Please enable JavaScript.
Coggle requires JavaScript to display documents.
Sparse Matrix (sparse matrix multiplication (\( \bar{c}_{*, j} = \sum…
Sparse Matrix
:pineapple:presentation of sparse matrix
:one:从列开始计算,并且保存non-zero的row_index和当前列以及当前列之前的所有non-zero元素的个数
\(B_v\)记录所有non-zero元素
\(B_c\)表示第i个列及之前列的所有non-zero元素的个数
\( B_i \)表示这个元素的row_index
:two:LibLinear
\( *BS \)表示该列的起始地址,是指针的指针
DN表示dummy node
最后两个DN中,第一个是分隔用,第二个是说明该列没有non-zero元素
Node包含row_index和当前value值
sparse cnn
sparse matrix multiplication
\(C = A \times B, \text{where } A \in R^{m \times k}, B \in R^{k \times n}\)
\(A\) dense, \(B\) sparse
\( \bar{c}_{*, j} = \sum_{i=1}^k{ \bar{a}_{*,i} \bar{b}_{i, j} }, 1 \leq j \leq 8 \)
\(B\) is sparse, so we need to know what role non-zero value plays in matrix multiplication
\(\bar{b}_{i,j}\) means that it times the i-th column of \(A\) and saved to the j-th column of \(C\)
Spatially-sparse convolutional neural networks
input feature maps
can be considered to be sparse if padding
e.g.
:one:online handwriting
a sequence of strokes
