卷积核一定可以提升网络性能吗？-分类0，2

制作一个网络然后向这个网络上加1-9个卷积核，通过网络的分辨准确率来比较卷积核对网络性能的影响。

网络的结构是

(mnist 0 ,mnist2)81-con(3*3)*n-(49*n)-30-2-(1,0) || (0,1)

分类mnist的0和2，将28*28的图片压缩到9*9，向这个网络增加n个3*3的卷积核，网络隐藏层的节点数是30个，让0向（1，0）收敛，让2向（0，1）收敛。让n分别等于0-9.当n=0时，网络相当于一个三层的网络

(mnist 0 ,mnist2)81-30-2-(1,0) || (0,1)

用这个没有卷积核的3层网络作为参照比较网络性能的变化。

网络迭代的停止标准是：

|网络的输出值-目标函数|<δ,

让δ分别等于0.5到1e-6的34个值每个δ收敛199次，通过每次收敛的准确率的平均值和199次收敛的最大值来比较性能的变化。一共收敛了9*34*199次。

平均性能的表格

		1	2	3	4	5	6	7	8	9
δ	平均准确率p-ave	平均准确率p-ave	平均准确率p-ave	平均准确率p-ave	平均准确率p-ave	平均准确率p-ave	平均准确率p-ave	平均准确率p-ave	平均准确率p-ave	平均准确率p-ave
0.5	0.5523392	0.5039786	0.502023	0.5121532	0.5091611	0.5093609	0.5166738	0.5169935	0.5132447	0.5188143
0.4	0.9516094	0.7559143	0.7354191	0.7508267	0.6995165	0.6534412	0.6503317	0.6558139	0.6173537	0.649505
0.3	0.9548188	0.8792771	0.8562769	0.8324675	0.8165954	0.7935727	0.7344676	0.7343227	0.7617811	0.7652077
0.2	0.9455553	0.9126847	0.8743069	0.8599209	0.8596436	0.8487367	0.8041924	0.7903484	0.7557494	0.7660594
0.1	0.9601137	0.9143681	0.8736151	0.8528228	0.8806283	0.8796717	0.8396106	0.7770588	0.7376894	0.6889642
0.01	0.9393064	0.9245307	0.8586921	0.7835649	0.7043043	0.7821613	0.8381045	0.8263509	0.8017723	0.7628475
0.001	0.9764878	0.9358697	0.9156918	0.7677628	0.677308	0.6918139	0.6827552	0.6992168	0.6397644	0.6579093
9.00E-04	0.976268	0.9418714	0.9082715	0.7771262	0.6896435	0.6739163	0.6619629	0.6649675	0.6600498	0.6836219
8.00E-04	0.9763554	0.9388343	0.9299529	0.7895916	0.6767286	0.6480064	0.6607965	0.6492677	0.6606017	0.686042
7.00E-04	0.9764379	0.937663	0.9266137	0.8038253	0.7073439	0.6674051	0.6691085	0.6738489	0.6938295	0.748409
6.00E-04	0.9775193	0.9359047	0.9417115	0.842268	0.7357039	0.6842812	0.6739413	0.686047	0.7140099	0.772683
5.00E-04	0.9785883	0.9323032	0.9336768	0.8814125	0.7663092	0.7402345	0.7195346	0.761384	0.795848	0.8459095
4.00E-04	0.9786957	0.9397759	0.9376979	0.9120353	0.8389612	0.8006309	0.7972392	0.8230716	0.8615768	0.8792496
3.00E-04	0.9787057	0.9376355	0.9369512	0.9286942	0.8930937	0.8779609	0.8737425	0.9023822	0.9043228	0.9192658
2.00E-04	0.9807637	0.9394288	0.9160514	0.9294584	0.9271831	0.9271157	0.9328826	0.9386095	0.9462971	0.9588349
