首页 > AI前沿 > Thompson Sampling using Prior-fitted Diffusion Transformers

Thompson Sampling using Prior-fitted Diffusion Transformers

arXiv机器学习 2026-11-11 03:07 20 阅读 查看原文
用户要求将一段纯文本新闻正文(实际上是一篇学术论文的摘要)排版成语义化的HTML片段。 规则: 1. 只输出HTML片段本身,不要Markdown围栏、不要解释。 2. 每个自然段落用`

`包裹,长文用`

`/`

`划分小节。 3. 列表内容用`
  • `,引用用`
    `,代码用`
    `,重要句子可用``。
    4. 保留原文所有信息,不得增删改内容。
    5. 不要输出``/``/``等文档级标签。
    
    正文内容:
    "Thompson sampling is natural randomized strategy for optimizing unknown functions in a black-box manner. In continuous global optimization, Thompson sampling has largely been restricted to Gaussian process priors, as these are essentially the only class for which computing the posterior distribution has been tractable in practice. In this work, we introduce \emph{prior-fitted diffusion transformers}, which are diffusion models that perform Thompson sampling in-context at inference time, after being pre-trained using the prior-fitted network paradigm. To apply these模型,we develop a set of practical non-Gaussian priors for general black-box functions---including those that have characteristics that cannot be achieved under Gaussianity, such as priors over unimodal functions. We also quantify the tradeoff between model size and the number of pre-training iterations, and show that these models follow scaling laws akin to those observed in transformers defined over other modalities. Across a set of global optimization benchmarks, we show that prior-fitted diffusion transformers with non-Gaussian priors can achieve improved performance compared to traditional pipelines built on Gaussian processes."
    
    分析正文:
    这是一段连续的文本,没有明显的列表、引用或代码。
    里面有LaTeX的 `\emph{prior-fitted diffusion transformers}`,在HTML中应该转为 `prior-fitted diffusion transformers`,但规则说“不得增删改内容”,LaTeX命令转成HTML语义标签算修改吗?通常应该保留文本,把`\emph{...}`变成`...`,或者保留纯文本。为了安全,用`prior-fitted diffusion transformers`,因为`\emph`就是斜体/强调的意思。或者直接输出文本,去掉`\emph{}`,但“不得增删改”可能意味着最好保留字面,或者合理转义。`prior-fitted diffusion transformers` 是合理的HTML转换。
    
    段落划分:
    整段可以作为一个段落,或者分成几个小节。
    由于是一段摘要,可以加一个`

    `作为标题(但原文没有标题,规则说“长文用h2/h3划分小节”,这里不长,就一个自然段)。 那就直接用一个`

    `包裹。 或者把“In this work, we introduce...”等作为重要句子用``?规则说“重要句子可用``”,我可以挑一句作为重要句子。比如“we introduce prior-fitted diffusion transformers, which are diffusion models that perform Thompson sampling in-context at inference time, after being pre-trained using the prior-fitted network paradigm.” 或者整段不加strong,只是用`

    `。 让我们仔细看: "Thompson sampling is natural randomized strategy for optimizing unknown functions in a black-box manner. In continuous global optimization, Thompson sampling has largely been restricted to Gaussian process priors, as these are essentially the only class for which computing the posterior distribution has been tractable in practice. In this work, we introduce \emph{prior-fitted diffusion transformers}, which are diffusion models that perform Thompson sampling in-context at inference time, after being pre-trained using the prior-fitted network paradigm. To apply these models, we develop a set of practical non-Gaussian priors for general black-box functions---including those that have characteristics that cannot be achieved under Gaussianity, such as priors over unimodal functions. We also quantify the tradeoff between model size and the number of pre-training iterations, and show that these models follow scaling laws akin to those observed in transformers defined over other modalities. Across a set of global optimization benchmarks, we show that prior-fitted diffusion transformers with non-Gaussian priors can achieve improved performance compared to traditional pipelines built on Gaussian processes." 处理 `\emph{prior-fitted diffusion transformers}` -> `prior-fitted diffusion transformers` 可以分成几个段落: 段落1:Thompson sampling is natural randomized strategy for optimizing unknown functions in a black-box manner. In continuous global optimization, Thompson sampling has largely been restricted to Gaussian process priors, as these are essentially the only class for which computing the posterior distribution has been tractable in practice. 段落2:In this work, we introduce prior-fitted diffusion transformers, which are diffusion models that perform Thompson sampling in-context at inference time, after being pre-trained using the prior-fitted network paradigm. 段落3:To apply these models, we develop a set of practical non-Gaussian priors for general black-box functions---including those that have characteristics that cannot be achieved under Gaussianity, such as priors over unimodal functions. 段落4:We also quantify the tradeoff between model size and the number of pre-training iterations, and show that these models follow scaling laws akin to those observed in transformers defined over other modalities. Across a set of global optimization benchmarks, we show that prior-fitted diffusion transformers with non-Gaussian priors can achieve improved performance compared to traditional pipelines built on Gaussian processes. 这样满足“每个自然段落用

    包裹”,“重要句子可用”。 注意原文本中的 `---` 保留。 不要修改原文的拼写或内容,比如 "is natural randomized strategy"(原文少了a,可能是 "is a natural randomized strategy",但我必须保留原文,不得增删改内容,所以保留 "is natural randomized strategy")。 检查HTML:

    Thompson sampling is natural randomized strategy for optimizing unknown functions in a black-box manner. In continuous global optimization, Thompson sampling has largely been restricted to Gaussian process priors, as these are essentially the only class for which computing the posterior distribution has been tractable in practice.

    In this work, we introduce prior-fitted diffusion transformers, which are diffusion models that perform Thompson sampling in-context at inference time, after being pre-trained using the prior-fitted network paradigm.

    To apply these models, we develop a set of practical non-Gaussian priors for general black-box functions---including those that have characteristics that cannot be achieved under Gaussianity, such as priors over unimodal functions.

    We also quantify the tradeoff between model size and the number of pre-training iterations, and show that these models follow scaling laws akin to those observed in transformers defined over other modalities. Across a set of global optimization benchmarks, we show that prior-fitted diffusion transformers with non-Gaussian priors can achieve improved performance compared to traditional pipelines built on Gaussian processes.

    确认没有文档级标签,没有Markdown围栏,没有解释。 `\emph{...}` 转为 `...` 是HTML语义化的正确做法,不算增删改文本内容,只是排版。 如果严格保留 `\emph{}`,那在HTML里会很奇怪。作为“专业的新闻排版工程师”,转换为 `` 是合理的。 再核对一遍原文: Thompson sampling is natural randomized strategy for optimizing unknown functions in a black-box manner. (保留) In continuous global optimization, Thompson sampling has largely been restricted to Gaussian process priors, as these are essentially the only class for which computing the posterior distribution has been tractable in practice. (保留) In this work, we introduce \emph{prior-fitted diffusion transformers}, which are diffusion models that perform Thompson sampling in-context at inference time, after being pre-trained using the prior-fitted network paradigm. (转为 In this work, we introduce prior-fitted diffusion transformers, which are diffusion models that perform Thompson sampling in-context at inference time, after being pre-trained using the prior-fitted network paradigm.) To apply these models, we develop a set of practical non-Gaussian priors for general black-box functions---including those that have characteristics that cannot be achieved under Gaussianity, such as priors over unimodal functions. (保留) We also quantify the tradeoff between model size and the