Numerical Optimization: Numerical Geometry of Non-Rigid Shapes
Numerical Optimization: Numerical Geometry of Non-Rigid Shapes
Numerical Optimization: Numerical Geometry of Non-Rigid Shapes
Numerical Optimization
Alexander Bronstein, Michael Bronstein 2008 All rights reserved. Web: tosca.cs.technion.ac.il
Slowest
Longest
Shortest
Maximal
Minimal
Fastest Largest
Smallest
Optimization problems
Generic unconstrained minimization problem
where Vector space A solution The value is the minimum is the search space
Local minimum
Global minimum
Convex functions
A function defined on a convex set is called convex if
for any
and
Convex
Non-convex
Gradient
In multidimensional case, linearization of the function according to Taylor
The function
, denoted as
Gradient
In Euclidean space ( ), can be represented in standard basis
i-th place
which gives
10
11
12
Hessian
Linearization of the gradient
The function
is called the Hessian of
, denoted as
13
14
Optimization algorithms
Descent direction Step size
15
such that
Update iterate
Until convergence
Increment iteration counter Solution Descent direction Step size Stopping criterion
16
Stopping criteria
Near local minimum, (or equivalently )
17
Line search
Optimal step size can be found by solving a one-dimensional optimization problem
One-dimensional optimization algorithms for finding the optimal step size are generically called exact line search
18
19
Descent direction
How to descend in the fastest way? Go in the direction in which the height lines are the densest
Devils Tower
Topographic map
20
Steepest descent
Directional derivative: how much changes in the direction (negative for a descent direction)
21
Steepest descent
L2 norm
L1 norm
22
Until convergence
Update iterate
23
MATLAB
intermezzo
Steepest descent
24
Condition number
Condition number is the ratio of maximal and minimal eigenvalues of the Hessian
1
,
1
0.5
0.5
-0.5
-0.5
-1 -1
-0.5
0.5
-1 -1
-0.5
0.5
Problem with large condition number is called ill-conditioned Steepest descent convergence rate is slow for ill-conditioned problems
25
Q-norm
Change of coordinates
Q-norm
L2 norm
26
Preconditioning
Using Q-norm for steepest descent can be regarded as a change of coordinates, called preconditioning Preconditioner should be chosen to improve the condition number of
the Hessian in the proximity of the solution In system of coordinates, the Hessian at the solution is
(a dream)
27
is unknown in advance
28
(quadratic function in
Close to solution the function looks like a quadratic function; the Newton method converges fast
29
Newton method
Start with some Compute Newton direction
Until convergence
Update iterate
30
Frozen Hessian
Observation: close to the optimum, the Hessian does not change significantly Reduce the number of Hessian inversions by keeping the Hessian from previous iterations and update it once in a few iterations Such a method is called Newton with frozen Hessian
31
Cholesky factorization
Decompose the Hessian
where
Backward substitution
Complexity: , better than straightforward matrix inversion
32
Truncated Newton
Solve the Newton system approximately
A few iterations of conjugate gradients or other algorithm for the solution of linear systems can be used Such a method is called truncated or inexact Newton
33
Non-convex optimization
Using convex optimization methods with non-convex functions does not guarantee global convergence! There is no theoretical guaranteed global optimization, just heuristics
Local minimum
Global minimum
Good initialization
Multiresolution
34
Iterative majorization
Construct a majorizing function . Majorizing inequality: for all satisfying
35
Iterative majorization
Start with some Find such that
Update iterate
Until convergence
36
Constrained optimization
37
where
38
An example
Equality constraint Inequality constraint
Feasible set
Inequality constraint
A point
is active at point
if
, inactive otherwise
and of
39
Lagrange multipliers
Main idea to solve constrained problems: arrange the objective and constraints into a single function
and minimize it as an unconstrained problem is called Lagrangian and are called Lagrange multipliers
40
KKT conditions
If is a regular point and a local minimum, there exist Lagrange multipliers and Known as Karush-Kuhn-Tucker conditions Necessary but not sufficient! such that for all such that inactive constraints and for all
41
KKT conditions
Sufficient conditions:
If the objective
are convex
and the equality constraints then for all such that inactive constraints
minimizer)
42
Geometric interpretation
Consider a simpler problem: Equality constraint
43
Penalty methods
Define a penalty aggregate
where
and
44
Penalty methods
Inequality penalty
Equality penalty
45
Penalty methods
Start with some Find and initial value of
Until convergence
Set
Set Update
Solution