I am finding that for a segmentation problem, loss decreases over time a lot. To be able to learn, I need to increase learning rate instead of decreasing. Is it normal to do this or there might be some problem with my training schedule?
Can you show the plot of your loss? I don’t think you will have to increase it, but start off with even a higher one and then gradually decrease it. You will have to show a graph of both your losses to really know.
ok, I ll generate and update.
Moved to the advanced category.
I saw the problem was because I was using small learning rate to start with.