Training Longer reduces variance or bias?

I think training longer increases bias?

Hi, @rasgulla.c !

Normally, the longer you train, the more prone to overfitting (high variance) your training becomes. That’s why the early stopping is a good choice for simple models.

1 Like

Hi @rasgulla.c,
I think, theoretically, training longer or training a bigger network can reduce the training bias (see C3W1Video: Improving your Model Performance). But in reality, for a plain network, when it gets too deep, the training error tends to go up and that’s why ResNet was introduced (see C4W2Video: ResNets). Hope this help!