TLDR: This research introduces P-APO, a parallel version of the Artificial Protozoa Optimizer (APO) that uses NVIDIA’s CUDA architecture to leverage GPU acceleration. The study demonstrates that P-APO significantly reduces computation time, achieving up to 6.7 times speedup on benchmark functions and up to 2.67 times speedup in real-world applications like image thresholding, making the powerful APO algorithm more scalable and efficient for complex problems.
Metaheuristic algorithms are powerful tools for tackling complex problems, offering solutions that are close to optimal. However, a significant challenge with these algorithms is their increasing execution time as problem sizes and solution spaces grow. To achieve more promising results, these algorithms often require a large number of iterations, which can be very time-consuming. This issue has led researchers to explore parallel versions of these optimization algorithms to enhance their performance.
A recent study introduces a parallel implementation of the Artificial Protozoa Optimizer (APO), named P-APO, leveraging NVIDIA’s CUDA framework for GPU acceleration. The Artificial Protozoa Optimizer (APO) is a bio-inspired metaheuristic algorithm that draws inspiration from the survival behaviors of protozoa, including foraging, dormancy, and reproduction. It has demonstrated superior performance against many other state-of-the-art algorithms on challenging benchmark functions.
The core idea behind P-APO is to optimize the existing sequential APO algorithm to achieve higher performance by distributing computational tasks across multiple GPU threads. While APO is effective at finding optimal solutions for large and complex problems, its execution time can limit its scalability, especially for real-time applications. The parallel version addresses this by efficiently distributing the work of individual protozoa among GPU threads in each iteration.
The researchers implemented both the existing sequential APO and their proposed parallel P-APO. Their experimental results, tested on benchmark functions from CEC2022, showed a significant performance improvement. The P-APO achieved a speedup of up to 6.7 times compared to the sequential version. This means that tasks that previously took a long time can now be completed much faster, making the algorithm more practical for demanding applications.
Beyond benchmark functions, the P-APO was also tested on two real-world applications: Tension/Compression Spring Design in engineering optimization and Image Thresholding using Otsu’s method. While the spring design problem did not show a substantial speedup due to its relatively low computational cost, the image thresholding application demonstrated a notable reduction in execution time, achieving a speedup of up to 2.67 times. This indicates P-APO’s potential to handle real-world tasks more efficiently.
The implementation details highlight the use of a Google Colab environment with an NVIDIA Tesla T4 GPU, along with CUDA C++ for programming. The parallel approach maps each protozoa’s calculation to a CUDA thread, allowing for simultaneous updates of optimal values. This parallelization significantly reduces the overall computation time, making metaheuristic optimization more accessible and efficient for a wider range of problems.
Also Read:
- Hybrid Optimization Solver Delivers Unprecedented Speed and Trust for Power Grids
- AI Designs and Solves Game Levels in a Dynamic Unity Environment
In conclusion, the development of the Parallel Artificial Protozoa Optimizer (P-APO) marks a significant step forward in addressing the computational intensity of metaheuristic algorithms. By harnessing the power of GPU acceleration, P-APO offers a faster and more scalable solution for complex optimization problems, opening new possibilities for its application in various scientific and industrial fields. You can read the full research paper here.
