{"slug": "newton-matching-for-generative-modeling-a-unified-framework-for-fine-tuning-and", "title": "Newton Matching for Generative Modeling: A Unified Framework for Fine-Tuning and Sampling", "summary": "A new arXiv paper (2609.05727v1) introduces Newton Matching, a unified framework that treats fine-tuning and sampling in generative modeling as iterative optimization over canonical models rather than isolated losses. The authors prove that the reverse-KL Hessian equals the transported Fisher-Rao metric, so the Newton direction coincides with the negative Fisher-Rao gradient, yielding strict reverse-KL descent for step sizes 0 < η ≤ τ, global convergence under mild conditions, and local quadratic convergence for full steps (η = τ). The framework produces sample-wise tangential-update losses without importance sampling or full-trajectory backpropagation and recovers representative existing methods as exact realizations, critical-point-consistent approximations, or objective-altering variants.", "body_md": "arXiv:2609.05727v1 Announce Type: new \nAbstract: We develop Newton Matching, a unified framework for fine-tuning and sampling in generative modeling. The target is $\\pi\\propto\\mu e^{\\tau r}$, where $r$ is the reward, $\\tau>0$ the inverse temperature, and $\\mu$ denotes the pretrained model's terminal density for fine-tuning or the constant $1$ for sampling. We shift the paradigm from isolated losses to iterative optimization over canonical models: population minimizers of standard conditional matching for terminal densities. Under compatible smooth-realization assumptions, canonical velocities form a manifold diffeomorphic to the density manifold. Transporting the Fisher-Rao metric and mixture connection to this manifold, we show that the reverse-KL Hessian equals the metric, so the Newton direction coincides with the negative Fisher-Rao gradient. At terminal density $\\rho$, each stage takes a tangential step generated by the regularized reward $r-\\frac1\\tau\\log(\\rho/\\mu)$, followed by terminal-density-preserving canonicalization. This canonical retraction yields an exact finite-stepsize density characterization. For the ideal iteration, we prove strict reverse-KL descent away from the target for $0 < \\eta \\le \\tau$, global convergence under mild conditions, and local quadratic convergence for full steps ($\\eta=\\tau$). Covariance and gradient forms, each with forward or reverse regression-pair constructions, yield sample-wise tangential-update losses with the same population minimizer, without importance sampling or full-trajectory backpropagation. We develop approximate updates and define critical-point consistency as vanishing tangential displacement if and only if $\\rho=\\pi$. We recover representative methods as exact realizations, critical-point-consistent approximations, or objective-altering variants, enabling modular algorithm design. Our work advances the theory and algorithms of reinforcement learning for generative models.", "url": "https://wpnews.pro/news/newton-matching-for-generative-modeling-a-unified-framework-for-fine-tuning-and", "canonical_source": "https://arxiv.org/abs/2609.05727", "published_at": "2026-09-10 04:00:00+00:00", "updated_at": "2026-09-10 04:23:05.456030+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "generative-ai", "ai-research", "large-language-models"], "entities": ["Newton Matching", "arXiv", "Fisher-Rao metric", "reverse-KL Hessian"], "alternates": {"html": "https://wpnews.pro/news/newton-matching-for-generative-modeling-a-unified-framework-for-fine-tuning-and", "markdown": "https://wpnews.pro/news/newton-matching-for-generative-modeling-a-unified-framework-for-fine-tuning-and.md", "text": "https://wpnews.pro/news/newton-matching-for-generative-modeling-a-unified-framework-for-fine-tuning-and.txt", "jsonld": "https://wpnews.pro/news/newton-matching-for-generative-modeling-a-unified-framework-for-fine-tuning-and.jsonld"}}