1.00E-04	0.980931	0.9435847	0.9207244	0.9284419	0.9477307	0.9556655	0.9600662	0.9628136	0.9671569	0.970661
9.00E-05	0.980409	0.9463745	0.9198777	0.9400207	0.949439	0.9538298	0.9648616	0.961882	0.9693647	0.9697219
8.00E-05	0.9805963	0.9482127	0.9346559	0.9425932	0.9509601	0.9606856	0.9629584	0.9673242	0.968031	0.9698842
7.00E-05	0.9805189	0.9511449	0.9402679	0.9501708	0.957721	0.9614924	0.9692398	0.9690101	0.9715276	0.9699816
6.00E-05	0.9813631	0.9534327	0.9404153	0.9503007	0.962876	0.9671019	0.9707534	0.9715576	0.9721071	0.9730986
5.00E-05	0.9820649	0.9538448	0.9515645	0.9591546	0.9651638	0.9709657	0.9717874	0.973698	0.9731785	0.9733808
4.00E-05	0.9827467	0.953645	0.9557554	0.9651563	0.9691499	0.9737555	0.9740926	0.9746471	0.9739852	0.975219
3.00E-05	0.9830564	0.9588249	0.964387	0.9707584	0.9724492	0.9763055	0.9759883	0.9759009	0.9736655	0.9752915
2.00E-05	0.983526	0.9624015	0.9721995	0.9748769	0.975776	0.9766801	0.9769274	0.9769873	0.9756062	0.9740602
1.00E-05	0.9821723	0.9718698	0.9794749	0.977799	0.9770598	0.9779814	0.9764878	0.9773944	0.9761531	0.974707
9.00E-06	0.9822897	0.9740377	0.9801942	0.9780488	0.9786107	0.9772221	0.9770198	0.9759833	0.9749343	0.9747445
8.00E-06	0.9825819	0.9755562	0.978888	0.9779913	0.9770023	0.9771671	0.9770847	0.9765203	0.9760507	0.9765053
7.00E-06	0.9826743	0.9759308	0.9779514	0.9788755	0.9774069	0.9771871	0.9777766	0.9768724	0.977859	0.9755887
6.00E-06	0.9826368	0.9772346	0.9790528	0.9789454	0.9779214	0.9770747	0.9771497	0.9769199	0.9764478	0.9761681
5.00E-06	0.9826618	0.9778415	0.9786532	0.9784184	0.9771247	0.978346	0.9778215	0.9772096	0.9766701	0.9759308
4.00E-06	0.9829016	0.9796822	0.9805738	0.9779464	0.9781812	0.9785108	0.9785833	0.9789554	0.9774369	0.9764828
3.00E-06	0.9834086	0.9801243	0.9796522	0.9786957	0.9785083	0.9790953	0.979405	0.9794799	0.9787356	0.9779764
2.00E-06	0.9839805	0.9801667	0.9800768	0.9799894	0.9800469	0.9802292	0.9806188	0.9806488	0.9797047	0.9782211
1.00E-06	0.9850945	0.9819625	0.9821623	0.9810084	0.9816928	0.9825095	0.98199	0.9821398	0.9816953	0.9817527

没有加卷积核的3层网络的平均准确率大于任何加卷积核的网络，也就是增加卷积核对这个网络的平均性能没有任何正面价值。平均性能随卷积核变化的一个大致的规律，有2个卷积核的网络的平均性能最大，当卷积核数量超过2个以后随着卷积核数量的增加网络的平均性能有明显的下降趋势。

按照平均性能排序如下

81*30*2>2>3>4>5>6>7>8>9>1

 

