Compare the parameters after each SI

[DuyLe]

I am implementing a function that evaluates the performance of SN nodes participating in training. I have 2 questions:

1. Assuming there is a node with poor computation, is there any way I can compare the parameters between nodes in the end of each SI?
2. With the finding of the poorly calculated node, can I adjust the weights of the SN nodes by resetting the NodeWeightage parameter in SwarmCallback for the next SI?

Hope to get your reply. Thank you!

[Vinky_99 · Accepted solution]

Hello @DuyLe

1. Comparing Parameters between nodes after each SI:

Yes, it is possible to compare the parameters between nodes after each SI. One way to do this is by collecting the parameter values from each node after each iteration and then comparing them. Another way is to use a parameter server that aggregates the parameter updates from all nodes and then distributes the updated parameters to each node for the next iteration.

2. Adjusting the weights of the SN nodes by resetting the NodeWeightage parameter:

Yes, it is possible to adjust the weights of the SN nodes by resetting the NodeWeightage parameter in SwarmCallback for the next SI. The NodeWeightage parameter is used to assign a weight to each node based on its performance in the previous iteration. By resetting the NodeWeightage parameter for the poorly performing node, the weight assigned to that node will be reduced, which will in turn reduce its influence on the final parameter updates. This can help to improve the overall performance of the swarm. However, it is important to note that resetting the NodeWeightage parameter should be done carefully and only after confirming that the node's poor performance is not due to other factors such as network latency or hardware issues.

I hope it helps! Let me know