Efficient Multiplayer Battle Game Optimizer for Adversarial Robust Neural Architecture Search

TL;DR

MBGO exhibited imbalances between exploitation and exploration in its movement phase, motivating the development of EMBGO. EMBGO unifies the movement and battle phases and introduces a differential mutation toward the current best and the population mean, plus Lévy flight for outside-zone exploration, to sustain diversity and accelerate convergence. Across CEC2017/2020/2022 benchmarks, eight engineering problems, and ARNAS, EMBGO outperforms or competes aggressively with twelve established metaheuristics, demonstrating robust, scalable optimization performance. The work also analyzes computational complexity and discusses limitations in high-dimensional or certain hybrid/composite problems, outlining future enhancements such as cooperative coevolution and PSO-inspired selection to broaden applicability. The authors provide open-source code to enable reproducibility and further research in adversarially robust neural architecture search contexts.

Abstract

This paper introduces a novel metaheuristic algorithm, known as the efficient multiplayer battle game optimizer (EMBGO), specifically designed for addressing complex numerical optimization tasks. The motivation behind this research stems from the need to rectify identified shortcomings in the original MBGO, particularly in search operators during the movement phase, as revealed through ablation experiments. EMBGO mitigates these limitations by integrating the movement and battle phases to simplify the original optimization framework and improve search efficiency. Besides, two efficient search operators: differential mutation and Lévy flight are introduced to increase the diversity of the population. To evaluate the performance of EMBGO comprehensively and fairly, numerical experiments are conducted on benchmark functions such as CEC2017, CEC2020, and CEC2022, as well as engineering problems. Twelve well-established MA approaches serve as competitor algorithms for comparison. Furthermore, we apply the proposed EMBGO to the complex adversarial robust neural architecture search (ARNAS) tasks and explore its robustness and scalability. The experimental results and statistical analyses confirm the efficiency and effectiveness of EMBGO across various optimization tasks. As a potential optimization technique, EMBGO holds promise for diverse applications in real-world problems and deep learning scenarios. The source code of EMBGO is made available in \url{https://github.com/RuiZhong961230/EMBGO}.

Figures (14)

• Figure 1: Safe zones from representative multiplayer battle games. (a). The safe zone from PUBG. (b). The safe zone from APEX.
• Figure 2: Convergence curves of the fitness value on 30-D CEC2017 benchmark functions.
• Figure 3: Convergence curves the population diversity on 30-D CEC2017 benchmark functions.
• Figure 4: The flowchart of our proposed EMBGO.
• Figure 5: A visual demonstration of the differential mutation and Lévy flight in a 2-D search space is provided. The worst, best, and centroid individuals of the population are represented by blue, red, and orange dots, respectively. The light blue circle depicts the safe zone. Individuals located within the safe zone employ the differential mutation, while those outside the safe zone utilize the Lévy flight to construct offspring individuals.