Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
184 tokens/sec
GPT-4o
7 tokens/sec
Gemini 2.5 Pro Pro
45 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
38 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

GPU-accelerated Evolutionary Many-objective Optimization Using Tensorized NSGA-III (2504.06067v1)

Published 8 Apr 2025 in cs.NE

Abstract: NSGA-III is one of the most widely adopted algorithms for tackling many-objective optimization problems. However, its CPU-based design severely limits scalability and computational efficiency. To address the limitations, we propose {TensorNSGA-III}, a fully tensorized implementation of NSGA-III that leverages GPU parallelism for large-scale many-objective optimization. Unlike conventional GPU-accelerated evolutionary algorithms that rely on heuristic approximations to improve efficiency, TensorNSGA-III maintains the exact selection and variation mechanisms of NSGA-III while achieving significant acceleration. By reformulating the selection process with tensorized data structures and an optimized caching strategy, our approach effectively eliminates computational bottlenecks inherent in traditional CPU-based and na\"ive GPU implementations. Experimental results on widely used numerical benchmarks show that TensorNSGA-III achieves speedups of up to $3629\times$ over the CPU version of NSGA-III. Additionally, we validate its effectiveness in multiobjective robotic control tasks, where it discovers diverse and high-quality behavioral solutions. Furthermore, we investigate the critical role of large population sizes in many-objective optimization and demonstrate the scalability of TensorNSGA-III in such scenarios. The source code is available at https://github.com/EMI-Group/evomo

Summary

  • The paper introduces TensorNSGA-III, a novel fully tensorized and GPU-accelerated implementation of NSGA-III designed to enhance performance for many-objective optimization problems.
  • TensorNSGA-III achieves significant computational speedups, up to 3629 times over the CPU version, by leveraging GPU parallelism and tensorized operations while maintaining the original NSGA-III selection and variation mechanisms.
  • The method demonstrates exceptional scalability for large population sizes and effectiveness in practical applications like robotic control tasks, enabling more extensive searches and broader feasibility for complex problems.

GPU-accelerated Evolutionary Many-objective Optimization Using Tensorized NSGA-III

The paper under consideration presents a novel approach to improving the performance of the Non-dominated Sorting Genetic Algorithm III (NSGA-III) for many-objective optimization problems by leveraging the computational capabilities of GPUs. The proposed method, termed TensorNSGA-III, is a fully tensorized implementation that utilizes the GPU's inherent parallelism to address scalability and efficiency challenges associated with traditional CPU-based versions.

NSGA-III is a widely recognized algorithm in solving many-objective optimization problems (MaOPs). However, its practical deployment is often hindered by its computational inefficiencies when scaled to larger problems or deployed in real-world applications requiring significant computational resources. The primary innovation of TensorNSGA-III lies in its ability to retain the exact selection and variation mechanisms of the original NSGA-III while achieving significant computational acceleration—recording speedups up to 3629 times over the CPU version. This is accomplished through a reformation of the selection process using tensorized data structures and an optimized caching strategy, effectively mitigating the bottlenecks observed in conventional CPU and naive GPU implementations.

The authors tested TensorNSGA-III across several numerical benchmarks and multiobjective robotic control tasks, confirming its efficiency in discovering diverse and high-quality solutions in multiobjective domains. The experiments demonstrated TensorNSGA-III's exceptional scalability, particularly in accommodating large population sizes, a critical aspect for MaOPs. In comparisons with the CPU version, traditional GPU versions, and other tensorized algorithms, TensorNSGA-III consistently outperformed its counterparts in computation time across varied problem instances.

Strong numerical results underpin the effectiveness of TensorNSGA-III, particularly notable in contexts such as the DTLZ benchmark problems, where even at population sizes as large as 12,800, the algorithm exhibited only marginal increases in execution time, thereby substantiating its scalability claim. Furthermore, the application in robotic control tasks, using a neuroevolutionary framework, reinforced the practical viability of the approach in real-world scenarios, showcasing superior performance in generating diverse controller policies.

The implications of this research are significant both in theoretical and applied realms of AI and optimization. From a theoretical perspective, TensorNSGA-III contributes to the ongoing discourse on enhancing evolutionary algorithms through advanced computational paradigms like tensorization. It opens avenues for further exploration into optimizing other complex algorithms using similar strategies.

Practically, the demonstrated acceleration permits more extensive and diversified solution searches, making many-objective optimization feasible for high-dimensional and computationally intensive applications that were previously constrained by performance bottlenecks. The utilization of GPU power in such a manner aligns with contemporary trends toward leveraging parallel computing to handle larger computational tasks more efficiently.

The future of AI developments in this domain may see an increased focus on integrating tensor-based computations with evolutionary strategies, enabling more powerful and efficient solutions for complex optimization challenges. The open-source dissemination of TensorNSGA-III encourages broader engagement from the research community, potentially accelerating innovations and adaptations across diverse fields requiring robust multiobjective optimization solutions.

Github Logo Streamline Icon: https://streamlinehq.com