Back to feed
arXiv cs.LG
arXiv cs.LG
6/26/2026
\chisao{}: A GPU-Native Parallel Optimizer for Multimodal Black-Box Functions via Convergence-Anticonvergence Oscillation

\chisao{}: A GPU-Native Parallel Optimizer for Multimodal Black-Box Functions via Convergence-Anticonvergence Oscillation

Short summary

ChiSao is a GPU-native parallel optimizer that finds all modes of multimodal black-box functions through convergence-anticonvergence oscillation—freezing confirmed peaks while exploring via adaptive momentum. On 42 benchmark functions, it achieves 100% mode recovery where CPU methods fail at high dimensions, with 34×-39× speedups. Open-source on PyPI.

  • •GPU-native algorithm exploits parallelism by oscillating between convergence (freeze modes) and anti-convergence (adaptive exploration)
  • •100% mode recovery on benchmarks where basin-hopping and CMA-ES collapse at dimensions ≥8; 34×-39× speedup over baselines
  • •Available open-source on PyPI; handles derivative-free black-box functions with noise robustness

Generated with AI, which can make mistakes.

#research-breakthrough#ai-tools#open-source
Read full article at arXiv cs.LG

Is this a good recommendation for you?

Explore more

Performance Analysis and Optimization of 3D Generative Diffusion Models across GPU Architectures
arXiv cs.LG
MODE: Modality-Decomposed Expert-Level Mixed-Precision Quantization for MoE Multimodal LLMs
arXiv cs.LG
Beyond Trajectory Imitation: Strategy-Guided Policy Optimization for LLM Reasoning
arXiv CS.AI
BaLoRA: Bayesian Low-Rank Adaptation of Large Scale Models
arXiv cs.LG