Search papers, labs, and topics across Lattice.
This study critiques the limitations of existing global optimization benchmarks, which are outdated and biased due to their reliance on a narrow set of analytical functions. By introducing black-box adversarial attack (BBAA) tasks as a new benchmark, the authors evaluate the performance of various evolutionary algorithms and metaheuristics in high-dimensional optimization scenarios. The findings reveal that these methods can effectively tackle BBAA problems, highlighting a significant shift towards aligning global optimization techniques with contemporary machine learning challenges.
Black-box adversarial attacks could redefine global optimization benchmarks, revealing the true potential of evolutionary algorithms in high-dimensional spaces.
Existing global optimization benchmark suites are of a moderate size and are based on a small number of analytical functions that date back even to the 1970s. This causes a risk of biasing the development of global optimization methods. We argue that the tasks related to the black-box adversarial attack (BBAA) can serve as valuable global optimization benchmark in many-dimensional space. We demonstrate the efficiency of several types of evolutionary algorithms and other metaheuristics in solving example BBAA problems. Thus, we take a step towards convergence of global optimization methods to the challenges and needs that arise in the modern machine learning field.
