vlfeat / matconvnet

MatConvNet: CNNs for MATLAB
Other
1.4k stars 753 forks source link

Multi-task training #1146

Open fshamsafar opened 6 years ago

fshamsafar commented 6 years ago

Dear all,

I am training a dagg network with two objective functions, i.e. a multi-tasking network. I have some queries in this regard and I really appreciate your help.

  1. If we notate loss 1 with L1 and loss 2 with L2, is the network considering the total loss as L1+L2?
  2. Should the two loss functions reside in a similar range? One of the loss functions yields values around 100 and the other values less than 5. For example, if L1 = 120, L2 is 1.5. Do I have to take them in a similar range?
  3. If I want to give more importance to one objective function, is there a way to make the total loss function something like a L1 + b L2? Where I can make such modification?

Thanks

haoliyoupai09 commented 6 years ago

I would like to know, have you solved your problem now? Could you share me the solution? Thank you very much.