Increases in loss after training restart.
Posted: Sat Mar 14, 2020 2:20 pm
This question is not related to the visual results really.
Seems that after restarting training, I see the loss factor going up sharply, then coming down after a few hundred iterations. Same model, dame data.
Curious to why this is.
Would it be similar to walking away from a task, and coming back with "new eyes"? A bit of randomness when loading up the data in the beginning?