本文制作了一个三分类的网络来分类mnist数据集的0,1,2.并同时制作了一个力学模型,用来模拟这个三分类的过程,并用这个模型解释分类的原理。
上图可以用下列方程描述
只要ωx0,ωx1,ωx2,ωx012这四个数已知这个方程组是可以解的。
现在设计一个网络来计算ωx0
制作一个网络分类mnist 0和一张图片x,让左右两个网络实现参数共享,让x向1,0,0收敛,让mnist 0向0,1,0收敛
将这个网络简写成
d2(mnist 0, x)81-con(3*3)-49-30-3-(3*k) ,k∈(0,1)
具体进样顺序 |
|
||
δ=0.5 |
|
||
初始化权重 |
|
||
迭代次数 |
|
||
X |
1 |
判断是否达到收敛 |
|
mnist 0-1 |
2 |
判断是否达到收敛 |
|
梯度下降 |
|
||
X |
3 |
判断是否达到收敛 |
|
mnist 0-2 |
4 |
判断是否达到收敛 |
|
梯度下降 |
|
||
…… |
|
||
X |
9997 |
判断是否达到收敛 |
|
mnist 0-4999 |
9998 |
判断是否达到收敛 |
|
梯度下降 |
|
||
…… |
|
||
如果4999图片内没有达到收敛标准再次从头循环 |
|||
X |
9999 |
判断是否达到收敛 |
|
mnist 0-1 |
10000 |
判断是否达到收敛 |
|
梯度下降 |
|
||
…… |
|
||
每当网路达到收敛标准记录迭代次数和对应的准确率测试结果 |
|||
将这一过程重复199次 |
|
||
δ=0.01 |
|
||
… |
|
||
δ=3e-6 收敛条件是 |
|
if (Math.abs(f2[0]-y[0])< δ && Math.abs(f2[1]-y[1])< δ && Math.abs(f2[2]-y[2])< δ )
因为对应每个δ都有一个n与之对应,所以可以得到一条稳定的n(δ)曲线。让nx0等于ωx0。
用同样的办法制作另外三个网络
d2(mnist 1, x)81-con(3*3)-49-30-3-(3*k) ,k∈(0,1)
制作一个网络分类mnist 1和一张图片x,让左右两个网络实现参数共享,让x向1,0,0收敛,让mnist 1向0,1,0收敛,得到ωx1.
d2(mnist 2, x)81-con(3*3)-49-30-3-(3*k) ,k∈(0,1)
制作一个网络分类mnist 2和一张图片x,让左右两个网络实现参数共享,让x向1,0,0收敛,让mnist 2向0,0,1收敛,得到ωx2.
计算mnist0,mnist1和mnist2相对一个参照物x的频率,让他们的频率可以相互比较。
d3(mnist 0,1,2)81-con(3*3)-49-30-3-(3*k) ,k∈(0,1)
制作一个网络分类mnist 0,1,2让三个网络实现参数共享,让mnist 0向1,0,0收敛,让mnist 1向0,1,0收敛,让mnist 2 向0,0,1收敛,得到ω012.
