hello everyone !
I have trained various neural networks, often with the same parameters and training environnement, but I often ended up with imperfect AIs which were always slightly different from their peers. Especially when the environnment contained some randomness.
I just had the idea of making different neural networks with the same training, let’s say a dozen, and to make them vote the decision to take.
My hope being that the different AIs, who are right around 90% of the time, would not be wrong all at the same time.
I wanted to know if anyone ever tried anything similar ? is the idea good or just stupid (and if the latter, why ?)
It would take long to train those, so if anyone tried before and can tell me about the result, that would be nice (and save me much time ! ^^)
Have a nice day !