What is the best approach to update aggregation_weights based on other DXO's aggregated result in NVFlare?

Healthcare Clara Train SDK
#1

Hi,

I understand that aggregation weights in aggregator should be dictionary to map contributor name to its aggregation weights, and normally should be fixed.

What if for every round, server’s aggregation weights should be different and based on other DXO’s aggregated result?

What would be the best way to approach this problem?

Thanks!

#2

Hi, this would require some modifications to the aggregator code. You can use your own version easily by adopting the nvflare aggregators to your needs.