【NNDL作业】 宽卷积具有交换性

习题5-2 证明宽卷积具有交换性,即公式: 

rot180(W) \widetilde{\otimes }X = rot180(X) \widetilde{\otimes }W

证明:

不失一般性,设:

W= \begin{pmatrix} w_{11} &w_{12} & \cdots &w_{1v} \\ w_{21} &w_{22} & \cdots &w_{2v} \\ \vdots &\vdots & \ddots & \vdots \\ w_{u1} &w_{u2} &\cdots& w_{uv} \end{pmatrix},  

X= \begin{pmatrix} x_{11} &x_{12} & \cdots &x_{1n} \\ x_{21} &x_{22} & \cdots &x_{2n} \\ \vdots &\vdots & \ddots & \vdots \\ x_{m1} &x_{m2} &\cdots& x_{mn} \end{pmatrix}

则:

rot(W)= \begin{pmatrix} w_{uv} &w_{u,v-1} & \cdots &w_{u1} \\ w_{u-1,v} &w_{u-1,v-1} & \cdots &w_{u-1,1} \\ \vdots &\vdots & \ddots & \vdots \\ w_{1v} &w_{1,v-1} &\cdots& w_{11} \end{pmatrix}

rot(X)= \begin{pmatrix} x_{mn} &x_{m,n-1} & \cdots &x_{m1} \\ x_{m-1,n} &x_{m-1,n-1} & \cdots &x_{m-1,1} \\ \vdots &\vdots & \ddots & \vdots \\ x_{1n} &x_{1,n-1} &\cdots& x_{11} \end{pmatrix}

W两端各补m-1n-1 个零,得到\widetilde{W}:
  \widetilde{W}= \begin{pmatrix} 0_{11} &0_{12} &\cdots &0_{1,n-1} &0_{1,n} &0_{1,n+1} &\cdots &0_{1,v+n-1} &0_{1,v+n} &0_{1,v+n+1} &\cdots &0_{1,v+2n-2} \\ 0_{21} &0_{22} &\cdots &0_{2,n-1} &0_{2,n} & & & & & & & \vdots\\ \vdots &\vdots &\ddots &\vdots &\vdots & & & & &\ddots & & \vdots\\ 0_{m-1,1} &0_{m-1,1} &\cdots &0_{m-1,n-1} &0_{m-1,n} & & &0_{m-1,v+n-1} & & & & \vdots\\ 0_{m,1} & & & &w_{11} &w_{12} &\cdots &w_{1v} & & & & \vdots \\ \vdots & & & &w_{21} &w_{22} &\cdots &w_{2v} & & & & \vdots \\ \vdots & &\ddots & &\vdots &\vdots &\ddots &\vdots & &\ddots & & \vdots \\ 0_{u+m-1,1} & & & &w_{u1} &w_{u2} &\cdots &w_{uv} & & & & \vdots \\ 0_{u+m-1,1} & & & & & & & & & & \vdots \\ \vdots& &\ddots & & & & & & &\ddots & & \vdots \\ \vdots& & & & & & & & & & & \vdots \\ 0_{u+2m-2,1}& & & & & & & & & & & 0_{u+2m-2,v+2n-2} \end{pmatrix}
X两端各补u-1v-1个零 得到\widetilde{X}:

 \widetilde{X}= \begin{pmatrix} 0_{11} &0_{12} &\cdots &0_{1,v-1} &0_{1,v} &0_{1,v+1} &\cdots &0_{1,v+n-1} &0_{1,v+n} &0_{1,v+n+1} &\cdots &0_{1,n+2v-2} \\ 0_{21} &0_{22} &\cdots &0_{2,v-1} &0_{2,v} & & & & & & & \vdots\\ \vdots &\vdots &\ddots &\vdots &\vdots & & & & &\ddots & & \vdots\\ 0_{u-1,1} &0_{u-1,1} &\cdots &0_{u-1,v-1} &0_{u-1,n} & & &0_{u-1,v+n-1} & & & & \vdots\\ 0_{u,1} & & & &x_{11} &x_{12} &\cdots &x_{1n} & & & & \vdots \\ \vdots & & & &x_{21} &x_{22} &\cdots &x_{2n} & & & & \vdots \\ \vdots & &\ddots & &\vdots &\vdots &\ddots &\vdots & &\ddots & & \vdots \\ 0_{u+m-1,1} & & & &x_{m1} &x_{m2} &\cdots &x_{mn} & & & & \vdots \\ 0_{u+m-1,1} & & & & & & & & & & \vdots \\ \vdots& &\ddots & & & & & & &\ddots & & \vdots \\ \vdots& & & & & & & & & & & \vdots \\ 0_{m+2u-2,1}& & & & & & & & & & & 0_{m+2u-2,n+2v-2} \end{pmatrix}

由定义:

rot180(W) \widetilde{\otimes }X = rot180(W) \otimes\widetilde{X}

rot180(X) \widetilde{\otimes }W = rot180(X) \otimes\widetilde{W}

rot180(W) ,\widetilde{X},rot180(X) ,\widetilde{W}代入,计算可得:

rot180(X)\otimes \widetilde{W}=\begin{pmatrix} x_{11}w_{11}&x_{11}w_{12}+x_{12}w_{11} &\cdots &x_{1n}w_{1v} \\ x_{11}w_{21}+x_{21}w_{11} &x_{11}w_{22}+x_{12}w_{21}+x_{21}w_{12}+x_{22}w_{11} &\cdots &x_{1n}w_{2v}+x_{2n}w_{1v} \\ \vdots&\vdots &\ddots &\vdots \\ x_{m1}w_{u1}&x_{m1}w_{u2}+x_{m2}w_{u1} &\cdots &x_{mn}w_{un} \end{pmatrix}

 rot180(W)\otimes \widetilde{X}=\begin{pmatrix} w_{11}x_{11}&w_{11}x_{12}+w_{12}x_{11} &\cdots &w_{1v} x_{1n}\\ w_{21}x_{11}+w_{11}x_{21} &w_{22}x_{11}+w_{21}x_{12}+w_{12}x_{21}+w_{11} x_{22}&\cdots &w_{2v}x_{1n}+w_{1v}x_{2n} \\ \vdots&\vdots &\ddots &\vdots \\ w_{u1}x_{m1}&w_{u2}x_{m1}+w_{u1}x_{m2} &\cdots &w_{un} x_{mn}\end{pmatrix}

【更正】两个矩阵右下角的元素w_{un}应为w_{uv}

乘法交换律可知: x_{ij}w_{pq} = w_{pq}x_{ij},所以矩阵内的对应位置的元素均相等,即两个矩阵相等。

因此:rot180(W) \otimes\widetilde{X} = rot180(X) \otimes\widetilde{W}

进一步可得:rot180(W) \widetilde{\otimes }X = rot180(X) \widetilde{\otimes }W

原式得证. 

  • 8
    点赞
  • 7
    收藏
    觉得还不错? 一键收藏
  • 0
    评论

“相关推荐”对你有帮助么?

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值