再比较网络的最大性能的变化

		1	2	3	4	5	6	7	8	9
δ	最大值p-max	最大值p-max	最大值p-max	最大值p-max	最大值p-max	最大值p-max	最大值p-max	最大值p-max	最大值p-max	最大值p-max
0.5	0.9284294	0.8489066	0.7897614	0.7659046	0.7997018	0.888171	0.8533797	0.833499	0.8136183	0.8210736
0.4	0.9671968	0.97167	0.9771372	0.972664	0.972664	0.9741551	0.9741551	0.9741551	0.9761431	0.9736581
0.3	0.9686879	0.971173	0.972664	0.972167	0.9681909	0.973161	0.97167	0.9686879	0.973161	0.973161
0.2	0.9627237	0.9756461	0.971173	0.9741551	0.9736581	0.9746521	0.9736581	0.972664	0.9736581	0.97167
0.1	0.9706759	0.9691849	0.9786282	0.97167	0.9761431	0.9766402	0.9756461	0.9756461	0.9681909	0.9671968
0.01	0.9776342	0.9751491	0.9761431	0.9776342	0.9761431	0.9706759	0.972167	0.9756461	0.9736581	0.972167
0.001	0.9816103	0.9796223	0.9801193	0.9741551	0.9776342	0.9786282	0.9791252	0.9826044	0.9806163	0.9801193
9.00E-04	0.9806163	0.9786282	0.9791252	0.9751491	0.9771372	0.9781312	0.9811133	0.9806163	0.9806163	0.9791252
8.00E-04	0.9816103	0.9796223	0.9811133	0.9791252	0.9746521	0.9801193	0.9786282	0.9806163	0.9806163	0.9816103
7.00E-04	0.9816103	0.9811133	0.9786282	0.9781312	0.9796223	0.9771372	0.9781312	0.9796223	0.9801193	0.9831014
6.00E-04	0.9816103	0.9791252	0.9806163	0.9796223	0.9771372	0.9781312	0.972167	0.9791252	0.9806163	0.9801193
5.00E-04	0.9816103	0.9791252	0.9811133	0.9811133	0.9811133	0.9796223	0.9776342	0.9816103	0.9831014	0.9845924
4.00E-04	0.9831014	0.9811133	0.9791252	0.9801193	0.9796223	0.9811133	0.9791252	0.9816103	0.9826044	0.9840954
3.00E-04	0.9835984	0.9786282	0.9826044	0.9811133	0.9806163	0.9835984	0.9831014	0.9821074	0.9840954	0.9845924
2.00E-04	0.9831014	0.9845924	0.9840954	0.9811133	0.9831014	0.9855865	0.9850895	0.9850895	0.9865805	0.9860835
1.00E-04	0.9831014	0.9835984	0.9835984	0.9840954	0.9845924	0.9865805	0.9860835	0.9875746	0.9875746	0.9880716
9.00E-05	0.9845924	0.9816103	0.9826044	0.9860835	0.9860835	0.9860835	0.9870775	0.9875746	0.9865805	0.9875746
8.00E-05	0.9845924	0.9835984	0.9845924	0.9855865	0.9865805	0.9870775	0.9870775	0.9875746	0.9860835	0.9865805
7.00E-05	0.9845924	0.9821074	0.9831014	0.9855865	0.9865805	0.9875746	0.9875746	0.9870775	0.9870775	0.9875746
6.00E-05	0.9845924	0.9835984	0.9845924	0.9855865	0.9870775	0.9865805	0.9875746	0.9880716	0.9865805	0.9875746
5.00E-05	0.9850895	0.9826044	0.9855865	0.9865805	0.9880716	0.9875746	0.9895626	0.9875746	0.9875746	0.9870775
4.00E-05	0.9850895	0.9826044	0.9855865	0.9875746	0.9885686	0.9875746	0.9880716	0.9890656	0.9875746	0.9875746
3.00E-05	0.9855865	0.9885686	0.9880716	0.9875746	0.9880716	0.9880716	0.9880716	0.9885686	0.9880716	0.9880716
2.00E-05	0.9855865	0.9880716	0.9875746	0.9875746	0.9900596	0.9875746	0.9875746	0.9890656	0.9890656	0.9885686
1.00E-05	0.9855865	0.9875746	0.9880716	0.9895626	0.9875746	0.9885686	0.9885686	0.9895626	0.9890656	0.9885686
9.00E-06	0.9850895	0.9870775	0.9885686	0.9880716	0.9885686	0.9890656	0.9885686	0.9885686	0.9895626	0.9885686
8.00E-06	0.9850895	0.9865805	0.9875746	0.9885686	0.9895626	0.9885686	0.9885686	0.9885686	0.9890656	0.9890656
7.00E-06	0.9845924	0.9875746	0.9885686	0.9880716	0.9885686	0.9880716	0.9885686	0.9880716	0.9880716	0.9885686
6.00E-06	0.9845924	0.9880716	0.9885686	0.9885686	0.9880716	0.9880716	0.9885686	0.9885686	0.9885686	0.9900596
5.00E-06	0.9845924	0.9875746	0.9880716	0.9895626	0.9890656	0.9880716	0.9900596	0.9890656	0.9885686	0.9885686
4.00E-06	0.9845924	0.9880716	0.9885686	0.9890656	0.9890656	0.9880716	0.9895626	0.9895626	0.9885686	0.9885686
3.00E-06	0.9855865	0.9875746	0.9880716	0.9890656	0.9880716	0.9885686	0.9895626	0.9880716	0.9890656	0.9900596
2.00E-06	0.9870775	0.9875746	0.9885686	0.9890656	0.9885686	0.9890656	0.9885686	0.9895626	0.9875746	0.9890656
1.00E-06	0.9875746	0.9890656	0.9900596	0.9905567	0.9895626	0.9915507	0.9885686	0.9895626	0.9895626	0.9890656