计算ωx
所以ωx,ω0,ω1,ω2都可以被解出来。
再根据质量关系
如果假设mx=1。m0,m1,m2也都可以被解出来
具体解得的数据
δ |
ω123 |
ωx0 |
ωx1 |
ωx2 |
ωx |
ω0 |
ω1 |
ω2 |
||
0.1 |
7587.95 |
2472.46 |
2934.42 |
3140.44 |
2055.39 |
3324.81 |
#NUM! |
#NUM! |
||
1.00E-02 |
11022.2 |
3210.84 |
3790.5 |
4027.85 |
2635.91 |
4469.02 |
#NUM! |
#NUM! |
||
1.00E-03 |
18193.2 |
4364.11 |
5471.13 |
5291.57 |
3579.12 |
6091.61 |
#NUM! |
#NUM! |
||
9*1e-4 |
19061.9 |
4436.95 |
5452.92 |
5168.63 |
3569.43 |
6578.86 |
#NUM! |
#NUM! |
||
8*1e-4 |
19311.7 |
4632.25 |
5613.24 |
5432.4 |
3723.26 |
6889.13 |
#NUM! |
#NUM! |
||
7*1e-4 |
20254.4 |
4830.88 |
5765.85 |
5547.04 |
3839.11 |
7484.61 |
#NUM! |
#NUM! |
||
6*1e-4 |
21368.1 |
5128.38 |
5991.87 |
6051.89 |
4084.89 |
7877.29 |
#NUM! |
#NUM! |
||
5*1e-4 |
22162.8 |
6119.18 |
6673.56 |
6826.39 |
4711.76 |
10931.1 |
#NUM! |
#NUM! |
||
4*1e-4 |
22926.2 |
7350.78 |
7738.91 |
8741.98 |
5745.78 |
12195.5 |
17949.2 |
#NUM! |
||
3*1e-4 |
25926.6 |
9288.86 |
10599.9 |
11598.3 |
7640.9 |
12855 |
38570.7 |
#NUM! |
||
2*1e-4 |
29002.3 |
13223 |
14411.3 |
15514.8 |
10781.1 |
18781.3 |
31213.9 |
#NUM! |
||
1.00E-04 |
37411.5 |
25290.2 |
27430.9 |
29153.8 |
22370.7 |
29764.2 |
38931.9 |
53083.3 |
||
9*1e-5 |
39277 |
26452.8 |
30887 |
33227.6 |
24944.5 |
28272.6 |
45208.3 |
69954.5 |
||
8*1e-5 |
39619.2 |
30812.7 |
34556.4 |
35642 |
29511.8 |
32302.4 |
43574.7 |
48439.4 |
||
7*1e-5 |
42224.3 |
34396 |
39146.5 |
39702.1 |
34088 |
34712.5 |
47430.5 |
49492.7 |
||
6*1e-5 |
46435.3 |
39686.3 |
46285.1 |
45362.6 |
41007.1 |
38485.4 |
54320.9 |
51485.4 |
||
5*1e-5 |
45628.2 |
45913.8 |
52943.6 |
52850.1 |
56574 |
39643.5 |
49933 |
49776.4 |
||
4*1e-5 |
52063.7 |
58538.8 |
66635.4 |
62407.9 |
82495.7 |
47853.1 |
57402.7 |
52229.9 |
||
3*1e-5 |
54023.4 |
73604.5 |
86087.4 |
82569 |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
||
2*1e-5 |
65447.1 |
105869 |
121513 |
114573 |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
||
1.00E-05 |
80455.8 |
199074 |
235384 |
193000 |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
||
9*1e-6 |
85210.4 |
217783 |
255118 |
224412 |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
||
8*1e-6 |
90364.4 |
235942 |
292894 |
242478 |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
||
7*1e-6 |
92539.4 |
267342 |
312441 |
276817 |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
||
6*1e-6 |
100522 |
298437 |
370803 |
305591 |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
