Scholarlink

This chapter outlines a practical path toward fully explainable FL: we connect client-level attributions and visual explanations (e.g., LayerCAM-based heat maps) with representation-level diagnostics and lightweight, server-side screening that informs aggregation, fairness, and privacy decisions. Emphasizing deployment-ready tools with minimal overhead, we show how explainability improves robustness to poisoning, accelerates convergence under Non-IID data, and supports energy- and carbon-aware orchestration, thereby turning transparency from a reporting afterthought into a core mechanism for trustworthy FL.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Towards Fully Explainable Federated Learning

  • 章节

  • 发表日期：

    2026年

查看全文
  • Kai Li,
  • Xin Yuan,
  • Wei Ni

摘要

This chapter outlines a practical path toward fully explainable FL: we connect client-level attributions and visual explanations (e.g., LayerCAM-based heat maps) with representation-level diagnostics and lightweight, server-side screening that informs aggregation, fairness, and privacy decisions. Emphasizing deployment-ready tools with minimal overhead, we show how explainability improves robustness to poisoning, accelerates convergence under Non-IID data, and supports energy- and carbon-aware orchestration, thereby turning transparency from a reporting afterthought into a core mechanism for trustworthy FL.