Initially, I was thinking of adding relative noise, e.g., 1% of the signal. However, after realizing that the signal is in log-scale already, we shouldn't use "relative noise" here, instead, we should use
ode_data += randn(size(ode_data)) .* noise
BTW, adding noise is pretty useful to regularize the parameters.
https://github.com/jiweiqi/CellBox.jl/blob/47e8652a3d38d7189fca12b763ab6818f310c586/spilt/network.jl#L62
Initially, I was thinking of adding relative noise, e.g., 1% of the signal. However, after realizing that the signal is in log-scale already, we shouldn't use "relative noise" here, instead, we should use
ode_data += randn(size(ode_data)) .* noise
BTW, adding noise is pretty useful to regularize the parameters.