W1_Quiz_Large NN Models vs Traditional Learning

I want to discuss this point. As we can see the image’s red line shows the traditional algorithm’s performance and the green line shows the Large NN model.

  1. As we can see the question statement amount of data is not mentioned.
  2. red line (traditional algorithms) is below the green line (Large NN model) even if the amount is data is low.

so I selected the answer true because of the above reason and my answer is marked wrong.

Hi. Actually Andrew explained for small training sets. Traditional learning algo do better in some cases. The diagram is just a hand draft.


Yes, in week 1 Prof Ng has talked about simple learning with smaller set of models. But from week 2 onward, he has tried to explain the techniques that can be used with large data size to train the larger neural network models.

Here’s a link that could be useful.

1 Like

Hi @Muhammad_Usama_Amin

in addition to the previous responses.

When dealing with a rather limited data amount and structured data, classic machine learning comes in handy as it allows valuable domain knowledge to be built into the model as handcrafted features, which can be helpful:

  • not only with respect to interpretability and explainability of the predictions
  • but also in terms of the ability to create powerful (but less complex) models with a fairly limited amount of data.

On the other hand: If you know that you will work with Big Data & highly unstructured and high dimensional data in e.g. computer vision (videos, images) or large language models: Deep Learning seems to be an effective tool to solve your challenges, see also this thread.

These sources might be helpful for you:

Please let us know if you have any open questions, @Muhammad_Usama_Amin.

Best regards