首页 > AI前沿 > Who Pays for Byzantine Robustness? Validity, Data-Share Dictatorships, and Fairness When Only a Few Hospitals Collaborate

Who Pays for Byzantine Robustness? Validity, Data-Share Dictatorships, and Fairness When Only a Few Hospitals Collaborate

arXiv机器学习 2026-11-17 23:08 8 阅读 查看原文
用户需要我作为专业的新闻排版工程师,将一段纯文本的新闻正文(实际上是学术摘要)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。
    
    分析正文内容:
    正文是一段关于临床联邦学习(FL)的学术摘要,没有明显的列表、引用或代码,主要是几个自然段落和研究发现。
    
    让我分析正文结构:
    1. "Clinical federated learning (FL) collaborations typically connect only three to six hospitals, yet Byzantine-robust aggregation rules are designed and benchmarked on synthetic partitions across dozens of equal-sized clients."
    2. "We conduct an empirical audit of eight standard aggregation rules: FedAvg, coordinate-wise median, trimmed mean, Krum, Multi-Krum, Bulyan, geometric median (GeoMed), and Centered Clipping (CC)."
    3. "We evaluate both unweighted and sample-size-weighted variants across three cross-silo medical datasets from the FLamby benchmark: Fed-Heart-Disease ($n = 4$ hospitals), Fed-TCGA-BRCA ($n = 6$ regions), and Fed-ISIC2019 ($n = 6$ centres)."
    4. "We execute 54,214 training runs across enumerated attacker coalitions ($f \in \{1, 2\}$) paired with exact-twin leave-out controls that decompose performance drops into honest data loss versus adversarial attack damage."
    5. "We report three findings."
    6. "First, client-count validity checks fail to capture breakdown in skewed medical consortia: while $n = 6, f = 1$ satisfies every count condition except Bulyan's, assigning sample-size weights turns median and GeoMed into a clean-model dictatorship whenever a single hospital holds more than half the training data, and causes weighted median, GeoMed, and CC to collapse when that dominant centre attacks."
    7. "Second, exact-twin leave-out controls reveal that when the largest hospital attacks, the apparent failure of count-based robust rules is predominantly caused by the loss of honest training data (7.5 to 10.4 points on ISIC) rather than adversarial perturbation (2.3 to 3.8 points on ISIC)."
    8. "Conversely, weight-dependent rules suffer large true attack damage (16.1 to 25.1 points on ISIC)."
    9. "Third, unweighted robust rules consistently discount whichever institution generates the most atypical parameter updates, penalizing the smallest hospital on Heart but penalizing larger institutions on TCGA and ISIC."
    
    我需要将其合理分段,并适当添加小标题来划分小节,因为正文较长。
    
    考虑结构:
    - 引言/背景
    - 方法/评估
    - 研究发现
    
    但规则要求不得增删改内容,所以小标题不能添加原文没有的文字。原文没有明确的小标题,所以我不能凭空创造小标题文字。那么我应该用

    包裹每个自然段落,如果需要划分小节,我可以用

    /

    ,但标题内容必须来自原文或者不添加额外文字? 规则说"长文用

    /

    划分小节",但没有说必须添加原文没有的标题。为了遵守"不得增删改内容",我不应该添加原文中没有的小标题文字。所以最好就只用

    来包裹段落,把原文的每个自然段落分开。 原文看起来可以分成几个逻辑段落: 段落1:背景和问题 "Clinical federated learning (FL) collaborations typically connect only three to six hospitals, yet Byzantine-robust aggregation rules are designed and benchmarked on synthetic partitions across dozens of equal-sized clients." 段落2:研究内容和评估方法 "We conduct an empirical audit of eight standard aggregation rules: FedAvg, coordinate-wise median, trimmed mean, Krum, Multi-Krum, Bulyan, geometric median (GeoMed), and Centered Clipping (CC). We evaluate both unweighted and sample-size-weighted variants across three cross-silo medical datasets from the FLamby benchmark: Fed-Heart-Disease ($n = 4$ hospitals), Fed-TCGA-BRCA ($n = 6$ regions), and Fed-ISIC2019 ($n = 6$ centres)." 段落3:实验执行 "We execute 54,214 training runs across enumerated attacker coalitions ($f \in \{1, 2\}$) paired with exact-twin leave-out controls that decompose performance drops into honest data loss versus adversarial attack damage." 段落4:研究发现总起 "We report three findings." 段落5:发现一 "First, client-count validity checks fail to capture breakdown in skewed medical consortia: while $n = 6, f = 1$ satisfies every count condition except Bulyan's, assigning sample-size weights turns median and GeoMed into a clean-model dictatorship whenever a single hospital holds more than half the training data, and causes weighted median, GeoMed, and CC to collapse when that dominant centre attacks." 段落6:发现二 "Second, exact-twin leave-out controls reveal that when the largest hospital attacks, the apparent failure of count-based robust rules is predominantly caused by the loss of honest training data (7.5 to 10.4 points on ISIC) rather than adversarial perturbation (2.3 to 3.8 points on ISIC). Conversely, weight-dependent rules suffer large true attack damage (16.1 to 25.1 points on ISIC)." 段落7:发现三 "Third, unweighted robust rules consistently discount whichever institution generates the most atypical parameter updates, penalizing the smallest hospital on Heart but penalizing larger institutions on TCGA and ISIC." 关于"重要句子可用",我可以考虑把关键研究发现或核心结论用强调。但要小心不要改变内容,只是用标签包裹。 关于数学公式 $n = 4$ 等,这是文本,我可以保留原样。 现在考虑是否需要列表、引用或代码。原文没有这些元素,所以不需要使用

      、
      、
      。
      
      关于划分小节,由于不能添加原文没有的文字作为标题,我可以不用

      /

      ,或者用已有的关键词?但规则说"长文用

      /

      划分小节",如果不用标题,可能不符合要求。但添加新文字又违反"不得增删改内容"。 折中方案:我可以不使用

      /

      ,因为原文没有明确的章节标题,强制添加会违反规则4。只用

      包裹每个自然段落即可。或者,我可以把"We report three findings."作为

      ?但它是句子,不是标题。 更合理的解释是:规则允许使用

      /

      ,但如果原文没有合适的标题内容,就不使用。为了遵守"不得增删改内容",我应该避免添加任何原文没有的文字作为标题。 所以最终结构:多个

      标签,每个包含一个或多个自然段落。重要句子可以用。 让我确定哪些是重要句子: - "We report three findings." 可能不需要强调 - 每个发现的核心结论可以强调 但为了不过度处理,我可以选择强调关键发现句。 让我构建HTML:

      Clinical federated learning (FL) collaborations typically connect only three to six hospitals, yet Byzantine-robust aggregation rules are designed and benchmarked on synthetic partitions across dozens of equal-sized clients.

      We conduct an empirical audit of eight standard aggregation rules: FedAvg, coordinate-wise median, trimmed mean, Krum, Multi-Krum, Bulyan, geometric median (GeoMed), and Centered Clipping (CC). We evaluate both unweighted and sample-size-weighted variants across three cross-silo medical datasets from the FLamby benchmark: Fed-Heart-Disease ($n = 4$ hospitals), Fed-TCGA-BRCA ($n = 6$ regions), and Fed-ISIC2019 ($n = 6$ centres).

      We execute 54,214 training runs across enumerated attacker coalitions