Federated Tumor Segmentation Challenge 2021 - METU-FL Team Challenge Repository

This repository contains METU-FL team implementations for the Federated Tumor Segmentation Challenge 2021.

You can reach our related work here: https://link.springer.com/chapter/10.1007/978-3-031-09002-8_36

The list of given and implemented functions for Task 1 is listed below.

Aggregation function determines how the server side merges the collaborator model updates.

Aggregation Function Name	Source
weighted_average_aggregation	Given by FeTS initiative
fedNova_simplified	METU-FL
make_aggregation_with_improved_nodes	METU-FL
FedAvgM	METU-FL
coordinatewise_median_aggregation	METU-FL

Hyperparameter choice function determines the parameters for each FL round.

Function Name	Source
constant_hyper_parameters	Given by FeTS initiative
lrscheduling_hyper_parameters	METU-FL
adaptive_epoch	METU-FL
adaptive_epoch_with_lr_scheduling	METU-FL

Collaborator choice function determines which collaborators are chosen to train in each FL round.

Function Name	Source
all_collaborators_train	Given by FeTS initiative
choose_random_nodes_with_faster_ones	METU-FL
random_collaborators_train	METU-FL
random_single_collaborator_train	METU-FL

Validation metrics show the metrics to be computed each FL round.

Function Name	Source
channel_sensitivity	Given by FeTS initiative
sensitivity	Given by FeTS initiative
specificity	Given by FeTS initiative

Our challenge combination is as below:

• Hyperparameter choice function: adaptive_epoch_with_lr_scheduling
  adaptive_epoch_with_lr_scheduling is an adaption of AdaComm with a learning scheduling scheme. The number of epochs per round decays according to the decrease in initial loss, and the learning rate decays according to the performance metric (average DICE score).

@inproceedings{MLSYS2019_c8ffe9a5,
 author = {Wang, Jianyu and Joshi, Gauri},
 booktitle = {Proceedings of Machine Learning and Systems},
 editor = {A. Talwalkar and V. Smith and M. Zaharia},
 pages = {212--229},
 title = {Adaptive Communication Strategies to Achieve the Best Error-Runtime Trade-off in Local-Update SGD},
 url = {https://proceedings.mlsys.org/paper/2019/file/c8ffe9a587b126f152ed3d89a146b445-Paper.pdf},
 volume = {1},
 year = {2019}
}

• Collaborator choice function: all_collaborators_train
  All collaborators participate in each FL round.

• Aggregation function: fedAvgM
  Federated Averaging with server momentum uses accumulated gradients for the weight update.

@article{hsu2019measuring,
  title={Measuring the effects of non-identical data distribution for federated visual classification},
  author={Hsu, Tzu-Ming Harry and Qi, Hang and Brown, Matthew},
  journal={arXiv preprint arXiv:1909.06335},
  year={2019}
}

Validation metrics: We used the given default validation metrics but considering limited computational resources, we did not include Hausdorff Distance (include_validation_with_hausdorff = False).