Scribd
Upload
78 views

Local Search

Local search algorithms are used to find solutions to optimization problems where the path to the goal is irrelevant and only the goal state itself is the solution. Local search algorithms keep a single "current" state and try to improve it through incremental changes, using little memory. Examples of local search algorithms include hill climbing, simulated annealing, and genetic algorithms.

Uploaded by

DUDEKULA VIDYASAGAR
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
78 views

Local Search

Local search algorithms are used to find solutions to optimization problems where the path to the goal is irrelevant and only the goal state itself is the solution. Local search algorithms keep a single "current" state and try to improve it through incremental changes, using little memory. Examples of local search algorithms include hill climbing, simulated annealing, and genetic algorithms.

Uploaded by

DUDEKULA VIDYASAGAR
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 34

Beyond Classical Search

Search Algorithms So Far


• Designed to explore search space
systematically:
– keep one or more paths in memory
– record which have been explored and which have
not
– a path to goal represents the solution

CS 420: Artificial Intelligence 7


Local Search Algorithms
• In many optimization problems, the path to the goal is irrelevant; the goal
state itself is the solution

• State space = set of "complete" configurations


• Find configuration satisfying constraints, e.g., n-queens

• In such cases, we can use local search algorithms


• keep a single "current" state, try to improve it

– use very little memory – usually a constant amount


– find reasonable solutions in large or infinite state spaces for which systematic
solutions are unsuitable
– useful for solving optimization problems, e.g. Darwinian evolution, no “goal
test” or “path cost”

CS 420: Artificial Intelligence 8


Example: n-Queen Problem
• Put n queens on an n × n board with no two queens on the
same row, column, or diagonal

CS 420: Artificial Intelligence 9


State Space Landscape
• Problem: depending on initial state, can get
stuck in local maxima/minima

CS 420: Artificial Intelligence 10


Example: Initial State

GOLD

Assume the objective function measures the straight-line distance


CS 420: Artificial Intelligence 11
Example: Local Minima

GOLD

<< I am stuck

Assume the objective function measures the straight-line distance


CS 420: Artificial Intelligence 12
Example: A Plausible Solution
Making some “bad” choices is actually not that bad

GOLD

Assume the objective function measure the straight-line distance


CS 420: Artificial Intelligence 13
Hill-Climbing Search
• "Like climbing Everest in thick fog with
amnesia"

CS 420: Artificial Intelligence 14


Example: 8-Queen

• h = number of pairs of queens that are attacking each other, either directly or
indirectly

• h = 17 for the above state

CS 420: Artificial Intelligence 15


Example: 8-Queen

• A local minimum with h = 1

CS 420: Artificial Intelligence 16


More on Hill Climbing
• Complete? Optimal?
• Hill climbing is sometimes called greedy local search
• Although greedy algorithms often perform well, hill climbing gets stuck
when:
– Local maxima/minima
– Ridges
– Plateau (shoulder or flat local maxima/minima)

• The steepest-ascent hill climbing solves only 14% of the randomly-


generated 8-queen problems with an avg. of 4 steps
• Allowing sideways move raises the success rate to 94% with an avg. of 21
steps, and 64 steps for each failure

CS 420: Artificial Intelligence 17


Variants of Hill Climbing
• Stochastic hill climbing:
– chooses at random from among uphill moves
– converges more slowly, but finds better solutions in some landscapes
• First-choice hill climbing:
– generate successors randomly until one is better than the current
– good when a state has many successors
• Random-restart hill climbing:
– conducts a series of hill climbing searches from randomly generated
initial states, stops when a goal is found
– It’s complete with probability approaching 1

CS 420: Artificial Intelligence 18


More on Random-Restart Hill Climbing

• Assume each hill climbing search has a probability p of


success, then the expected number of restarts required is 1/p
• For 8-queen problem, p = 14%, so we need roughly 7
iterations to find a goal
• Expected # of steps = cost_to_success + (1-p)/p *
cost_to_failure

• Random-restart hill climbing is very effective for n-queen


problem
• 3 million queens can be solved < 1 min

CS 420: Artificial Intelligence 19


Some Thoughts
• NP-hard problems typically have an exponential number of
local maxima/minima to get stuck on
• A hill climbing algorithm that never makes “downhill” (or
“uphill”) moves is guaranteed to be incomplete
• A purely random walk – moving to a successor chosen
uniformly at random – is complete, but extremely inefficient
• What should we do?
• Simulated annealing (P115)

CS 420: Artificial Intelligence 20


What is Simulated Annealing?
• The process used to harden metals and glass
by heating them to a high temperature and
then gradually cooling them, thus allowing the
material to reach a low-energy crystalline
state
Ping-Pong Ball Example

CS 420: Artificial Intelligence 22


Simulated Annealing
• Idea: escape local maxima by allowing some "bad" moves but
gradually decrease their frequency

CS 420: Artificial Intelligence 23


Analysis of Simulated Annealing

• One can prove: If T decreases slowly enough, then simulated


annealing search will find a global optimum with probability
approaching 1

• Widely used in VLSI layout, airline scheduling, etc.

