I had already trained the CNN with SGD, and it is training well. However, once I am training the model with Adam solver, after 100k
iteration almost, it is starting to increase the loss value. Could you please help me to interpret this?
The following shows the solver.prototxt
:
momentum: 0.99
momentum2: 0.999 #+
test_interval: 1000
test_iter: 40
weight_decay: 0.0005
base_lr: 0.0001