虽然增加卷积核对网路的平均性能没有正面价值，但有卷积核网络的最大性能都好于原始的3层网络，表明卷积核对网络的最大性能有显著提升。

 

用收敛标准=1e-6的值排序

5>3>2>4=7=8>1=9>6>81*30*2

 

加上前面做的0-1，0-5，0-6的实验已经做了4组，将所有的数据总结成表格

平均性能	0-1	81*30*2>2>3>4>5>6>7>8>9>1			没有正面价值
	0-5	2>3>4>5>6>7>8>9>81-30-2>1			明显提升
	0-6	81*30*2>2>9>8>1>7>6>5>4>3			没有正面价值
	0-2	81*30*2>2>3>4>5>6>7>8>9>1			没有正面价值
最大性能	0-1	81*30*2=1=2=3=4=5=6=7=8=9			没有提升
	0-5	2>3>4>5>6>7>8>9>1>81-30-2			提升明显
	0-6	81*30*2=2=7>8>1=3=4>6>5			无明显提升
	0-2	5>3>2>4=7=8>1=9>6>81*30*2			提升明显

对网络平均性能有提升的只有0-5，对网络最大性能有明显提升的只有0-5和0-2.而对于0-1和0-6.增加卷积核对网络的平均性能和最大性能都没有任何正面价值。表明卷积核对网络性能的影响不是绝对的，与测试集有很大的关系。并不是所有的测试集都适于加卷积核。

