{
  "abstract": "Actor--critic methods rely on critic gradients to improve continuous-action policies, yet finite-sample guarantees must control both value and gradient errors and their propagation through the actor. We develop a neural actor-critic procedure that combines a regularized minimax Bellman estimator over sigmoid networks with a finite reflected Langevin sampler driven by the fitted critic gradient. Under uniform Bellman regularity and covered reset sampling, a positive high-order action-derivative penalty yields value and action-gradient risk bounds with classical nonparametric minimax exponents up to logarithmic factors. We propagate estimation, mixing, discretization, temperature, and policy-iteration errors to obtain return guarantees, with refinements for bounded multimodal potentials and strongly concave regimes. Empirical studies across controlled simulations, logged continuous-control data, and a simulated decision task show favorable performance relative to the considered baselines.",
  "abstract_shown": true,
  "affiliations": [],
  "article_type": null,
  "arxiv": "2611.03352",
  "author_total": 4,
  "authors": [
    {
      "name": "Kunqi Huang"
    },
    {
      "name": "Lican Kang"
    },
    {
      "name": "Hongrui Shu"
    },
    {
      "name": "Jincheng Ying"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/c0nxdv5dzxzwb.bib",
  "cited_by": 0,
  "doi": "10.48550/arxiv.2611.03352",
  "fulltext_url": null,
  "id": "c0nxdv5dzxzwb",
  "issns": [],
  "landing_url": "https://arxiv.org/abs/2611.03352",
  "language": null,
  "licence": "LicenseRef-arXiv-nonexclusive-1.0",
  "licence_label": "arXiv licence",
  "licence_source": "article",
  "licence_url": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
  "metadata_licence": "CC0-1.0",
  "native_subjects": [
    [
      "arxiv",
      "cs.LG"
    ],
    [
      "arxiv",
      "stat.ML"
    ]
  ],
  "openalex": null,
  "pdf_url": "https://arxiv.org/pdf/2611.03352",
  "pmcid": null,
  "pmid": null,
  "published": "2026-11-02",
  "publisher": null,
  "ris": "https://research.draconitic.com/paper/c0nxdv5dzxzwb.ris",
  "sources": [
    {
      "first_seen": "2026-11-03T12:21:21Z",
      "native_id": "2611.03352",
      "source": "arxiv",
      "url": "https://arxiv.org/abs/2611.03352"
    }
  ],
  "subjects": [
    {
      "field": "mathematics",
      "origin": "arxiv:stat.ML",
      "subfield": "statistics"
    }
  ],
  "title": "Neural Actor--Critic Learning with Minimax Critics and Reflected Langevin Sampling",
  "title_text": "Neural Actor--Critic Learning with Minimax Critics and Reflected Langevin Sampling",
  "type": "preprint",
  "updated_at": "2026-11-09T00:20:32Z",
  "url": "https://research.draconitic.com/paper/c0nxdv5dzxzwb",
  "venue": "arXiv",
  "versions": []
}