Overfitting is a common issue but deep-learning practitioners generally tend to overfit
the model and address the issue through regularization techniques etc. rather than
searching for optimal solution by gradually increasing the parameters.
When the training loss is high and the gap between training and validation loss is also
high, then the model has high variance and has only an over-fitting issue.
It is a common practice in deep learning projects having image, text, etc. data, to have
higher training to validation dataset split ratio for huge datasets when compared to the
small dataset.
Fig: 1