logo
科言猫学术研究的AI总结
language
手机端手机端
  • Phone Feature微信小程序
简体中文简体中文
notice消息通知
博客博客
delete
arrow
返回

FedRNL: Federated Rationalization with Soft Parameter Sharing

delete2026-01-01
delete0
PRE
AI
J
Jiahui Jiang
H
Haozhao Wang
L
Lei Wu *
R
Ruixuan Li
DOI：10.1007/978-3-032-05981-9_21delete
delete原文链接
delete原文求助
delete分享
delete收藏
摘要

摘要

En 中文
可解释性在自然语言处理中至关重要，能够增强透明度和信任度。理性化模型通过提取关键输入片段（即理性依据）来解释决策，同时保持预测性能。另一方面，联邦学习（FL）近年来已成为训练机器学习模型的关键范式，因为它可以利用来自多个客户端的训练数据，而无需上传其原始数据。基于此，我们首先提出以联邦学习的方式训练理性化模型。然而，我们发现简单地将两者结合会导致严重性能下降，这是由于客户端之间的数据异构性，其中存在不一致的理性依据生成。为解决此问题，我们提出FedRNL，该模型引入软共享机制来对齐生成器和预测器的编码器，确保浅层一致性和深层泛化。一个编码器损失最小化特征差异，而分层聚合策略分别更新服务器上的生成器和预测器，增强模型稳定性。大量实验表明，FedRNL相较于现有的通用异构性缓解方法显著提升了性能。
Keyword:
Federated learning
Rationalization
Non-IID

期刊

M
MACHINE LEARNING AND KNOWLEDGE DISCOVERY IN DATABASES. RESEARCH TRACK, ECML PKDD 2025, PT II
IF:
0
论文数:
28
被引数:
0

机构

H
huazhong university of science & technology
学者数:
6.1K
论文数: 1.6K
被引数: 0
引用论文

引用论文

From Insights to Actions: The Impact of Interpretability and Analysis Research on NLP
errProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing
IF0
err2024-01-01
err0
PREAI
errMosbach,Marius; Gautam,Vagrant; Vergara Browne,Tomás; Klakow,Dietrich; Geva,Mor
err分享
err收藏
Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation使用RNN编码器-解码器学习统计机器翻译的短语表示
errProceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP)
IF0
err2014-01-01
err0
errOAAI
errKyunghyun Cho; Bart van Merrienboer; Caglar Gulcehre; Dzmitry Bahdanau; Fethi Bougares; Holger Schwenk; Yoshua Bengio
err分享
err收藏
FedNLR: Federated Learning with Neuron-wise Learning Rates
errProceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining
IF0
err2024-08-25
err0
PREAI
errWang,Haozhao; Zheng,Peirong; Han,Xingshuo; Xu,Wenchao; Li,Ruixuan; Zhang,Tianwei
err分享
err收藏
Improving Global Generalization and Local Personalization for Federated Learning改进联邦学习的全局泛化和局部个性化
errIEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
IF8.9
err2025-01-01
err0
PREAI
errMeng, Lei; Qi, Zhuang; Wu, Lei; Du, Xiaoyu; Li, Zhaochuan; Cui, Lizhen; Meng, Xiangxu
err分享
err收藏
Can Rationalization Improve Robustness?
errProceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies
IF0
err2022-01-01
err0
PREAI
errChen,Howard; He,Jacqueline; Narasimhan,Karthik; Chen,Danqi
err分享
err收藏
Human-like Explanation for Text Classification With Limited Attention Supervision
err2021 IEEE International Conference on Big Data (Big Data)
IF0
err2021-12-15
err0
PREAI
errZhang,Dongyu; Sen,Cansu; Thadajarassiri,Jidapa; Hartvigsen,Thomas; Kong,Xiangnan; Rundensteiner,Elke
err分享
err收藏
Toward Few-Label Vertical Federated Learning
errACM TRANSACTIONS ON KNOWLEDGE DISCOVERY FROM DATA
IF4.8
err2024-06-19
err0
errOAAI
errZhang, Lei; Fu, Lele; Liu, Chen; Yang, Zhao; Yang, Jinghua; Zheng, Zibin; Chen, Chuan
err分享
err收藏
MARTA: Leveraging Human Rationales for Explainable Text Classification
errProceedings of the AAAI Conference on Artificial Intelligence
IF0
err2021-05-18
err0
errOAAI
errInes Arous; Ljiljana Dolamic; Jie Yang; Akansha Bhardwaj; Giuseppe Cuccu; Philippe Cudré-Mauroux
err分享
err收藏
Latent aspect rating analysis on review text data
errProceedings of the 16th ACM SIGKDD international conference on Knowledge discovery and data mining
IF0
err2010-07-25
err0
PREAI
errHongning Wang; Yue Lu; Chengxiang Zhai
err分享
err收藏
The Mythos of Model Interpretability模型可解释性的神话
errQueue
IF0
err2018-06-01
err0
errOAAI
errZachary C. Lipton
err分享
err收藏
学者 查看更多内容