# Fed-FBD: Federated Functional Block Diversification for Isolation, Privacy, and Surgical Unlearning

Source: [Fed-FBD: Federated Functional Block Diversification for Isolation, Privacy, and Surgical Unlearning](<https://www.sota2.com/research/paper/6a2b9274d33dfb23e783af28>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-06-13T05:00:05.486Z

Authors: Weijie Chen, Alan B. McMillan

Published: 2026-06-10T21:06:10.000Z

## Abstract

Federated learning (FL) enables collaborative model training without sharing raw patient data, but standard approaches such as FedAvg treat each client as a black box and provide no mechanism for isolating an adversarial contributor, auditing per-client influence, or honoring a departed participant's right to be forgotten. We present Fed-FBD (Federated Functional Block Diversification), a modular federated architecture that decomposes a ResNet backbone into six functional blocks (the stem, four residual groups, and the classification head) and maintains a warehouse of N color variants, each assembled from independently tracked and contributor-stamped blocks. Fed-FBD provides three capabilities absent in FedAvg: (i) architecturally guaranteed block-level isolation, so that an adversarial or mislabelled client cannot contaminate the clean colous; (ii) privacy-by-design, where membership inference advantage is already indistinguishable from chance before any privacy mechanism is applied; and (iii) surgical machine unlearning of a departed participant's contribution at sub-second cost and without retraining. Experiments on six MedMNIST-2D datasets, PathMNIST at 224x224, and CIFAR-10 show that Fed-FBD trades a modest 0.3%-3.1% IID accuracy gap on the adequately sized datasets for these guarantees, remains within 0.8%-4.0% of FedAvg at Dirichlet alpha=1.0 on three of four datasets, and confines all six adversarial attacks we study to the poisoned client's own blocks with at most +/-0.01 AUC drift on the clean colors.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Image Classification on DermaMNIST 28x28 (val)](<https://www.sota2.com/research/sota/image-classification-on-dermamnist-28x28-val.md>): Image Classification; DermaMNIST 28x28 (val); AUC: 89.04.

- [Binary Image Classification on PneumoniaMNIST 28x28 (val)](<https://www.sota2.com/research/sota/binary-image-classification-on-pneumoniamnist-28x28-val.md>): Binary Image Classification; PneumoniaMNIST 28x28 (val); AUC: 95.2.

- [Binary Image Classification on BreastMNIST 28x28 (val)](<https://www.sota2.com/research/sota/binary-image-classification-on-breastmnist-28x28-val.md>): Binary Image Classification; BreastMNIST 28x28 (val); AUC: 61.36.

- [Image Classification on BloodMNIST 28x28 (val)](<https://www.sota2.com/research/sota/image-classification-on-bloodmnist-28x28-val.md>): Image Classification; BloodMNIST 28x28 (val); AUC: 0.9933.

- [Image Classification on PathMNIST 28x28 (val)](<https://www.sota2.com/research/sota/image-classification-on-pathmnist-28x28-val.md>): Image Classification; PathMNIST 28x28 (val); AUC: 96.82.

- [Image Classification on PathMNIST 224x224 (val)](<https://www.sota2.com/research/sota/image-classification-on-pathmnist-224x224-val.md>): Image Classification; PathMNIST 224x224 (val); AUC: 98.8.

- [Image Classification on CIFAR-10 (val) (AUC)](<https://www.sota2.com/research/sota/image-classification-on-cifar-10-val-auc.md>): Image Classification; CIFAR-10 32x32 (val); AUC: 94.74.

- [Image Classification on RetinaMNIST 28x28 (val)](<https://www.sota2.com/research/sota/image-classification-on-retinamnist-28x28-val.md>): Image Classification; RetinaMNIST 28x28 (val); AUC: 64.03.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2606.12679>)

[PDF](<https://arxiv.org/pdf/2606.12679.pdf>)
