[ad_1]
I'm using tensorflow to train MLP & CNN with AdamOptimizer as a default parameters.
I found that it works well but after some training step, it converges to fixed results.
For example, there's only two categories,
training step 0: 0.5
training step 1000:0.9
training step 2000: 0.953
..........
training step 100000: 0.99995
training step 110000: 0.5
.................
training step 200000: 0.5
after some step, it is converged to some values that indicating all label1 or all label2
What is the reason and how can I solve it?
[ad_2]
لینک منبع
