Quote:
Originally Posted by Hasbinlulz
[You must be logged in to view images. Log in or Register.]
Yeah, except that a simulation only needs to be as complex as will produce consistent and predictive results, you fucking tool.
Read the last paragraph, and see if your fancy letters can help you ferret out the reason why this study is valid rather than invalidated by lack of infinite complexity, as you seem to desire:
|
It's not difficult for a neural network to fit training data. It's a much different beast for it to actually be predictive. In fact, the low error they have achieved in training would make me suspect of overfitting the data, which is a common problem with such systems. They didn't even mention the training method they used for the neural networks. They should require all papers that model the environment to give some testable predictions of the future. At least then we'd be able to tell which ones are possibly valid after a few years.