||
5*1e-6 |
103184 |
355890 |
424557 |
344916 |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
||
4*1e-6 |
112972 |
431037 |
509263 |
409437 |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
||
3*1e-6 |
127752 |
539694 |
641035 |
523975 |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
||
2*1e-6 |
138964 |
|
936074 |
15953.4 |
#VALUE! |
#VALUE! |
#VALUE! |
#VALUE! |
由于统计精度的问题只得到了一部分有效的点
由图ω0,ω1,ω2的曲线有可能有交点。
δ |
mx |
m0 |
m1 |
m2 |
mx0 |
mx1 |
mx2 |
m123 |
k |
0.1 |
1 |
0.38217 |
#NUM! |
#NUM! |
1.38217 |
0.98124 |
0.85672 |
0.22012 |
4224629 |
1.00E-02 |
1 |
0.34788 |
#NUM! |
#NUM! |
1.34788 |
0.96716 |
0.85653 |
0.17157 |
6947999 |
1.00E-03 |
1 |
0.34521 |
#NUM! |
#NUM! |
1.34521 |
0.85591 |
0.91498 |
0.11611 |
1.3E+07 |
9*1e-4 |
1 |
0.29437 |
#NUM! |
#NUM! |
1.29437 |
0.85698 |
0.95384 |
0.10519 |
1.3E+07 |
8*1e-4 |
1 |
0.29209 |
#NUM! |
#NUM! |
1.29209 |
0.87993 |
0.93949 |
0.11151 |
1.4E+07 |
7*1e-4 |
1 |
0.2631 |
#NUM! |
#NUM! |
1.2631 |
0.88667 |
0.95801 |
0.10778 |
1.5E+07 |
6*1e-4 |
1 |
0.26891 |
#NUM! |
#NUM! |
1.26891 |
0.92954 |
0.91119 |
0.10964 |
1.7E+07 |
5*1e-4 |
1 |
0.1858 |
#NUM! |
#NUM! |
1.1858 |
0.99697 |
0.95283 |
0.13559 |
2.2E+07 |
4*1e-4 |
1 |
0.22197 |
0.10247 |
#NUM! |
1.22197 |
1.10247 |
0.86399 |
0.18843 |
3.3E+07 |
3*1e-4 |
1 |
0.3533 |
0.03924 |
#NUM! |
1.3533 |
1.03924 |
0.86802 |
0.26057 |
5.8E+07 |
2*1e-4 |
1 |
0.32951 |
0.1193 |
#NUM! |
1.32951 |
1.1193 |
0.96574 |
0.41455 |
1.2E+08 |
1.00E-04 |
1 |
0.5649 |
0.33018 |
0.1776 |
1.5649 |
1.33018 |
1.1776 |
1.07268 |
5E+08 |
9*1e-5 |
1 |
0.77842 |
0.30445 |
0.12715 |
1.77842 |
1.30445 |
1.12715 |
1.21002 |
6.2E+08 |
8*1e-5 |
1 |
0.83468 |
0.45869 |
0.37119 |
1.83468 |
1.45869 |
1.37119 |
1.66456 |
8.7E+08 |
7*1e-5 |
1 |
0.96435 |
0.51652 |
0.47437 |
1.96435 |
1.51652 |
1.47437 |
1.95524 |
1.2E+09 |
6*1e-5 |
1 |
1.13534 |
0.56988 |
0.63438 |
2.13534 |
1.56988 |
1.63438 |
2.33961 |
1.7E+09 |
5*1e-5 |
1 |
2.03652 |
1.28369 |
1.29178 |
3.03652 |
2.28369 |
2.29178 |
4.61198 |
3.2E+09 |
4*1e-5 |
1 |
2.97196 |
2.06537 |
2.49473 |
3.97196 |
3.06537 |
3.49473 |
7.53207 |
6.8E+09 |
3*1e-5 |
1 |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
2*1e-5 |
1 |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
1.00E-05 |
1 |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
9*1e-6 |
1 |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
8*1e-6 |
