Particle Swarm Optimization Acceleration via Reduced Resolution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Particle swarm optimization algorithms face sequential dependencies, where the location of a particle in the next iteration cannot be determined until the end of the current iteration, leading to sequential computation and potential bottlenecks in resource utilization.
Innovation Solution
The approach involves reducing the resolution of particle locations by rounding values or eliminating insignificant figures, allowing for the determination of potential full resolution locations to be reduced to a single reduced resolution location, which can be calculated for fitness prior to the completion of the current iteration, thereby accelerating the process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If fitness calculations are performed for all particles at the current iteration before determining next iteration locations, then calculation accuracy is maintained, but computational productivity is reduced due to sequential dependencies
Solution Approach 1:
The patent applies preliminary action by determining particle locations for the next iteration before the current iteration fully completes. Specifically, when a particle's location at the next iteration can be determined independently of other particles' fitness calculations (e.g., when personal best and global best are already known), that particle's next location is calculated in advance, allowing parallel processing and improving computational throughput while maintaining accuracy for particles that require full iteration completion.
2Measurement precision
If all particle locations are determined with full resolution, then solution precision is maintained, but computational complexity increases
Solution Approach 1:
The patent applies local quality by differentiating the treatment of particles based on their individual computational dependencies. Some particles have their next iteration locations determined with full resolution immediately, while others wait for fitness calculations to complete. This localized differentiation optimizes computational complexity by avoiding unnecessary waiting for particles that can be determined independently, while maintaining full precision where required.
Data Source
AI summary
Embodiments of the inventive subject matter include determining a plurality of potential full resolution locations for a particle representation for a second iteration of a particle swarm optimization, wherein the particle representation is associated with both a first full resolution location and a first reduced resolution location for a first iteration of the particle swarm optimization that has not yet completed and the second iteration is later than the first iteration. Embodiments further include determining that the plurality of full resolution locations reduces to a second reduced resolution location for the second iteration. Embodiments further include submitting the second reduced resolution location for fitness calculation prior to the first iteration completing.