• An example:
http://foghorn.cadlab.lafayette.edu/fp/fpIntro.html

CS 420: Artificial Intelligence 24


Local Beam Search
• Idea:
– Keep track of k states rather than just one
– Start with k randomly generated states
– At each iteration, all the successors of all k states are generated
– If any one is a goal state, stop; else select the k best successors from
the complete list and repeat
• Is it the same as running k random-restart searches?
• Useful information is passed among the k parallel search
threads
• Stochastic beam search: similar to natural selection, offspring
of a organism populate the next generation according to its
fitness

CS 420: Artificial Intelligence 25


Genetic Algorithms
• A successor state is generated by combining two
parent states

• Start with k randomly generated states (population)

• A state is represented as a string over a finite alphabet 16257483


(often a string of 0s and 1s or digits)

• Evaluation function (fitness function). Higher values


for better states

• Produce the next generation of states by selection,


crossover, and mutation

CS 420: Artificial Intelligence 26


Genetic Algorithms crossover point

• Fitness function: number of non-attacking pairs of queens (min


= 0, max = 8 × 7/2 = 28)
• 24/(24+23+20+11) = 31%
• 23/(24+23+20+11) = 29% etc

CS 420: Artificial Intelligence 27


Example: 8-Queen

CS 420: Artificial Intelligence 28


Example: TSA
• http://www.obitko.com/tutorials/genetic-
algorithms/

CS 420: Artificial Intelligence 29


More on Genetic Algorithms
• Genetic algorithms combine an uphill
tendency with random exploration and
exchange of information among parallel
search threads
• Advantages come from “crossover”, which
raise the level of granularity

CS 420: Artificial Intelligence 30


A Genetic Algorithm

CS 420: Artificial Intelligence 31


Exercise
• 4.11 Give the name of the algorithm that
results from each of the following special
cases:
– Local beam search with k = 1
– Local beam search with one initial state and no
limit on the number of states retained
– Simulated annealing with T = 0 at all times (and
omitting the termination test)
– Genetic algorithm with population size N = 1
CS 420: Artificial Intelligence 32
Searching With Partial Information
• We have covered: Deterministic, fully observable  single-state problem
– agent knows exactly which state it will be in
– solution is a sequence
• Deterministic, non-observable  multi-state problem
– Also called sensorless problems (conformant problems)
– agent may have no idea where it is
– solution is a sequence
• Nondeterministic and/or partially observable  contingency problem
– percepts provide new information about current state
– often interleave search, execution
– solution is a tree or policy
• Unknown state space  exploration problem (“online”)
– states and actions of the environment are unknown

CS 420: Artificial Intelligence 33


Example: Vacuum World
• Single-state, start in #5.
Solution?
– [Right, Suck]

• Multi-state, start in #[1, 2, …,


8]. Solution?
– [Right, Suck, Left, Suck]

CS 420: Artificial Intelligence 34


Sensorless (multi-
state problem)

CS 420: Artificial Intelligence 35


Contingency Problem
• Contingency, start in #5 & 7.
– Nondeterministic: suck may dirty
a clean carpet
– local sensing: dirt, location only
at current location
– Solution?
– Percept: [Left, Clean]  [Right, if
dirty then Suck]

CS 420: Artificial Intelligence 36


Exam 2
• Closed book and closed notes
• Coverage:
– Informed Search
– Local Search
• Format:
– True/False
– Short-Answer
– Problem Solving
Informed Search
• Understand what is informed search
• Understand best-first, greedy, and A*
• Understand the optimality and completeness
of each search
• Be able to draw search tree
• Understand the heuristic function in A*, under
what condition will A* be optimal?
– Admissible heuristic
– Consistent heuristic
CS 420: Artificial Intelligence 38
Local Search
• Hill climbing (HC)
– Steepest HC, stochastic HC, 1st choice HC, random-restart HC
• HC with side-moves allowed
• Simulated annealing
• Local beam search
• Genetic algorithms

• Understand each search strategy


• Understand how to define the objective function for local
search

CS 420: Artificial Intelligence 39

You might also like