Is it possible to adjust the loss weight of the multi output layer according to the input variable value?

[정성엽]

My existing model has a multi-label output and a total of three outputs.

I created a model by modifying train_val.prototxt with a loss weight of 0.5 0.5 2.0 for each output.

The above model works fine.

I would like to compare the input variables and change the output loss weight every time to make a better model.

For example, if it is similar to input_1, input_3, it is 1: 0: 2.0, and if it is similar to input_2, input_3, it is 0: 1: 2.0.

Is there just a way to use the custom loss layer?

Is there a way to make a good custom loss layer?