ADMM for Generalized Lasso Inference
L1 penalized inference is central to the field of compressed sensing. I derive an ADMM solver for the generalized Lasso problem and discuss three important special cases; standard lasso, variable fusion, and fused lasso. Further, I provide python code and demonstrate it on a simulated dataset. Read More...
Alternating Directions Method of Multipliers
Complex optimization problems can often be approached by splitting them up into tractable subproblems. I describe an approach where local solutions are coordinated to find a global solution. Read More...
Interactive Transposed Convolution Tool
I present an interactive tool that visualizes how altering parameters values of the transposed convlution affects the resulting output. Further, I discuss the connection between transposed convolution and standard convolution, and provide Pytorch Code that demonstrates their relationship. Read More...