Helios: A Co-designed Landscape-Aware Optimization System Bridging Serial Intelligence and GPU Parallelism

The efficacy of state-of-the-art optimization algorithms often stems from complex, serial decision-making logic that is fundamentally incompatible with the throughput-oriented nature of modern GPU architectures. This conflict presents a major bottleneck for solving large-scale problems in scientific computing. We present Helios (Heterogeneous, Evolutionary, and Landscape-aware Interacting Optimization System) to address this algorithm–architecture challenge through two co-designed implementations: Helios-AS (Adaptive Serial), a novel CPU-based algorithm that employs heterogeneous agent roles and an adaptive control mechanism that dynamically triggers specialized search procedures based on real-time analysis of the fitness landscape; and Helios-MP (Massively Parallel), its GPU-native counterpart. Our core contribution is a co-design approach for translating the intent of serial adaptive logic into parallel-friendly operators, enabling complex control behavior to be re-expressed using high-throughput, population-level mechanisms. Validated on a suite of challenging benchmarks and the Lennard-Jones atomic cluster problem, we show that Helios-AS achieves solution quality comparable to established state-of-the-art methods such as CMA-ES. Furthermore, we show that Helios-MP matches this high solution quality while delivering speedups of up to 15. Helios provides both a validated optimization system for computational science and a concrete blueprint for mapping adaptive algorithmic behavior to massively parallel hardware.