当前位置:网站首页>Influence of black and white pixel distribution on iteration times
Influence of black and white pixel distribution on iteration times
2022-07-27 10:58:00 【Black elm】
( A , B)---81*30*2---(1,0)(0,1)
Classify with neural network A and B, Give Way A yes mnist Of 0, Give Way B yes mnist Of 1-9. Reduce it to 9*9. But let B Central African 0 The number of values is equal to 14,14*0.9,14*0.8,…,14*0.2. Compare in Africa 0 When the number of values is the same ,0 And B The number of iterations of classification and non 0 The relationship between the number of values .
1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | ||
14 | 8028.492 | 7170.111 | 4563.92 | 6931.015 | 3712.191 | 5297.548 | 7865.568 | 4449.216 | 7524.784 | 1.00E-05 |
14*0.9 | 10512.76 | 7085.307 | 3957.085 | 5011.256 | 4287.045 | 4568.633 | 5539.92 | 3349.618 | 5888.221 | 1.00E-05 |
14*0.8 | 12175.1 | 8909.668 | 5329.327 | 5052.623 | 3973.759 | 3833.618 | 4698.231 | 3198.503 | 4814.271 | 1.00E-05 |
14*0.7 | 14806.39 | 9178.673 | 3901.99 | 4177.176 | 3760.663 | 3874.573 | 5672.623 | 3730 | 5220.161 | 1.00E-05 |
14*0.6 | 18403.16 | 12387.02 | 4450.814 | 4873.387 | 3520 | 4857.91 | 4257.085 | 4653.035 | 4988.834 | 1.00E-05 |
14*0.5 | 18569.19 | 16936.02 | 5504 | 4610 | 7360.563 | 5293.357 | 5994.452 | 5058.241 | 7303.035 | 1.00E-05 |
14*0.4 | 32946.66 | 21557.49 | 8226.271 | 6770 | 8412 | 6036 | 8845.538 | 6021.759 | 7886.734 | 1.00E-05 |
14*0.3 | 77674.51 | 32116.68 | 10478.2 | 8889.276 | 10278.59 | 9077.477 | 15703.26 | 9730.201 | 10279.68 | 1.00E-05 |
14*0.2 | 104475.7 | 59634.69 | 15644.66 | 15570.22 | 18435.36 | 16665.61 | 28967.59 | 15053.71 | 17749.77 | 1.00E-05 |
such as 14*0.5 That's ok , Show in B Random reservation in 7 About a non 0 value , The convergence error is 1e-5, convergence 199 Time Statistical average . So every line B The number of white dots in is the same . The difference in the number of iterations depends only on the distribution of white dots .
Statistics can find some obvious rules ,

When B by 3,4,5,6,8,9 when , The shape of the iteration number curve looks very consistent .
7 The value of is only slightly larger than the above situation .

But when B=1,2 The number of iterations is much larger ,

So the shape of the iteration number curve is in the case of equal value , Approximately divided into 3 Kind of :1,2 and 3-9.
| 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 |
| 44.4698 | 27.7134 | 41.7294 | 17.9278 | 15.9034 | 16.9892 | 17.8176 | 14.0056 | 19.4652 | 15.9438 |
This phenomenon and 1-9 Central African 0 The proportion of values is highly correlated ,3-9 Non - 0 The number of values is about 16.8 individual , This value is less than 1 Of 27.7 Less than 2 Of 41.7 individual .
3<1<2.
But this order is related to the number of iterations
3<2<1
Not the same. .
because
C(16,2)<C(27,2)<C(41.2)
If the number of iterations is inversely proportional to entropy , I think it's in 0 In the external environment, from the distribution form, such as 2 Of 41 Click here 3 The number of points may be smaller than that from the distribution form, such as 1 Of 27 Choose... From the points 3 The possibility of a point . Or equivalently think of form 0 hold 2 Medium 41 There are fewer points , But the 1 Of 27 There are more points .
C(16,2) <C(41-m .2) <C(27+n ,2)
Or equivalent understanding form 0 hold 2 Shrunk , But the 1 Magnified . Therefore, the difference between forms may be expressed equivalently by the cardinality in the permutation .
边栏推荐
- 学习笔记-简易服务器实现
- Li Kou brush question 02 (sum of three numbers + sum of maximum subsequence + nearest common ancestor of binary tree)
- 全校软硬件基础设施一站式监控 ,苏州大学以时序数据库替换 PostgreSQL
- Data types and variables
- MySQL log management, backup and recovery
- Camera switching
- How to smooth the online and offline of Web Services
- 招聘顶尖人才!旷视科技“MegEagle创视者计划”正式启动
- Distributed block device replication: client
- Apache22 and opencms9.5 integrated configuration
猜你喜欢

Awesome! VMware esxi installation record, with download

Antd table+checkbox default value display

MySQL数据表的高级操作

文档智能多模态预训练模型LayoutLMv3:兼具通用性与优越性

Open source project - taier1.2 release, new workflow, tenant binding simplification and other functions

How to build a data index system is the most effective. It will take you a quick start from 0 to 1

Set up Samba service

flask_restful中的输出域(Resource、fields、marshal、marshal_with)

Gamer questions

Tdengine helps Siemens' lightweight digital solution simicas simplify data processing process
随机推荐
Object array de duplication
Have you ever seen this kind of dynamic programming -- the stock problem of state machine dynamic programming (Part 1)
BeautifulSoup的使用
TDengine 商业生态合作伙伴招募开启
MIMO array 3D imaging technology based on mobile terminal
基于Spark封装的二次开发工程edata-base,介绍
Codeforces Round #807 (Div 2.) AB
颜值爆表!推荐两款JSON可视化工具,配合Swagger使用真香
How to smooth the online and offline of Web Services
7z用法
php生成文字图片水印
Background color style modification on table hover in antd
antd table+checkbox 默认值显示
招聘顶尖人才!旷视科技“MegEagle创视者计划”正式启动
The difference between scalar, vector, matrix and tensor in deep learning
想要一键加速ViT模型?试试这个开源工具!
ASP. Net core dependency injection journey: 1. Theoretical concepts
It is thought-provoking: is syntax really important? Qiu Xipeng group proposed a powerful baseline for aspect based emotional analysis
MySQL数据表的高级操作
对象数组去重