Short-term memory
• Handled by treating the tabu list as a circular list
• size : t
• Perform effectively for driving the search beyond local optima and obtaining progressively improved solutions
• Aspiration criteria
Long-term memory
• Endow the memory structures with a flexibility to choose the “most attractive” moves by evaluation function
• Doesn’t restrict evaluations to measures of “ascent” and “descent” , but more adaptive and varied measures
Tabu List Strategies for Single Attribute Moves
TL is a vector of attributes which impart a tabu classification to moves that contain these attributes
TL = (e(1) , e(2) ,… e(q))
Identify the list of solutions (x(1) , x(2) ,… x(q)) such that , for each i , e(i) is the attribute associated with the move applied to x(i) to prevent this move from being reversed to return to x(i)
Tabu List Strategies for Single Attribute Moves
• e(i) & e(i)
ex : if x(i) → x(i+1) , e(i) set xj =1
and x(i+1) → x(i) , e(i) set xj =0
• in the sequence e(p) ,… e(q) , if any e(r) is followed by an element e(s) such that e(r) = e(s) the e(r) is said to be canceled by the first such e(s)
Tabu Status Based on Cancellation Sequences
•active tabu list , ATL , consists only of the element
of TL that have not been canceled
ATL = (e(p) ,… e(q))
•An element e(p+1) is added to ATL , where e(q+1) =
e(q) , and e(q+1) cancle an earlier element e(i) of
ATL as a result of e(q+1) = e(i)
Tabu Status Based on Cancellation Sequences
•The structure of ATL , upon adding e(q+1) but
berore dropping e(i) , may be depicted as follows : ATL = (e(p) ,… e(h) , e(i) , e(j)… e(q) , e(q+1) )
•Cancellation Sequence , or C-Sequence : lies between the canceling element e(q+1) and the
canceled element e(i)
Tabu Status Based on Cancellation Sequences
•startseq(e) & endseq(e)
startseq(e) denote the element f on ATL that
starts the C-Sequence terminated by e
endseq(e) denote the element g on ATL that
end the C-Sequence initial by e