On the Potential Information Leakage in Loss Computation During Communication Rounds #3
Description
Dear Author,
I highly appreciate your research work and the impressive achievement of reducing communication costs by 87.57% on the PEMS-BAY dataset.
However, while reading the code, I noticed that during the computation of communication rounds, the loss is calculated by first generating predictions through the model and then computing the loss based on these predictions. This procedure seems to implicitly “inform” the model of the target values it should predict in advance. Could this potentially affect the experimental results?
