Week 10: Particle Swarm Algorithms #2 Flashcards
What are some applications to PSO algorithms?
- FPGA placement
- Neural network training
- Robot motion planning
- Clustering
How does PSO compare to SA in terms of convergence?
What about for problems with larger dimensionality?
- PSO has a faster convergence than SA
- For problems with larger dimensionality, PSO gets stuck in local minima
How does PSO compare to SA in terms of initial solution?
What about for computational cost?
- PSO will start at a better solution than SA due to the initialization of many particles
- PSO also has a higher computational cost than SA due to handling of a complete swarm
What is the motivation behind using adaptive PSO in terms of parameters?
- The purpose is to have a user call PSO without having to specify any parameters
- The adaptive algorithm would tune the parameters according to feedback from the environment
In Adaptive PSO, what does a “tribe” refer to?
- A tribe is a group of connected particles
In Adaptive PSO:
What do tribes need in between them?
What is the reason for this?
- All the tribes should have some type of connection between them to inform one another of their findings
- This will help in deciding which is the global minimum among all the different solutions that was found by the different tribes
In PSO, what does a “good” particle mean?
What does “G” refer to in context to a tribe?
- A “good” particle is a particle that has its pbest improved in the last iteration, otherwise, it’s “neutral”
- “G” is the number of good particles in a tribe
In Adaptive PSO, what does each particle memorize?
What is an “excellent” particle known as?
- Each particle memorizes the last two performance variations
- A particle with both variations as improvements is an excellent particle
In Adaptive PSO, under what condition is a tribe marked as “good” or “bad”?
A tribe is marked as “good” depending on the value of “G”:
r = uniform random number between 0 and 1 T = number of particles in the tribe
If (r < G/T): tribe is “good”
otherwise: tribe is “bad”
In Adaptive PSO, what happens at the end of a generation for:
- A good tribe
- A bad tribe
- A Good tribe deletes its worst particle to conserve the number of performed function evaluations
- A Bad tribe generates a new random particle simultaneously, and all new particles form a new tribe
In Adaptive PSO, whats the general idea behind forming new tribes?
- All new particles form a new tribe
- Each particle gets connected to the tribe that generated it through its best particle
- The idea is to start with a single particle, and then generate other particles to form other tribes accordingly
What happens in Adaptive PSO in terms of tribe creation if:
- The algorithm isn’t performing well?
- The algorithm is performing very well?
- Larger and larger tribes will be generated to increase the swarm search power
- Good tribes will start to occur, and they will start removing their worst particles reducing the tribes size, possibly to complete extinction
What are the 3 general approaches for cooperative PSO?
- Concurrent PSO
- Cooperative PSO
- Hybrid Cooperative PSO
Describe the Concurrent PSO approach
- Two different swarms are updated in parallel, both using different algorithms
- The swarms exchanged their gbest values every pre-determined number of iterations
- Both swarms track the better gbest
Describe the Cooperative PSO approach (CPSO)
- Have different swarms optimizing different variables of the problem (different dimensions of the solution)
- The fitness of any particle is determined by its value and the value of the best particles in all other swarms
- Performs best if the problem variables are independent
- In order to generate a solution, combine the highest fitness individuals from each swarm