1 |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
7*1e-6 |
1 |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
6*1e-6 |
1 |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
5*1e-6 |
1 |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
4*1e-6 |
1 |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
3*1e-6 |
1 |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
#NUM! |
2*1e-6 |
1 |
#VALUE! |
#VALUE! |
#VALUE! |
#VALUE! |
#VALUE! |
#VALUE! |
#VALUE! |
#VALUE! |
1.00E-06 |
1 |
#VALUE! |
#VALUE! |
#VALUE! |
#VALUE! |
#VALUE! |
#VALUE! |
#VALUE! |
#VALUE! |
1.00E-07 |
1 |
#VALUE! |
#VALUE! |
#VALUE! |
#VALUE! |
#VALUE! |
#VALUE! |
#VALUE! |
#VALUE! |
这条质量曲线里m1和m2的值也纠缠在一起。
虽然频率和质量的曲线从图上看都有纠缠但,这几个纠缠的点不完全重合,如果从频率和质量两个维度去分类仍然可能将这几个对象区分开。
这个实验虽然只得到了7组有价值的数据,但也证实了这种模拟是可能的。
实验数据 |
学习率 0.1 |
权重初始化方式 |
Random rand1 =new Random(); |
int ti1=rand1.nextInt(98)+1; |
int xx=1; |
if(ti1%2==0) |
{ xx=-1;} |
tw[a][b]=xx*((double)ti1/x); |
第一层第二层和卷积核的权重的初始化的x分别为1000,1000,200 |
d2(mnist 0, x)81-con(3*3)-49-30-3-(3*k) ,k∈(0,1)的数据
x0---3k |
|||||||||
f2[0] |
f2[1] |
f2[2] |
迭代次数n |
平均准确率p-ave |
δ |
耗时ms/次 |
耗时ms/199次 |
耗时min/199次 |
最大准确率p-max |
0.498028 |
0.500578 |
0.267648 |
17.9196 |
0.503767 |
0.5 |
806.3417 |
160464 |
2.6744 |
0.822695 |
0.609062 |
0.391232 |
0.001381 |
1721.543 |
0.5064 |
0.4 |
1103.322 |
219561 |
3.65935 |
0.969267 |
0.715655 |
0.284307 |
0.001383 |
1913.513 |
0.673654 |
0.3 |
1133.392 |
225547 |
3.759117 |
0.997636 |
0.819332 |
0.180886 |
0.001309 |
2203.241 |
0.753759 |
0.2 |
287.1608 |
57146 |
0.952433 |
0.997636 |
0.913426 |
0.086324 |
0.001203 |
2472.457 |
0.785112 |
0.1 |
1594.548 |
317330 |
5.288833 |
0.998109 |
0.992089 |
0.007913 |
9.45E-04 |
3210.839 |
0.75202 |
0.01 |
1374.216 |
273486 |
4.5581 |
0.997636 |
0.999266 |
7.34E-04 |
6.94E-04 |
4364.106 |
0.697801 |
0.001 |
1594.709 |
317364 |
5.2894 |
0.997636 |
0.999359 |
6.42E-04 |
6.69E-04 |
4436.95 |
0.71783 |
9.00E-04 |
1835.452 |
365264 |
6.087733 |
0.99669 |
0.999431 |
5.69E-04 |
6.39E-04 |
4632.246 |
0.690243 |
8.00E-04 |
2244.447 |
446649 |
7.44415 |
0.99669 |
0.999527 |
4.74E-04 |
5.96E-04 |
4830.879 |
0.677335 |
7.00E-04 |
2293.94 |
456497 |
7.608283 |
0.995272 |
0.999606 |
3.95E-04 |
5.60E-04 |
5128.377 |
0.722364 |
6.00E-04 |
2407.397 |
479072 |
7.984533 |
0.995745 |
0.999698 |
3.02E-04 |
4.79E-04 |
6119.181 |
0.739387 |
5.00E-04 |
2638.276 |
525025 |
8.750417 |
0.997636 |
0.999792 |
2.08E-04 |
3.92E-04 |
