Discussion on Symbolic Fitting Variability

[yixi0527]

I've observed that when applying the same model to a given dataset, the symbolic fitting formulas produced can vary. While I understand that this variability arises from factors such as the selection of data points and inherent randomness during the optimization process, I would like to delve deeper into the fundamental reasons behind this phenomenon.

Is there a method to reliably identify a deterministic formula that effectively describes a relatively simple dataset? Furthermore, is it feasible to discover such a formula using any random seed while maintaining the same network structure?

I encourage everyone to join this discussion and share your insights!

Here are the results of symbolic formulas obtained using the same dataset, the same network structure, but different random seeds.