Thanks for your great work!
I'm confused about the training of RL, in the code it seems you use the "accumulate_gradients_dagnn.m" to update parameters for RL, which is the same with SL stage. However, in the paper you stated that use "stochastic gradient ascent" to maximize tracking score. Can you help to explain it? Appreciate for the help!
Thanks for your great work!
I'm confused about the training of RL, in the code it seems you use the "accumulate_gradients_dagnn.m" to update parameters for RL, which is the same with SL stage. However, in the paper you stated that use "stochastic gradient ascent" to maximize tracking score. Can you help to explain it? Appreciate for the help!