7350.779 |
0.781313 |
4.00E-04 |
2925.864 |
582255 |
9.70425 |
0.998109 |
0.99985 |
1.50E-04 |
2.96E-04 |
9288.859 |
0.773073 |
3.00E-04 |
3337.955 |
664269 |
11.07115 |
0.997163 |
0.999908 |
9.19E-05 |
1.98E-04 |
13223.02 |
0.806277 |
2.00E-04 |
3833.729 |
762927 |
12.71545 |
0.997163 |
0.994943 |
0.005057 |
9.89E-05 |
25290.15 |
0.806339 |
1.00E-04 |
5484.06 |
1091329 |
18.18882 |
0.997636 |
0.999972 |
2.78E-05 |
8.91E-05 |
26452.81 |
0.785402 |
9.00E-05 |
5786.945 |
1151603 |
19.19338 |
0.997636 |
0.994949 |
0.005051 |
7.92E-05 |
30812.7 |
0.752778 |
8.00E-05 |
5980.07 |
1190035 |
19.83392 |
0.998582 |
0.994954 |
0.005046 |
6.92E-05 |
34396 |
0.774002 |
7.00E-05 |
8870.95 |
1765335 |
29.42225 |
0.998109 |
0.994957 |
0.005043 |
5.94E-05 |
39686.29 |
0.775178 |
6.00E-05 |
9581.141 |
1906647 |
31.77745 |
0.998582 |
0.994962 |
0.005038 |
4.94E-05 |
45913.83 |
0.764458 |
5.00E-05 |
11592.85 |
2306978 |
38.44963 |
0.996217 |
0.98994 |
0.01006 |
3.96E-05 |
58538.81 |
0.785692 |
4.00E-05 |
14235.63 |
2832906 |
47.2151 |
0.997636 |
0.989942 |
0.010058 |
2.96E-05 |
73604.5 |
0.74317 |
3.00E-05 |
18005.08 |
3583011 |
59.71685 |
0.99669 |
0.984921 |
0.015079 |
1.97E-05 |
105868.7 |
0.728406 |
2.00E-05 |
24788.76 |
4932971 |
82.21618 |
0.998582 |
0.994972 |
0.005028 |
9.89E-06 |
199073.6 |
0.710966 |
1.00E-05 |
45093.17 |
8973556 |
149.5593 |
0.995745 |
0.979897 |
0.020103 |
8.91E-06 |
217782.7 |
0.707915 |
9.00E-06 |
50139.76 |
9977813 |
166.2969 |
0.998109 |
0.989948 |
0.010052 |
7.92E-06 |
235942 |
0.692995 |
8.00E-06 |
45157.56 |
8986357 |
149.7726 |
0.992435 |
0.994973 |
0.005027 |
6.93E-06 |
267341.6 |
0.700398 |
7.00E-06 |
60661.26 |
12071590 |
201.1932 |
0.997636 |
0.999999 |
1.47E-06 |
5.94E-06 |
298437.4 |
0.694805 |
6.00E-06 |
65289.42 |
12992611 |
216.5435 |
0.994799 |
0.999999 |
1.04E-06 |
4.96E-06 |
355890.1 |
0.682433 |
5.00E-06 |
75109.63 |
14946820 |
249.1137 |
0.98487 |
0.999999 |
8.47E-07 |
3.97E-06 |
431037.4 |
0.668444 |
4.00E-06 |
82579.55 |
16433350 |
273.8892 |
0.996217 |
0.994974 |
0.005026 |
2.98E-06 |
539693.7 |
0.692377 |
3.00E-06 |
116352.7 |
23154181 |
385.903 |
0.987234 |
d2(mnist 1, x)81-con(3*3)-49-30-3-(3*k) ,k∈(0,1)的数据
x1---3k |
|||||||||
f2[0] |
f2[1] |
f2[2] |
迭代次数n |
平均准确率p-ave |
δ |
耗时ms/次 |
耗时ms/199次 |
耗时min/199次 |
最大准确率p-max |
0.494146 |
0.50332 |
0.300855 |
16.51759 |
0.519529 |
0.5 |
1215.678 |
241922 |
4.032033 |
0.899291 |
0.391208 |
0.608624 |