比较迭代次数

		1	2	3	4	5	6	7	8	9
δ	迭代次数n	迭代次数n	迭代次数n	迭代次数n	迭代次数n	迭代次数n	迭代次数n	迭代次数n	迭代次数n	迭代次数n
0.5	8.8643216	16.246231	13.809045	14.904523	15.175879	12.135678	11.834171	11.924623	13.065327	11.130653
0.4	208.9397	1685.9246	1134.1307	882.45729	751.93467	664.0804	594.83417	526.31658	481.18593	445.8191
0.3	268.72864	1923.0955	1299.8291	1047.608	869.26633	788.86935	699.34171	648.02513	593.94472	548.1809
0.2	324.01508	2120.0452	1440.6281	1160.9749	988.66332	872.42714	790.13568	712.67839	664.07538	630.42211
0.1	410.58794	2318.6683	1637.1307	1291.804	1105.392	971.68342	880.61307	814.58794	764.27136	718.45729
0.01	685.54774	2832.7236	2076.397	1656.4171	1424.8995	1282.6281	1206.7035	1101.2513	1039.4724	999.18593
0.001	1447.1307	3803.3216	2899.7387	2436.8945	2234.598	2142.9648	2086.7085	2070.9849	2123.7638	2199.7889
9.00E-04	1450.9146	3783.5276	2924.2462	2495.8342	2291.7538	2173.5628	2137.2211	2139.3467	2256.804	2365.5226
8.00E-04	1494.9246	3831.3266	3005.1759	2545.6231	2314.8342	2233.1558	2231.5327	2247.4975	2312.9698	2508.6784
7.00E-04	1552.7136	3962.8894	3032.8241	2621.9497	2432.1558	2312.2362	2360.9447	2437.2261	2532.6633	3025.8291
6.00E-04	1711.6281	4070.799	3151.9698	2738.2462	2520.1809	2427.5578	2559.2161	2598.5427	2897.8191	3426.1709
5.00E-04	1968.7136	4228.1005	3268.7487	2908.4422	2656.392	2614.2312	2638.9397	3189.1759	3705.1457	4812.1407
4.00E-04	2165.0553	4379.6533	3413.5879	3049.1709	2878.2161	2938.7839	3379.3819	4073.3719	5242.4171	7248.4221
3.00E-04	2324.0804	4616.3869	3568.7487	3236.6734	3262.6985	3474.1357	4279.6533	6453.8191	8586.4523	11609.673
2.00E-04	2720.8442	5050.8744	4034.0854	3941.7839	4741.608	5935.9347	8696.7739	11151.116	15180.141	19449.472
1.00E-04	3165.7085	6349.6985	5698.9447	6692.2211	9883.7035	14487.849	20128.899	26472.593	31021.317	34488.161
9.00E-05	3407.9095	6705.2513	6425.6432	7671.8342	10057.864	16786.427	22643.764	27280.819	33107.07	36141.161
8.00E-05	3486.7638	6955.8141	6875.5779	8632.7638	13695.608	18656.789	24313.794	30661.899	32898.834	37901.487
7.00E-05	3668.6834	7528.9196	7582.6533	10520.673	14594.503	21585.935	27430.905	33478.191	37926.065	39971.744
6.00E-05	3962.5628	8184.2915	8908.6583	10364.412	17772.01	26005.452	33621.568	37461.166	40297.015	43397.05
5.00E-05	4167.6784	8865.0201	9484.1759	13928.859	20620.387	29064.899	36309.246	40110.719	43048.417	45538.367
4.00E-05	4515.8291	9644.5578	11563.161	18354.367	27010.352	33951.97	40799.658	44108.271	45917.915	48308.945
3.00E-05	5059.8794	11485.899	15935.819	23537.467	31983.07	40540.769	44894.171	48028.648	49944.307	52696.226
2.00E-05	6524.8744	15515.141	23050.367	31693.899	39692.101	46629.558	50373.94	54465.653	55854.437	57249.005
1.00E-05	8527.8995	25127.07	34262.568	43816.402	51902.055	55930.221	60159.965	61394.462	63830.06	66139.307
9.00E-06	8746.4221	26410.513	37680.035	46022.296	54010.824	57544.231	61327.05	63699.201	64750.774	68351
8.00E-06	9474.0101	29507.945	39589.829	48614.332	54025.236	59733.015	62842.513	64384.889	66818.925	68970.196
7.00E-06	10301.668	31298.121	40019.106	50429.01	56361.678	62009	63321.663	66217.427	69258.372	69606.422
6.00E-06	11850.653	33845.357	44199.276	51939.93	59028.905	63013.839	65983.749	67526.186	69809.583	70851.06
5.00E-06	13068.734	36965.256	47009.317	56133.382	62123.925	66072.719	68282.146	71326.035	72951.161	74612.08
4.00E-06	14670.05	40347.784	49610.824	59106.07	63852.477	69004.804	71463.156	73956.101	77287.442	76819.804
3.00E-06	17112.472	44845.005	56031.894	62566.347	67823.025	73042.563	74778.497	77379.653	80787.312	81893.528
2.00E-06	23285.869	52058.754	62225.432	70379.01	74180.412	77512.573	81224.447	85341.327	87238.95	87805.156
1.00E-06	35222.241	66830.714	77563.397	82040.663	86059.04	90214.985	93471.638	96062.583	99403.251	102968.42

