Add Flower Baseline: FedHT

[chancejohnstone]

Paper

Tong, Q., Liang, G., Zhu, T. and Bi, J., 2020. Federated nonconvex sparse learning. arXiv preprint arXiv:2101.00052.

Link

https://arxiv.org/abs/2101.00052

Maybe give motivations about why the paper should be implemented as a baseline.

Myself and my coauthors used Fed-HT (and Fed-IterHT) in recent work to generate sparse, federated models. These two aggregation strategies were effective in generating highly predictive models while also allowing for constrained sparsity (with some sparsity threshold), instead of just sparsity through regularization. We'd like to implement this baseline for easier implementation in the future through Flower. The goal with this baseline is to implement Fed-HT and Fed-IterHT as a custom Flower aggregation strategy and recreate the methodology of the paper, to include applications using simulated and benchmark datasets.

Is there something else you want to add?

No

Implementation

To implement this baseline, it is recommended to do the following items in that order:

For first time contributors

• Read the first contribution doc
• Complete the Flower tutorial
• Read the Flower Baselines docs to get an overview:
  • How to use Flower Baselines
  • How to contribute a Flower Baseline

Prepare - understand the scope

• Read the paper linked above
• Decide which experiments you'd like to reproduce. The more the better!
• Follow the steps outlined in Add a new Flower Baseline.
• You can use as reference other baselines that the community merged following those steps.

Verify your implementation

• Follow the steps indicated in the EXTENDED_README.md that was created in your baseline directory
• Ensure your code reproduces the results for the experiments you chose
• Ensure your README.md is ready to be run by someone that is no familiar with your code. Are all step-by-step instructions clear?
• Ensure running the formatting and typing tests for your baseline runs without errors.
• Clone your repo on a new directory, follow the guide on your own README.md and verify everything runs.

[jafermarq]

HI @chancejohnstone 👋 It's great to see you want to add Fed-HT to the Flower Baselines collection ! Out of curiosity, what experiments from the paper are you planing to reproduce ? maybe parts of Fig 5? Please ping us if anytime if you have some doubts or if you encounter some problems.

[chancejohnstone]

@jafermarq That's a great idea; recreating parts of Figure 5 would be an effective way to showcase Fed-HT and Fed-IterHT. I would definitely try to start there.

[chancejohnstone]

@jafermarq Just wanted to give an update on progress for this baseline. Everything seems to be in order as far as the Fed-HT and FedIter-HT implementation is concerned. I have working examples using the datasets from the paper; the results don't currently match exactly due to a lack of knowledge on exact hyperparameters, e.g., weight decay, num_local_epochs, used for the plots. I have reached out to the authors to get those. Formatting and test scripts seem to run without error locally; I get a code rating at the end of running test-baseline.sh, but I'm not sure if that's the end of the script. The code fails the test check after commit and I'm not sure what the issue is. Happy to provide more details as necessary.