公平联邦学习包含协作公平，表现公平，模型公平等。现有的公平联邦学习多方协作可以粗略划分为3个小组：G...

可以，下面给出一版更符合论文综述语气、逻辑也更紧凑的表述。

中文润色版：

:::writing{variant="document" id="41827" title="公平联邦学习相关工作分类"} 公平联邦学习通常涵盖协作公平（collaboration fairness）、性能公平（performance fairness）以及模型公平/模型异质性（model fairness / model heterogeneity）等多个维度。现有面向多方协作的公平联邦学习方法，大致可以划分为以下三类。

第一类（G1）关注贡献或协作公平，旨在根据不同客户端的贡献程度、参与情况或效用差异，实现更加公平的协作与收益分配，代表性方法包括 q-FFL [27]、FedUFA [28] 和 FedGCR [29]。

第二类（G2）侧重于性能公平，其核心目标是缩小不同客户端之间的模型性能差距，并提升弱势客户端或低性能客户端的学习效果。代表性工作包括 FedFV [30]、FedLF [31]、FedHEAL [25]，以及采用自适应聚合机制的 FdePW [26] 和 FedAA [32]。

第三类（G3）主要关注模型异质性条件下的公平学习，即在不同客户端具有异构模型架构、参数规模或个性化模型需求的情况下实现有效且公平的协同训练。代表性方法包括 PHP-FL [33] 和 HeteroFL-LoRA [34]。其中，PHP-FL [33] 还进一步针对客户端参与概率不一致的场景进行了较为系统和细致的研究。 :::

对应的顶会论文风格英文可以写成：

:::writing{variant="document" id="73514" title="Conference-Style English Version"} Fairness in federated learning (FL) encompasses multiple dimensions, including collaboration fairness, performance fairness, and model fairness under system/model heterogeneity. Existing studies on fairness-aware multi-party FL can be broadly categorized into three groups.

G1: Contribution/Collaboration Fairness. This line of work aims to ensure fair collaboration and reward allocation by accounting for heterogeneous client contributions, participation patterns, or utilities. Representative methods include q-FFL [27], FedUFA [28], and FedGCR [29].

G2: Performance Fairness. Methods in this category focus on mitigating performance disparities across clients, particularly by improving the learning outcomes of disadvantaged or underperforming clients. Representative approaches include FedFV [30], FedLF [31], FedHEAL [25], as well as adaptive aggregation-based methods such as FdePW [26] and FedAA [32].

G3: Fairness under Model Heterogeneity. This category investigates fair and effective collaborative learning when clients employ heterogeneous model architectures, parameter scales, or personalized model configurations. Representative methods include PHP-FL [33] and HeteroFL-LoRA [34]. Notably, PHP-FL [33] further provides a systematic investigation of FL scenarios with heterogeneous client participation probabilities. :::

如果用于 Related Work / Literature Review，英文里建议使用第二版这种 “This line of work / Methods in this category / Notably” 的写法，比逐条罗列更接近 NeurIPS、ICML、AAAI、IJCAI 等顶会常见的综述风格。