首页 > AI前沿 > Projected Penalty-Gradient Tracking for Communication-Efficient Byzantine-Resilient Federated Learning

Projected Penalty-Gradient Tracking for Communication-Efficient Byzantine-Resilient Federated Learning

arXiv机器学习 2026-11-17 23:28 9 阅读 查看原文
用户现在需要我把给定的英文新闻正文排版成语义化HTML片段,遵循要求:每个自然段落用p包裹,重要句子用strong?不对,先仔细看原文,原文是一个长的学术摘要?首先,整体是一个段落?不对,看内容,先梳理:首先开头介绍研究的内容,然后讲现有方法的问题,然后提出框架,然后讲框架的适配和两个算法,然后是理论结果,然后是实验结果。按照要求,保留所有信息,每个自然段落用p,长文用h2h3?哦原文是一个连续的段落?不对,看用户给的正文,是一整段?不对,仔细看,用户给的正文是一段连续的英文?等下看: “We investigate communication-efficient federated learning over heterogeneous datasets in the presence of Byzantine adversaries capable of uploading arbitrary information. Penalty-based methods bound the influence of each worker through the bounded gradient of a penalty function; however, the bound is lost once the uploads are compressed, since a Byzantine worker can accumulate an unbounded state at the server. To retain the Byzantine resilience under compressed uploads, we propose a projected penalty-gradient tracking framework, where each honest worker uploads a compressed residual of its penalty gradient and the server projects each reconstructed state onto the known penalty-gradient ball. The framework accommodates any contractive compressor and yields two algorithms: fault-resilient proximal gradient with consistent sparsification (FRPG-CS), whose shared mask is protected by freshly drawn reserve rows, and lazily aggregated FRPG (LAG-FRPG), whose uploads are event-triggered. For smooth and strongly convex losses, we establish a finite-time bound on the expected objective gap, where the Byzantine error floor is independent of the compressor and the compression error decays faster than the optimization error. We also bound the expected number of lazy uploads and derive certificates that separate the finite-penalty model bias from the optimization error. Experiments on MNIST and Fashion-MNIST show that the proposed algorithms attain the highest accuracy among twelve methods under the Min-Max attack, whereas a compressed-momentum benchmark and a local-update benchmark remain preferable in attack-free accuracy and total traffic, respectively.” 哦这是一整段?不对,按照要求,每个自然段落用p,那如果有小节的话用h2h3?不对,有没有需要划分的小节?比如可以划分?等下用户说“长文用

/

划分小节”,那这段内容可以划分吗?不对,首先,先确定,有没有需要加的标题?比如首先,研究背景?不对,不对,要保留原文所有信息,不能增删改内容,所以不能随便加标题?哦不对,等下规则是“长文用

/

划分小节”,但如果原文是连续的段落,那怎么划分?不对,再仔细看规则,规则4是“保留原文所有信息,不得增删改内容”,所以不能添加原文没有的标题文字?哦对,不能增,所以不能随便加h2的标题内容,那那原文是一个自然段?不对,等下看原文的句子,是不是可以?不对,等下,再看,哦不对,用户给的正文是一段?那是不是直接用p包裹?不对,不对,再看,有没有重要的句子?比如提出的框架?不对,规则说“重要句子可用”,那可以给核心的结论或者方法加strong?不对,要准确,不能改内容,只是加标签。等下,先理清楚: 首先,原文是一个连续的段落,那按照规则,每个自然段落用p包裹,那是不是就一个

