集成指数梯度与网格搜索的算法公平性优化方法  

Optimizing Algorithmic Fairness Through an Integrated Framework of Exponential Gradient and Grid Search Methods

在线阅读下载全文

作  者:焦婉妮[1] JIAO Wanni(Wuhan Digital Engineering Institute,Wuhan 430205)

机构地区:[1]武汉数字工程研究所,武汉430205

出  处:《舰船电子工程》2025年第1期132-134,140,共4页Ship Electronic Engineering

摘  要:论文创新地提出了集成指数梯度和网格搜索算法公平性的优化框架,实现了最优的动态参数调整,即在初步的指数梯度缩减后,根据模型的初步表现来调整网格搜索的范围和粒度。这种方法能够在不牺牲模型性能的前提下提高优化效率,减少了需要探索的参数空间大小。提出的集成框架实现了精确控制优化路径和调节不同公平性约束的权重,与传统算法相比,论文方法在广义熵指数、均等机会差异和平均奇偶校验差异等公平性测度指标上显示出了更优的性能。实验中采用了成年人经济收入公开数据集进行了验证,表明该框架在加速收敛和提高决策公平性方面有显著效果,展示了其在复杂多变环境下机器学习模型公平性的适应性和高效性。This paper innovatively proposes an integrated framework for optimizing fairness in algorithms using exponential gradient reduction and grid search,achieving optimal dynamic parameter adjustment.After initial exponential gradient reduction,the range and granularity of grid search are adjusted based on the preliminary performance of the model.This method enhances opti⁃mization efficiency without sacrificing model performance by reducing the size of the parameter space that needs to be explored.The proposed integrated framework allows for precise control of the optimization path and adjustment of weights for different fairness con⁃straints.Compared to traditional algorithms,our method shows superior performance in fairness metrics such as generalized entropy index,equal opportunity difference,and average odds difference.Experiments conduct on the adult income public dataset demon⁃strate significant improvements in convergence speed and decision-making fairness,showcasing the adaptability and efficiency of the model fairness framework in complex and varying environments.

关 键 词:算法公平性 指数梯度缩减 网格搜索 机器学习 公平性测度指标 

分 类 号:TP181[自动化与计算机技术—控制理论与控制工程]

 

参考文献:

正在载入数据...

 

二级参考文献:

正在载入数据...

 

耦合文献:

正在载入数据...

 

引证文献:

正在载入数据...

 

二级引证文献:

正在载入数据...

 

同被引文献:

正在载入数据...

 

相关期刊文献:

正在载入数据...

相关的主题
相关的作者对象
相关的机构对象