cz4e / lyn Goto Github PK

代码的参数没有改，如下：
learning_rate = 0.01
size = 512
delta = 0.03
dL = 0.02
batch_size = 64
batch = 10
c = 3e8
Hz = 0.4e12
最后的结果是这样的，好像并没有收敛：
epoch :1 loss:8.5502 acc:0.9375
epoch :2 loss:3.3796 acc:0.9688
epoch :3 loss:3.6199 acc:0.9688
epoch :4 loss:3.8715 acc:0.9688
epoch :5 loss:4.0078 acc:0.9688
epoch :6 loss:0.0015 acc:1.0000
epoch :7 loss:0.0005 acc:1.0000
epoch :8 loss:0.0000 acc:1.0000
epoch :9 loss:4.0107 acc:0.9688
epoch :10 loss:8.0005 acc:0.9375
epoch :11 loss:4.0004 acc:0.9688
epoch :12 loss:9.9267 acc:0.9219
epoch :13 loss:7.9175 acc:0.9375
epoch :14 loss:11.9953 acc:0.9062
epoch :15 loss:7.9595 acc:0.9375
epoch :16 loss:2.0000 acc:0.9844
epoch :17 loss:10.0000 acc:0.9219
epoch :18 loss:8.0024 acc:0.9375
epoch :19 loss:5.3605 acc:0.9531
epoch :20 loss:2.0257 acc:0.9844

你好，我想请问一下经过5层物理神经网络之后，其实光会衰减很多，一般用什么样的光源能带来比较好的效果呢？

Hello Lyn,

I'm quite confused that you did not include the Rayleigh Sommerfeld in your code. And the results seem to be quite different from what is expected. May we talk further in details about this?

Rulin

Hello,
I'm confused about your definition about _propogation_phase_out, could you please explain how it works?

Thank you!
Zikang Xu

有些部分表示看不懂，方便留下联系方式咨询吗，请问code里面的u0(input data?)和theta表示什么