C1W2 - Can adding data hurt? - What is meant by large model

[alechewitt]

In the video “Can adding data hurt”, Andrew states you need a large model in order to ensure adding data won’t hurt.

What is meant by a large model?

Is this referring to a neural network that has many layers? Or, a a network with a lot of parameters? Or something else?

Many thanks in advance

[TMosh]

@alechewitt

Is this referring to a neural network that has many layers? Or, a a network with a lot of parameters?

Both.

[saou_a]

Hi @alechewitt ,

FIrst of all, in most cases, adding more layers to a neural network will also increase the number of parameters.

Large model means that it has the potential to learn more complex relationships from the data, in that sence, it could mitigate the risk of adding data.

On the other hand, adding too much data to a model not “large enough” could lead to overfitting, meaning it memorizes the training data too well and performs poorly on unseen data.

@saou_a

[TMosh]

In my experience, adding more data helps reduce overfitting, rather than causing it.

These snips are from an article from AWS.

“Data augmentation” increases the amount of training data.