Tabu Search: Intermediate Concepts

1
Tabu Search Intermediate Concepts
2
Short Term Memory Issues

• List size
• No agreement on a best size
• Generally want smaller sizes for stringent
  restrictions
• Sizes can vary by specific list
• Tabu restrictions
• Context dependent
• Can have multiple types of restrictions
• Not all necessarily are in effect simultaneously

3
Short Term Memory Issues

• Aspiration criteria
• Can have multiple criteria
• Each criteria may have its own tenure
• Again, not all need be in effect at any one time
• Just about anything goes as an aspiration
  criteria
• Litmus test is does the aspiration criteria work

4
What if Short Term Fails?

• Short term memory functions and tabu search often
  sufficient for many problems
• There are cases where the approach requires more
  power
• Maybe the best so far curve has leveled off
• Maybe we know better solutions are available
• The other memory components of tabu search can
  now come into play
• Approached failed to drive search into new areas

5
Bridging Data Structure

• Name provided to distinguish how data structures
  are used and exploited
• These allow movement between memory functions
• Based on auxiliary data captured by search
• Analyzed on-line and events triggered
• Go to intermediate memory functions
• Go to long-term memory functions
• Change tenure, restrictions, aspiration criteria,
  etc.

6
Tabu Search Functions
Current Focus
7
Purpose of Functions
8
Intermediate Tabu Search

• Primarily a return to good areas
• Elite List of solutions
• Return to elite solutions and intensify
• Good areas include
• Strong solutions visited
• Strong solutions evaluated but not visited
• Solutions constructed via some set of rules
• Resume or restart search from a new area when
  search in an old area bogs down

9
Intermediate Tabu Search
Apply Short-Term Tabu Search
When Improvement Stops, Get Member of Elite List
Continue with the Short-Term Tabu Search from
new Solution
Continue until Stopping Criteria is Met
Ref OR/MS Encyclopedia
10
Elite Set Approaches

• Voss (1993) erases memory and returns to previous
  solutions different from current solution
• Glover (1990) tracks quality neighbors not picked
• An unvisited neighbor strategy
• Tsubakitani and Evans (1998) jump to new random
  solutions
• Kinney et al. (2003) generate list of good
  solutions

11
Tabu Search Functions
Current Focus
12
Purpose of Functions
13
Long-Term Tabu Search

• Short and Intermediate functions may not provide
  enough diversification
• Want really new solutions move to new portions
  of the search space
• Long-term memory functions
• Counts of move characteristics
• Move distance evaluations
• Relax constraints-- Strategic Oscillation
• Randomization of solutions

14
Long-Term Tabu Search
Apply Short-Term Tabu Search
When Some Trigger Occurs, Initiate Long-Term
Functions
Continue Short-Term Tabu Search from new Solution
Continue until Stopping Criteria is Met
Ref OR/MS Encyclopedia
15
Diversification

• Strategic oscillation
• Systematically cross boundaries ordinarily fixed
• Allow movement to regions previously not reached
• Vocabulary building
• Generate new solutions based on knowledge of
  solutions visited
• Exploit data structures (memory) containing
  search information
• e.g., examine frequency counts and fix variables
  to values

16
Diversification

• Move distance
• Force moves within the search that cause a large
  change is the structure of the solution
• Essentially a new evaluation function focused on
  change versus quality of solution neighborhood
• Also called move influence
• Path relinking
• Choose two (good) solutions
• Generate a path between the solutions
• Tunnel through infeasible region if necessary

17
Some Key Concepts

• Candidate Move and Candidate List
• Tabu Tenure and Tabu List
• Aspiration Criteria
• Do not ignore good solutions, make exceptions
• Aggressive Exploration
• Explicit Memory
• Keep track of lots of information
• Intensification and Diversification
• Move between the two and search the space

18
Questions?