Lecture 3 - Metaheuristics Flashcards
What is the null hypothesis?
A null hypothesis states the results are due to chance and are not significant in terms of supporting the idea being investigated
What is a p-value?
A number describing how likely it is that your data would have occurred by random chance i.e. that the null hypothesis is true
What does it tell us when two notched boxplots do not overlap?
You can conclude that with 95% confidence, that one is better than the other.
What is a metaheuristic?
A high-level problem independent algorithmic framework that provides a set of guidelines or strategies to develop heuristic optimisation algorithms
How do heuristics escape from local optima?
Iterate with different solutions, or restart - however, restarting could be costly
Change the search landscape - change the objective function, or mix different neighbourhoods
Use memory
Accept non-improving moves - allow search using candidate solutions with equal or worse evaluation function value than the one in hand.
What are some example termination criteria?
A fixed maximum number of iterations, or moves, or a fixed amount of CPU time is exceeded.
Consecutive number of iterations since the last improvement in the best objective function value.
Evidence can be given that an optimal solution has been obtained
No feasible solution can be obtained for a fixed number of steps/time.
How do you deal with infeasible solutions?
Use a problem domain specific repair operator
Penalise each constraint violation for the infeasible solutions such that they can’t be better than the worst feasible solution for a given instance.
What are the properties of Iterated Local Search?
The perturbation strength is crucial - too small may generate cycles, but too big may cause good properties of the local optima to be lost
Acceptance Criteria:
Extreme in terms of exploitation - accept only improving solutions
Extreme in terms of exploration - accept any solution
Otherwise, can use Simulated Annealing for move acceptance
Memory - very simple use - restart search if for a number of iterations no improved solution is found
How does Iterated Local Search work?
Generate an initial solution
Apply a number of random bit flips to the solution
Perform local search i.e. using something like Steepest Descent Hill Climbing
Perform move acceptance i.e. accept improving and equal moves
What is Tabu Search?
A stochastic local search algorithm which heavily relies on the use of an explicit memory of the search process:
Systematic use of memory to guide search process
Memory typically contains only specific attributes of previously seen solutions
Simple tabu search strategies exploit only short term memory, whereas more complex exploit long term memory
How does Tabu Search work?
IN each step, move to a ‘non-tabu’ best neighbouring solution (admissible neighbours) although it may be worse than the current one
To avoid cycles, tabu search tries to avoid revisiting previously seen solutions
Solutions which contain Tabu elements are forbidden for a certain number of iterations
Often, an additional aspiration criteria is used - specifies conditions under which tabu status may be overridden.
How can a variable be considered tabu, and when does it stop being considered tabu?
When a variable gets successfully flipped, mark the step count of when this happened. The variable is tabu if the number of iterations currently - the step count of the variable is <= T, where T is the length of the queue.
What is important about the Tabu Tenure?
Too low - risk of cycling
Too high - may restrict the search by too much
t=7 is often sufficient to prevent cycling, without restricting too much
What is scheduling?
Scheduling deals with the allocation of resources to tasks over given time periods and its goal is to optimise one or more objectives. The resources and tasks can take many different forms.
How can you classify a scheduling problem?
alpha|beta|gamma
alpha - machine characteristics (environments)
beta - processing/job characteristics
gamma - optimality criteria (objective to be minimised)
