The way the course is being built, it is a great series. Thanks to Prof. Constantine Caramanis One comment on this particular video is, instead of stating that solving the optimization problem is difficult, an example will be very helpful to appreciate the usefulness of the approximation of the functions. Motivation for going for iterative methods instead of solving directly the gradient equation will also help.
If I may ask, the graph you have of |10x1| + |x2| doesn't seem correct to me. For example, at (0,-3) we should have f(x1,x2) be (0,3) having no negative values or am I missing something?
excellent course
23:15 choose the subdifferential gx=[10m,-1], where m in [-1,0) ?
The way the course is being built, it is a great series. Thanks to Prof. Constantine Caramanis
One comment on this particular video is, instead of stating that solving the optimization problem is difficult, an example will be very helpful to appreciate the usefulness of the approximation of the functions. Motivation for going for iterative methods instead of solving directly the gradient equation will also help.
I from Vietnam. Thank for course
If I may ask, the graph you have of |10x1| + |x2| doesn't seem correct to me. For example, at (0,-3) we should have f(x1,x2) be (0,3) having no negative values or am I missing something?
How is the convergence estimated as exp(-ct) at 17:30?
its not exp() it's a constant raised to (-ct). If I understood it correctly.
It is equivalent:
let
K = (1-6*eta)
and
C = -ln(K) = - ln(1-6*eta),
then
K = exp(-C),
and
(1-6*eta)^t = K^t = exp(-Ct)