0.001349 |
2005.116 |
0.870722 |
0.4 |
1676.628 |
333650 |
5.560833 |
0.986761 |
0.285317 |
0.714582 |
0.001306 |
2308.02 |
0.885643 |
0.3 |
1746.588 |
347586 |
5.7931 |
0.986761 |
0.18278 |
0.817472 |
0.001167 |
2732.171 |
0.863255 |
0.2 |
1843.241 |
366837 |
6.11395 |
0.982033 |
0.085759 |
0.914178 |
0.00114 |
2934.422 |
0.860484 |
0.1 |
1889 |
375928 |
6.265467 |
0.989125 |
0.007483 |
0.992533 |
8.83E-04 |
3790.503 |
0.850812 |
0.01 |
2082.854 |
414490 |
6.908167 |
0.991017 |
7.48E-04 |
0.999251 |
5.74E-04 |
5471.126 |
0.821474 |
0.001 |
2046.261 |
407207 |
6.786783 |
0.988652 |
6.41E-04 |
0.999359 |
5.69E-04 |
5452.925 |
0.824933 |
9.00E-04 |
2477.819 |
493102 |
8.218367 |
0.98818 |
5.94E-04 |
0.999407 |
5.58E-04 |
5613.236 |
0.831267 |
8.00E-04 |
2513.638 |
500214 |
8.3369 |
0.987707 |
5.11E-04 |
0.999488 |
5.46E-04 |
5765.849 |
0.823774 |
7.00E-04 |
2560.799 |
509614 |
8.493567 |
0.990071 |
4.02E-04 |
0.999598 |
5.18E-04 |
5991.869 |
0.821353 |
6.00E-04 |
2605.503 |
518496 |
8.6416 |
0.990544 |
3.24E-04 |
0.999677 |
4.49E-04 |
6673.558 |
0.812816 |
5.00E-04 |
2740.698 |
545400 |
9.09 |
0.993853 |
2.49E-04 |
0.999751 |
3.80E-04 |
7738.915 |
0.839512 |
4.00E-04 |
2977.513 |
592525 |
9.875417 |
0.991962 |
1.54E-04 |
0.999846 |
2.89E-04 |
10599.87 |
0.826991 |
3.00E-04 |
3172.347 |
631297 |
10.52162 |
0.991962 |
9.67E-05 |
0.999903 |
1.96E-04 |
14411.34 |
0.801666 |
2.00E-04 |
4643.402 |
924037 |
15.40062 |
0.991017 |
4.43E-05 |
0.999956 |
9.90E-05 |
27430.87 |
0.817276 |
1.00E-04 |
7745.357 |
1541326 |
25.68877 |
0.990544 |
0.005061 |
0.994939 |
8.88E-05 |
30887.03 |
0.823458 |
9.00E-05 |
8030.497 |
1598069 |
26.63448 |
0.993381 |
0.010082 |
0.989918 |
7.89E-05 |
34556.39 |
0.813407 |
8.00E-05 |
9136.382 |
1818140 |
30.30233 |
0.992435 |
2.54E-05 |
0.999975 |
6.92E-05 |
39146.48 |
0.793236 |
7.00E-05 |
9865.337 |
1963209 |
32.72015 |
0.992908 |
0.010071 |
0.989929 |
5.92E-05 |
46285.07 |
0.796964 |
6.00E-05 |
11820.8 |
2352343 |
39.20572 |
0.991962 |
0.010067 |
0.989933 |
4.92E-05 |
52943.62 |
0.78607 |
5.00E-05 |
13333.96 |
2653462 |
44.22437 |
0.991489 |
0.020113 |
0.979887 |
3.94E-05 |
66635.38 |
0.786324 |
4.00E-05 |
12621.33 |
2511660 |
41.861 |
0.992435 |
0.010062 |
0.989938 |
2.96E-05 |
86087.45 |
0.765207 |
3.00E-05 |
16507.18 |
3284929 |
54.74882 |
0.991017 |
6.84E-06 |
0.999993 |
1.97E-05 |
121512.8 |
0.766105 |
2.00E-05 |
23216.28 |
4620054 |
77.0009 |
0.990071 |
