{"slug": "bridging-spherical-black-box-optimizers", "title": "Bridging Spherical Black-Box Optimizers", "summary": "Researchers at ICML 2026 presented a unified framework showing that parametric and nonparametric black-box optimizers are variations of a single update equation. By bridging this theoretical gap, they developed hybrid optimizers AdaPol and SchedPol that efficiently merge foundation models, finding generalized high-quality solutions at reduced computational cost.", "body_md": "We are pleased to present our research at ICML 2026, “Bridging Spherical Black-Box Optimizers”.\n\nWhen optimizing through simulators, external APIs, or in reinforcement learning, gradients are often unavailable. Black-Box Optimization (BBO) fills this gap, but the field has been historically split into two categories:\n\n-\nParametric Methods: Algorithms like Evolution Strategies (ES) scale to high dimensions but only find a single solution.\n\n-\nNonparametric Methods: Algorithms like Consensus-Based Optimization (CBO) find multiple solutions but fail in high dimensions.\n\nOur team asked a simple question: what if they are all doing the same thing?\n\nIn our paper, we showed that these distinct families are actually variations of a single update equation. By bridging this theoretical gap, we can now engineer custom hybrid optimizers for specific tasks.\n\nA key application of this is merging foundation models. Building on our previous work in Evolutionary Model Merging, we faced a computational challenge. Evaluating large language models at every step is resource-intensive, but using a smaller evaluation dataset causes standard unimodal optimizers to overfit.\n\nBy treating LLM merging as a multimodal problem and deploying our newly developed hybrid optimizers, AdaPol and SchedPol, we successfully navigated this issue. The algorithms identified multiple distinct optima on the smaller dataset, allowing us to find generalized, high-quality merges at a fraction of the compute cost.", "url": "https://wpnews.pro/news/bridging-spherical-black-box-optimizers", "canonical_source": "https://sakana.ai/bbob/", "published_at": "2026-07-03 15:00:00+00:00", "updated_at": "2026-07-07 00:50:39.843898+00:00", "lang": "en", "topics": ["machine-learning", "ai-research", "ai-agents", "large-language-models"], "entities": ["ICML 2026", "AdaPol", "SchedPol", "Evolution Strategies", "Consensus-Based Optimization"], "alternates": {"html": "https://wpnews.pro/news/bridging-spherical-black-box-optimizers", "markdown": "https://wpnews.pro/news/bridging-spherical-black-box-optimizers.md", "text": "https://wpnews.pro/news/bridging-spherical-black-box-optimizers.txt", "jsonld": "https://wpnews.pro/news/bridging-spherical-black-box-optimizers.jsonld"}}