Evolutionary Fitness Calculation with Dynamic Abort Timing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In genetic algorithm (GA) based evolutionary computation for optimizing the structure of optical SAW filters, the parallelization of fitness calculations leads to uneven calculation times among individuals, resulting in increased waiting times and reduced speed of evaluations due to fluctuations in calculation times, especially when neural-network (NN) prediction is used.
Innovation Solution
The proposed solution involves an arithmetic processing unit that aborts the eigen solution search for individuals with long calculation times and shifts them to the next generation, maintaining the variety of eigen solutions while reducing waiting times by controlling the abort timing based on the number of searches or calculation times.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If parallelization of fitness calculations is implemented, then calculation speed is improved, but waiting times increase due to uneven calculation times among individuals
Solution Approach 1:
The patent implements dynamic abort timing control where the threshold for aborting eigen solution searches is adjusted based on the generation number and calculation statistics. This allows the system to adaptively balance between completing calculations for all individuals and reducing waiting times, resolving the contradiction between calculation speed and waiting time in parallelized fitness evaluation
Solution Approach 2:
The system monitors calculation times across individuals and uses this feedback to dynamically adjust abort thresholds. By tracking the distribution of calculation times and adjusting the abort criterion accordingly, the system optimizes the balance between maintaining calculation speed through parallelization and reducing waiting times caused by uneven individual calculation durations
2Measurement precision
If eigen solution search is completed for all individuals, then fitness calculation accuracy is improved, but calculation time fluctuates significantly
Solution Approach 1:
The patent changes the parameter of eigen solution search completeness by introducing an abort mechanism. Instead of uniformly completing searches for all individuals, the system dynamically adjusts the search completion criterion based on calculation time thresholds, thereby controlling the trade-off between fitness calculation accuracy and total calculation time
Solution Approach 2:
The system applies partial action by allowing eigen solution searches to be aborted before completion for certain individuals. This partial completion strategy prevents excessive calculation time while maintaining sufficient accuracy for fitness evaluation, resolving the contradiction between complete search accuracy and manageable calculation time
3Loss of time
If abort timing is controlled based on number of searches, then waiting times are reduced, but variety of eigen solutions may be limited
Solution Approach 1:
The abort threshold is made dynamic rather than fixed, adjusting based on generation number and statistical properties of calculation times. This dynamic adjustment ensures that in early generations where diversity is crucial, more searches are completed, while in later generations where convergence is important, the abort mechanism more actively reduces waiting times, thus balancing variety preservation with time reduction
Data Source
AI summary
An arithmetic processing apparatus includes a memory and a processor. The processor coupled to memory and configured to determine an individual not to be evolved to an individual of a second generation from among a plurality of individuals in a first generation based on a predetermined reference for calculation completion of fitness calculation for each of the plurality of individuals, the second generation being a generation next to the first generation, and determine to cause the determined individual to evolve to an individual of a generation next or subsequent to the second generation.