0.020104 |
0.979896 |
9.88E-06 |
235383.6 |
0.757124 |
1.00E-05 |
45063.69 |
8967677 |
149.4613 |
0.991962 |
0.005028 |
0.994972 |
8.91E-06 |
255118.4 |
0.734806 |
9.00E-06 |
48584.32 |
9668280 |
161.138 |
0.990071 |
2.59E-06 |
0.999997 |
7.91E-06 |
292894 |
0.745213 |
8.00E-06 |
67311.87 |
13395064 |
223.2511 |
0.989598 |
0.010052 |
0.989948 |
6.92E-06 |
312441.3 |
0.73532 |
7.00E-06 |
52103.28 |
10368570 |
172.8095 |
0.986288 |
0.005027 |
0.994973 |
5.95E-06 |
370802.8 |
0.729998 |
6.00E-06 |
62282.38 |
12394211 |
206.5702 |
0.988652 |
0.010052 |
0.989948 |
4.96E-06 |
424557.5 |
0.738698 |
5.00E-06 |
73037.1 |
14534383 |
242.2397 |
0.986761 |
1.06E-06 |
0.999999 |
3.96E-06 |
509263.2 |
0.719527 |
4.00E-06 |
83283.84 |
16573502 |
276.225 |
0.992908 |
7.79E-07 |
0.999999 |
2.97E-06 |
641035.2 |
0.72789 |
3.00E-06 |
113768.1 |
22639857 |
377.331 |
0.98818 |
5.31E-07 |
0.999999 |
1.99E-06 |
936074.1 |
0.704451 |
2.00E-06 |
162290.3 |
32295803 |
538.2634 |
0.985343 |
d2(mnist 2, x)81-con(3*3)-49-30-3-(3*k) ,k∈(0,1)的数据
x2 |
|||||||||
f2[0] |
f2[1] |
f2[2] |
迭代次数n |
平均准确率p-ave |
δ |
耗时ms/次 |
耗时ms/199次 |
耗时min/199次 |
最大准确率p-max |
0.500087 |
0.239966 |
0.498985 |
21.40201 |
0.503132 |
0.5 |
749.8894 |
149243 |
2.487383 |
0.719682 |
0.390571 |
0.001166 |
0.609555 |
2119.417 |
0.67279 |
0.4 |
1158.015 |
230477 |
3.841283 |
0.919483 |
0.283649 |
0.001122 |
0.716415 |
2520.121 |
0.691709 |
0.3 |
1231.538 |
245076 |
4.0846 |
0.909046 |
0.182927 |
0.001091 |
0.817045 |
2697.608 |
0.70339 |
0.2 |
1264.744 |
251699 |
4.194983 |
0.916004 |
0.085491 |
0.001024 |
0.914714 |
3140.442 |
0.689823 |
0.1 |
1359.94 |
270631 |
4.510517 |
0.904573 |
0.007564 |
8.38E-04 |
0.992453 |
4027.849 |
0.66199 |
0.01 |
1514.744 |
301434 |
5.0239 |
0.919483 |
7.01E-04 |
6.68E-04 |
0.999298 |
5291.568 |
0.634952 |
0.001 |
1744.804 |
347216 |
5.786933 |
0.932406 |
6.23E-04 |
6.64E-04 |
0.999376 |
5168.633 |
0.627004 |
9.00E-04 |
1730.447 |
344375 |
5.739583 |
0.917495 |
5.42E-04 |
6.44E-04 |
0.999459 |
5432.402 |
0.657717 |
8.00E-04 |
1818.317 |
361846 |
6.030767 |
0.913519 |
4.73E-04 |
6.17E-04 |
0.999527 |
5547.035 |
0.644328 |
7.00E-04 |
1827.683 |
363710 |
6.061833 |
0.920477 |
3.61E-04 |
5.52E-04 |
0.999639 |
6051.889 |
0.633463 |
6.00E-04 |
1936.814 |
385426 |
6.423767 |
0.912028 |
2.64E-04 |
4.81E-04 |
0.999736 |
6826.392 |
0.626565 |
5.00E-04 |
1777.729 |
353783 |
5.896383 |
0.925447 |
1.91E-04 |
3.92E-04 |
