Martin Nocker
Martin Nocker
Convolution and AveragePool layers can be implemented using vector-matrix multiplication, where the vector is your (encrypted) CKKSVector matrix is a (rather big) sparse matrix simulating the convolution and averagepool, respectively....
Hello @napoay, the * (star) operator performs **element-wise** multiplication. If you want to do a vector-matrix multiplication, you should use `enc_y.mm(X)` or `enc_y @ X` or `enc_y.dot(X[0])` if you want...
Ok, I figured out, TenSEAL uses the diagonal method to compute vector-matrix products. This means for the identity matrix that all vectors except the first one are zero, resulting in...
Hello @dimk1, this would be very interesting! As far as i know, the only way to parallel TenSEAL is to parallel the outer computation, so for example for a neural...
Hey everybody, check out #403 where I showcase a solution to have multiple conv layers. Best, Martin
@scriptingxss @rockhoppersec Thanks in advance :)