Multitask learning questions

Course Q&A Deep Learning Specialization Structuring Machine Learning Projects
1

Hi

I have some questions for multitask learning.

  1. Can we use mixed data as input in deep learning model? Like time series data and image data both as input?

  2. So I can assign multiple labels to an image in multitask but what if one label is not 0/1 kind of binary label? But instead a continuous real number based label?

3.Also, Can I use different loss functions for different features then? Like logistic loss for 0/1 kind of labels and maybe MSE for a real number inside a single model?

  1. And if answers are yes, then will this be covered in remaining 2 courses of the specialization?

Regards
Wajid

2

Hi Wajid,

  1. Can we use mixed data as input in deep learning model? Like time series data and image data both as input?
    Ans: Absolutely. Refer here for a simple demonstration using Keras.

  2. So I can assign multiple labels to an image in multitask but what if one label is not 0/1 kind of binary label? But instead a continuous real number based label?
    Ans: Basically you are now solving a regression problem instead of a classification one, but can be done without any issues.

  3. Also, Can I use different loss functions for different features then? Like logistic loss for 0/1 kind of labels and maybe MSE for a real number inside a single model?
    Ans: Just like different kinds of inputs, the model can have different kinds of outputs as well with different loss functions. Check here for an example. The same is possible with other deep learning frameworks.

  4. And if answers are yes, then will this be covered in remaining 2 courses of the specialization?
    Ans: This one I am not too sure as I have not done this specific specialization, someone else hopefully can throw a light on this.

Let me know if this helps.

2 Likes