I have a tensor three dimension tensor $$H$$ of size 4x8x128. I divide this tensor into two sub tensors (ill be using matlab language to demonstrate)

$$H_1 = H(:,:,1:64)$$ $$H_2= H(:,:, 65:128)$$

Then I perform low rank ( example rank 3) canonical polyadic decomposition (CPD) of both $H_1$ and $H_2$ to obtain the factor matrices

$$ A_1, B_1, C_1$$
$$ A_2,B_2,C_2$$ Note that $A_1$ is if size 4x3 $B_1$ of size 8x3 and $C_1$ of size 64x3. The CPD is solved using alternate least square. I linked paper that shows how one can do that (page 5)


My question is does it make sense that somehow, the factor matrices $A_1$ and $A_2$ should be related? and what about $B_1$ and $B_2$? As you see the only difference between my two tensors $H_1$ and $H_2$ is the third dimension, does that mean only the third factor matrix should be different? If that is not that case, is there a way for me to actually show the relationship between the factor matrices? how can I compare the factor matrices? in other words, Id like to know what the relationship between the factor matrices are, and currently have no idea in mind on how I can do that...

Is there a correlation metric one can think of? Thank you


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.