Romain Goussault

Add parameter : fraction of partners used at each federated round

4

This is useful when the number of partners is large

AIS_Kriging_S contrib method fails

1

`AIS_Kriging_S` contrib method fails. It now removed from [the end to end tests.](https://github.com/SubstraFoundation/distributed-learning-contributivity/blob/58fa873cb731f8f066bb57917c636f611d4f39e8/tests/contrib_end_to_end_test.py#L66) How to reproduce: scenario = Scenario(2, [0.25, 0.75], epoch_count=4, minibatch_count=2, dataset_name='mnist', contributivity_methods=["AIS_Kriging_S"], dataset_proportion=0.1) exp = Experiment(scenarios_list=[scenario]) exp.run()...

Study Federated learning litterature and libraries

3

https://arxiv.org/pdf/1602.05629.pdf https://arxiv.org/pdf/1805.09767.pdf https://arxiv.org/pdf/1908.07873.pdf https://arxiv.org/pdf/1912.04977.pdf

Profile timing for minibatch federated learning

4

Increasing the number of minibatch seems to increase the time to do a federated run. This might be due to the process of: - creating the model - setting the...

Now each scenario is processed sequentially. Since scenarii are independant we could parallelize the computations. One thing to be wary of is the GPU memory if we learn too many...

What happens when we clip batch_size ?

2

The batch size is clipped here --> https://github.com/SubstraFoundation/distributed-learning-contributivity/blob/b72fa98c0b4db45d368f577d0f6d1a861b1610c2/scenario.py#L584 So if it's clipped it means that sometimes we don't use all the data in the dataset. But we don't give any...

@bowni @Thomas-Galtier @arthurPignet What would happen if we removed this threshold ? https://github.com/SubstraFoundation/distributed-learning-contributivity/blob/808ee93c8593d3b226d3d1aeaa370c6d9d9689f8/mplc/contributivity.py#L452

Signed-off-by: Romain Goussault Update outdated README.MD of docs folder