With the explosive development of artificial intelligence(AI),machine learning(ML),and high-performance comput-ing(HPC),the ever-growing data movement is asking for high density interconnects with higher bandwidth(BW)...With the explosive development of artificial intelligence(AI),machine learning(ML),and high-performance comput-ing(HPC),the ever-growing data movement is asking for high density interconnects with higher bandwidth(BW),lower power and lower latency[1−3].The optical I/O leverages silicon photonic(SiPh)technology to enable high-density large-scale integrated photonics.展开更多
文摘纵向联邦学习(vertical federated learning,VFL)常用于高风险场景中的跨领域数据共享,用户需要理解并信任模型决策以推动模型应用。现有研究主要关注VFL中可解释性与隐私之间的权衡,未充分满足用户对模型建立信任及调优的需求。为此,提出了一种基于人在回路(human-in-the-loop,HITL)的纵向联邦学习解释方法(explainable vertical federated learning based on human-in-the-loop,XVFL-HITL),通过构建分布式HITL结构将用户反馈纳入VFL的基于Shapley值的解释方法中,利用各参与方的知识校正训练数据来提高模型性能。进一步,考虑到隐私问题,基于Shapley值的可加性原理,将非当前参与方的特征贡献值整合为一个整体展示,从而有效保护了各参与方的特征隐私。实验结果表明,在基准数据上,XVFL-HITL的解释结果具有有效性,并保护了用户的特征隐私;同时,XVFL-HITL对比VFL-Random和直接使用SHAP的VFL-Shapley进行特征选择的方法,模型准确率分别提高了约14%和11%。
基金This work was supported by the National Natural Science Foundation of China(Grant Nos.61925505,92373209 and 62235017).
文摘With the explosive development of artificial intelligence(AI),machine learning(ML),and high-performance comput-ing(HPC),the ever-growing data movement is asking for high density interconnects with higher bandwidth(BW),lower power and lower latency[1−3].The optical I/O leverages silicon photonic(SiPh)technology to enable high-density large-scale integrated photonics.
