Paper formulates split federated unlearning and proposes SURE
Original title: Split Federated Unlearning
SURE uses dual-importance estimation and selective perturbation, matching retraining at modest extra cost.
The paper presents what it calls the first systematic formulation of Split Federated Unlearning for split federated learning, covering client-wise and class-wise removal, counterfactual retraining, and partition-specific update constraints. It argues that cross-partition coupling and historical influence make removal harder than in standard federated learning. The proposed SURE method combines dual-importance estimation, selective parameter perturbation, and importance-guided recovery. Adapted existing unlearning methods showed large forgetting-utility trade-offs, while split position and update constraints changed behavior.
Unconfirmed: The claim of being the first systematic SFU formulation is the authors' own.
- Unlearning covers both client-wise and class-wise removal.
- The analysis finds structural limits when model updates are restricted.
- Experiments span model architectures, data distributions, and update modes.
- SURE is described as competitive with mode-matched retraining.
From the original article
1 min read at arxiv.orgSplit federated learning (SFL) combines partitioned model execution with federated aggregation to support collaborative learning across distributed Web and edge environments. However, removing learned contributions in SFL is challenging due to cross-partition coupling and historical influence propagation through dual-side aggregation.
By Jia Chen · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.