0.999809 |
8741.98 |
0.630933 |
4.00E-04 |
2531.266 |
503722 |
8.395367 |
0.902584 |
1.25E-04 |
2.94E-04 |
0.999875 |
11598.28 |
0.618857 |
3.00E-04 |
3013.085 |
599619 |
9.99365 |
0.92495 |
6.72E-05 |
1.97E-04 |
0.999933 |
15514.78 |
0.584823 |
2.00E-04 |
3689.241 |
734159 |
12.23598 |
0.862326 |
0.00505 |
9.83E-05 |
0.99495 |
29153.83 |
0.565014 |
1.00E-04 |
6322.709 |
1258219 |
20.97032 |
0.778827 |
0.015097 |
8.82E-05 |
0.984903 |
33227.56 |
0.568439 |
9.00E-05 |
7009.352 |
1394878 |
23.24797 |
0.779821 |
0.010068 |
7.85E-05 |
0.989932 |
35641.99 |
0.57249 |
8.00E-05 |
8048.271 |
1601608 |
26.69347 |
0.903082 |
0.01509 |
6.88E-05 |
0.98491 |
39702.06 |
0.562949 |
7.00E-05 |
9952.528 |
1980553 |
33.00922 |
0.807157 |
0.025139 |
5.90E-05 |
0.974861 |
45362.58 |
0.558868 |
6.00E-05 |
11509.38 |
2290367 |
38.17278 |
0.878231 |
1.07E-05 |
4.91E-05 |
0.999989 |
52850.1 |
0.545721 |
5.00E-05 |
13502.02 |
2686904 |
44.78173 |
0.705765 |
0.010059 |
3.93E-05 |
0.989941 |
62407.9 |
0.544195 |
4.00E-05 |
15279.87 |
3040697 |
50.67828 |
0.693837 |
0.01508 |
2.95E-05 |
0.98492 |
82569.04 |
0.542946 |
3.00E-05 |
20272.58 |
4034248 |
67.23747 |
0.723658 |
0.015079 |
1.96E-05 |
0.984921 |
114572.7 |
0.537536 |
2.00E-05 |
19727.65 |
3925806 |
65.4301 |
0.783797 |
0.005027 |
9.84E-06 |
0.994973 |
193000.5 |
0.531015 |
1.00E-05 |
33317.63 |
6630210 |
110.5035 |
0.709742 |
0.010052 |
8.86E-06 |
0.989948 |
224411.6 |
0.530356 |
9.00E-06 |
37654.61 |
7493268 |
124.8878 |
0.663022 |
0.020102 |
7.87E-06 |
0.979898 |
242478.2 |
0.528997 |
8.00E-06 |
42188.21 |
8395457 |
139.9243 |
0.756461 |
1.29E-06 |
6.91E-06 |
0.999999 |
276817.3 |
0.529656 |
7.00E-06 |
47674.84 |
9487293 |
158.1216 |
0.785288 |
0.010051 |
5.92E-06 |
0.989949 |
305591.4 |
0.527146 |
6.00E-06 |
51288.63 |
10206438 |
170.1073 |
0.639165 |
0.005026 |
4.94E-06 |
0.994974 |
344915.9 |
0.524828 |
5.00E-06 |
59820.35 |
11904251 |
198.4042 |
0.604374 |
0.005026 |
3.95E-06 |
0.994974 |
409437.4 |
0.526517 |
4.00E-06 |
70332.83 |
13996238 |
233.2706 |
0.708748 |
4.95E-07 |
2.97E-06 |
1 |
523975 |
0.522715 |
3.00E-06 |
97274.69 |
19357665 |
322.6278 |
0.58002 |
d3(mnist 0,1,2)81-con(3*3)-49-30-3-(3*k) ,k∈(0,1)的数据
已经在《计算一个网络准确率达到99.9%的时间和需要的迭代次数---验证实例三分类minst0,1,2》给出了
d2(mnist 0, x)81-con(3*3)-49-30-3-(3*k) ,k∈(0,1)
d2(mnist 1, x)81-con(3*3)-49-30-3-(3*k) ,k∈(0,1)
d2(mnist 2, x)81-con(3*3)-49-30-3-(3*k) ,k∈(0,1)
的具体数据比较多有感兴趣的朋友可以到我的资源里下载。