propose $\textbf{AsyncBezier}$, which replaces linear aggregation with low-degree polynomial (B{é}zier) trajectories to bypass loss barriers, and $\textbf{OrthoDC}$, which orthogonally projects delayed updates to reduce interference. We establish framework-level convergence guarantees covering each variant given simple assumptions on their components. On three datasets spanning general-purpose and healthcare domains, including LEAF Shakespeare and FEMNIST, our approach consistently improves accuracy and client fairness over strong asynchronous baselines; finally, we show that these gains are preserved even when other methods are allocated a higher local compute budget.
APA
Licudi, A., Thakur, A., Molaei, S., Belgrave, D. & Clifton, D.A.. (2026). Aggregation on Learnable Manifolds for Asynchronous Federated Optimisation . Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 300:4942-4950 Available from https://proceedings.mlr.press/v300/licudi26a.html.

Related Material