 

迭代次数随着卷积核数量的增加而增加，在《计算多卷积核神经网络迭代次数》中已经得到的卷积核的数量与迭代次数的公式

 

ni/n(i-1)		b	1.1605951	1.057724	1.0489803	1.0482918	1.0360988	1.027719	1.034776	1.0358657
(i/(i-1))**0.5		a	1.4142136	1.2247449	1.1547005	1.118034	1.0954451	1.0801234	1.069045	1.0606602
a/b	α		1.2185245	1.157906	1.1007838	1.0665294	1.0572786	1.050991	1.0331173	1.023936
	X		2	3	4	5	6	7	8	9

 

用α和x拟合方程

α=1.3075677654798437*x**-0.11553016292685217

0.9715966417109875   ******  决定系数 r**2

计算迭代次数

	2	3	4	5	6	7	8	9
α	1.2069413	1.1517078	1.1140588	1.0857056	1.0630759	1.0443111	1.0283242	1.0144261
∏α		1.1517078	1.2830702	1.3930366	1.4809036	1.546524	1.5903281	1.6132704
计算		82482.182	85491.196	88036.813	90717.416	93828.37	97543.891	101989.61
实测		82040.663	86059.04	90214.985	93471.638	96062.583	99403.251	102968.42
误差		0.0053817	0.0065983	0.0241442	0.0294659	0.0232579	0.0187052	0.0095059

 

