Direct Search Flashcards
1
Q
what is direct search methods ?
A
- direct search methods are algorithms that do not attempt to approximate gradients
- derivative free optimization.
2
Q
Finite Differencing
A
Estimate the gradient using finite differences
3
Q
Disadvantage of Finite Differencing ?
A
- step size problem
- more computational costs
- not accurate with noise
4
Q
what is model-based derivative-free methods ?
A
- using trust region methods to compute the step
- switch from a linear to a quadratic model once enough points evaluated and become available
5
Q
What is coordinate search ?
A
- cycles through the n coordinate directions
- obtaining new iterates by performing a line search along each direction in turn.
6
Q
Disadvantage of coordinate search ?
Advantage ?
A
- iterate infinitely without ever approach to a stationary point
- slow than steepest descent method
advantage:
- no need derivative.
7
Q
Pattern search
A
??-direction set.
8
Q
Adv and Dis-adv of pattern search ?
A
Disadvantage:
-noise affect performance.
9
Q
Explain Nelder-Mead simplex method ?
A
- simplex of n+1 points
- different operations of the simplex to find a better point to remove the worst point in simplex.
- if not then shrink the simplex.
10
Q
Advantage and disadvantage of the simplex method.
A
Disadvantage:
- converge to non-stationary points
- bad for ill conditioning problem.
11
Q
What is implicit filtering ?
A
- steepest descent variant
- use finite difference estimate gradient
- good for noise and noise level decrease approach to optima.