?不对,等下,有没有可以划分的?不对,不能添加原文没有的内容,所以h2/h3的标题不能自己加,因为原文没有小标题,所以不能用h2h3?哦对,因为原文没有给出小节标题,所以不能自己添加,不然就增加了内容,违反规则4。那然后,重要的句子可以加strong,比如提出框架的那句?还有实验结果那句?还有理论结果的? 等下,逐句看: 第一句:We investigate communication-efficient federated learning over heterogeneous datasets in the presence of Byzantine adversaries capable of uploading arbitrary information. 这是研究背景,不需要strong? 第二句:Penalty-based methods bound the influence of each worker through the bounded gradient of a penalty function; however, the bound is lost once the uploads are compressed, since a Byzantine worker can accumulate an unbounded state at the server. 这是现有方法的问题,不需要? 第三句:To retain the Byzantine resilience under compressed uploads, we propose a projected penalty-gradient tracking framework, where each honest worker uploads a compressed residual of its penalty gradient and the server projects each reconstructed state onto the known penalty-gradient ball. 这是提出的核心框架,可以给“we propose a projected penalty-gradient tracking framework”加strong?不对,整句?或者核心部分。 第四句:The framework accommodates any contractive compressor and yields two algorithms: fault-resilient proximal gradient with consistent sparsification (FRPG-CS), whose shared mask is protected by freshly drawn reserve rows, and lazily aggregated FRPG (LAG-FRPG), whose uploads are event-triggered. 这是框架的适配和两个算法,可以给两个算法名?不对,只要重要句子,比如“yields two algorithms”后面的? 第五句:For smooth and strongly convex losses, we establish a finite-time bound on the expected objective gap, where the Byzantine error floor is independent of the compressor and the compression error decays faster than the optimization error. 这是理论贡献,可以加strong? 第六句:We also bound the expected number of lazy uploads and derive certificates that separate the finite-penalty model bias from the optimization error. 这是另一个理论贡献。 第七句:Experiments on MNIST and Fashion-MNIST show that the proposed algorithms attain the highest accuracy among twelve methods under the Min-Max attack, whereas a compressed-momentum benchmark and a local-update benchmark remain preferable in attack-free accuracy and total traffic, respectively. 这是实验结果,重要,可以加strong? 不对,但是要注意,不能增删内容,只是加标签。那现在,原文是一整段,所以用一个

包裹,然后把重要的句子用? 等下,不对,有没有可能用户给的正文是一个自然段,所以直接

里面加strong? 等下,再仔细看规则,规则2是“每个自然段落用

包裹,长文用

/

划分小节”,但因为原文没有给出小节标题,所以不能自己添加,所以不需要用h2h3,因为原文没有可以划分的小节内容,没有小标题,所以直接用p包裹整个段落,然后重要句子加strong。 那现在,怎么加strong?比如:

We investigate communication-efficient federated learning over heterogeneous datasets in the presence of Byzantine adversaries capable of uploading arbitrary information. Penalty-based methods bound the influence of each worker through the bounded gradient of a penalty function; however, the bound is lost once the uploads are compressed, since a Byzantine worker can accumulate an unbounded state at the server. To retain the Byzantine resilience under compressed uploads, we propose a projected penalty-gradient tracking framework, where each honest worker uploads a compressed residual of its penalty gradient and the server projects each reconstructed state onto the known penalty-gradient ball. The framework accommodates any contractive compressor and yields two algorithms: fault-resilient proximal gradient with consistent sparsification (FRPG-CS), whose shared mask is protected by freshly drawn reserve rows, and lazily aggregated FRPG (LAG-FRPG), whose uploads are event-triggered. For smooth and strongly convex losses, we establish a finite-time bound on the expected objective gap, where the Byzantine error floor is independent of the compressor and the compression error decays faster than the optimization error. We also bound the expected number of lazy uploads and derive certificates that separate the finite-penalty model bias from the optimization error. Experiments on MNIST and Fashion-MNIST show that the proposed algorithms attain the highest accuracy among twelve methods under the Min-Max attack, whereas a compressed-momentum benchmark and a local-update benchmark remain preferable in attack-free accuracy and total traffic, respectively.

不对,有没有问题?等下,检查有没有增删内容,没有,只是加了strong标签,段落用p包裹。对吗? 等下,再看,有没有需要划分的小节?比如原文是不是可以分成几个部分?但是因为原文没有小标题,所以不能加h2h3,因为加了的话就需要有文字,而原文没有,所以不能加,否则就增加了内容,违反规则4。对,因为规则4是不得增删改内容,所以不能自己添加标题文字,所以h2h3