The success of a learning algorithm depends on its ability to balance between fitting the data and avoiding overfitting.

如果你没有偶尔失败，那说明你没有在做任何真正创新的事情。