比较耗时

		1	2	3	4	5	6	7	8	9
δ	耗时 min/199	耗时 min/199	耗时 min/199	耗时 min/199	耗时 min/199	耗时 min/199	耗时 min/199	耗时 min/199	耗时 min/199	耗时 min/199
0.5	0.06415	0.1054333	0.1807833	0.2866167	0.3691167	0.4632	0.6012167	0.68235	0.7749	0.8857667
0.4	0.0743	0.4034333	0.5565167	0.74555	0.8677	1.0177333	1.2258	1.317	1.4365333	1.5769833
0.3	0.0762833	0.4464333	0.6118	0.8334833	0.9509833	1.1233167	1.3131	1.4768	1.5941	1.7541667
0.2	0.08025	0.4785167	0.66015	0.88305	1.0305833	1.1935333	1.3994	1.49935	1.6984167	1.8722667
0.1	0.0854667	0.5151333	0.7269333	0.9404167	1.1488	1.2794167	1.4967	0.6726667	1.81285	2.0303667
0.01	0.10295	0.6091	0.8742833	1.1268833	1.3877833	1.5445667	1.87975	1.98915	2.1972667	2.4928833
0.001	0.1482167	0.7868167	1.1538	1.5269333	1.9254333	2.2772667	2.8063167	3.1866833	3.7478167	4.4037333
9.00E-04	0.1486833	0.7794833	1.16055	1.5606167	1.969	2.3014333	2.8294333	3.2884667	3.9535667	1.89435
8.00E-04	0.1523833	0.7895	1.1887833	1.58825	1.9530167	2.3543833	2.89775	3.4075167	3.9764333	4.8350333
7.00E-04	0.15595	0.8114	1.1984167	1.62435	2.0165333	2.4227167	3.0292167	3.6387333	4.3147167	5.7025833
6.00E-04	0.1680833	0.8324333	1.2386	1.6883667	2.0880667	2.5167	3.18455	3.8301	4.8615833	6.2168833
5.00E-04	0.1823667	0.8596	1.2762667	1.7719167	2.1733333	2.6760833	3.2791167	4.5327833	6.0158667	8.5344833
4.00E-04	0.1947833	0.8926833	1.3244833	1.8532833	0.6664	2.9511333	4.0198333	5.58965	8.1587833	12.120433
3.00E-04	0.2015833	0.9296333	1.3770167	1.9949667	2.5713	3.3963333	4.93535	8.5260333	12.716067	18.7017
2.00E-04	0.2242667	1.07535	1.5356333	2.4373833	3.6019667	5.4545	7.9383167	14.16995	21.765317	29.262633
1.00E-04	0.2510167	1.3040167	2.0991	3.9648333	7.1169	12.690533	20.795567	31.041767	43.3882	54.798383
9.00E-05	0.2691833	1.3718833	2.3460667	4.2682	7.1659667	14.644033	23.287683	34.210233	46.565233	54.359333
8.00E-05	0.2690667	1.41815	2.4975833	4.7310667	9.6323167	16.210883	24.950917	35.979533	46.18115	59.05395
7.00E-05	0.2797	1.5245333	2.73555	5.70455	10.233417	18.691283	28.08415	40.165217	53.254283	62.658783
6.00E-05	0.29395	1.64815	3.18345	5.6891333	12.394317	22.412517	34.39305	43.790267	56.792533	66.610467
5.00E-05	0.3057167	0.7482667	3.3784667	7.4962667	14.319367	25.002167	38.500217	48.504167	58.532967	71.23395
4.00E-05	0.32555	1.90655	4.0798	8.5269167	18.6522	29.133117	41.577167	52.877717	63.498467	74.2049
3.00E-05	0.3573333	2.2570667	5.5666667	12.17225	22.0143	34.778783	45.4032	57.217	68.8245	83.56915
2.00E-05	0.4487167	2.9843667	7.9680833	16.277983	25.902283	40.653083	50.3917	67.005917	76.049067	87.9441
1.00E-05	0.5657	4.6324	12.738767	22.3845	35.64345	48.739217	61.586	70.405867	87.764633	101.07163
9.00E-06	0.5779	4.8042167	13.97685	23.499033	37.245083	48.107167	61.408567	75.853533	89.590633	106.89732
8.00E-06	0.6237333	5.3426	13.776983	24.87535	35.31665	50.382933	64.300433	77.051283	91.333683	105.29282
7.00E-06	0.6701667	5.6624333	11.376483	26.492033	38.42935	53.487317	64.339583	78.093217	96.4085	107.67667
6.00E-06	-1.043133	6.1118333	15.4332	27.433567	40.1873	52.854733	67.0226	79.72825	94.623517	109.36867
5.00E-06	0.8278667	6.6677333	16.386617	27.391317	42.329833	57.257867	69.3746	84.96685	100.74403	115.31163
4.00E-06	0.92955	7.2674	17.24095	30.8037	46.6526	57.598467	74.86045	88.066183	106.69952	118.55075
3.00E-06	1.0805333	8.0748	19.4551	32.2475	44.217	62.321083	78.466417	92.807433	112.71635	126.68888
2.00E-06	1.4429333	9.3541	22.58695	38.047467	52.679083	66.735667	82.635983	100.66913	121.80912	135.0271
1.00E-06	2.15025	11.933783	24.124117	40.834483	57.550183	77.955	97.3552	113.19602	138.3318	158.15827

计算迭代时间的表达式

用同样的办法

α=1.3075677654798437*x**-0.11553016292685217
	2	3	4	5	6	7	8	9
α	1.2069413	1.1517078	1.1140588	1.0857056	1.0630759	1.0443111	1.0283242	1.0144261
∏α		1.1517078	1.2830702	1.3930366	1.4809036	1.546524	1.5903281	1.6132704
计算		38.480969	53.179713	68.454014	84.646016	102.14023	121.35416	142.7457
实测		40.834483	57.550183	77.955	97.3552	113.19602	138.3318	158.15827
误差		0.0576355	0.0759419	0.1218778	0.1305445	0.0976694	0.1227313	0.0974503

 

综合上面的4个表格

卷积核并不是对所有的训练集都有正面价值，并不是所有的网络都可以通过增加卷积核的办法提升网络性能。