3 ms·
Not quite - see my response for to GP for what I was describing.
by szvsw 2y ago
Not quite - see my response for to GP for what I was describing.
- aghilmort 2y agoahi interesting so residual error derivatives in a sense? the diamond construct also feels evocative of dimer &/or branched manifold / lattice methods, be that Viterbi or otherwise 2.2 in op post is reminiscent of that, e.g., if we view the DCT reconstruction as implicit matched filter yes in theory should converge on similar result may quickly get into alternating conic optimization especially depending how the signal pair constructed, e.g., if one signal is an ECC &/or if L1 regularization is operating as an error squasher alternating op, definitely good stuff
- szvsw 2y ago> so residual error derivatives in a sense? Yep, loss = residuals*2 + lambda*sum(|a_i|), and we take the gradient of that with respect to the a_i to guide our gradient descent steps.