Subgradient approach

[michiexile]

Both L_1 and L_∞ optimizations can be reformulated to linear programs, which should be easier to solve.

Can we do the same kind of reformulation for a mixed optimization target?

https://math.stackexchange.com/questions/706935/minimizing-l-infty-norm-using-gradient-descent

[michiexile]

Linear programming is also implemented in scipy.optimize as scipy.optimize.linprog:
https://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.linprog.html