诱导偏差的几何学:一个完全可解决的数据模型,具有公平性影响
Stefano Sarao Mannelli1, Federica Gerace2, Negar Rostamzadeh3
1University of Gothenburg, Chalmers University of Technology, Data Science and AI, Computer Science and Engineering, Gothenburg, Sweden and School of Computer Science and Applied Mathematics, University of the Witwatersrand, Johannesburg, South Africa.
Physical review. E
|September 16, 2025
概括
机器学习 (ML) 模型可以通过训练数据来延续人类偏见. 这项研究使用一种可解决的模型来分析偏见的出现和测试缓解策略,发现合模型提供了更好的公平性准确性.
科学领域:
- 统计物理 统计物理
- 机器学习 机器学习
- 数据科学数据科学数据科学
背景情况:
- 机器学习 (ML) 模型可以继承和放大训练数据中存在的人类偏见.
- 了解偏差遗传机制对于开发公平的人工智能系统至关重要.
研究的目的:
- 调查数据几何学在ML偏差出现中的作用.
- 分析性地描述偏差遗传和评估缓解策略.
主要方法:
- 开发一个精确可解决的数据不平衡的高维模型.
- 应用统计物理工具来分析ML模型属性.
- 评估损失权衡和合学习模式策略.
主要成果:
- 从模型中获得的公平性评估可观察值的分析预测.
- 确定不同公平标准之间的不兼容性.
- 证明结合式学习模型实现了卓越的公平性-准确性权衡.
结论:
- 数据几何学显著影响了ML偏差.
- 现有的公平标准可能不兼容.
- 建议的结合学习策略提供了一种有效的方法来减轻ML模型中的偏差.
相关概念视频
Bias
7.2K
Bias refers to any tendency that prevents a question from being considered unprejudiced. In research, bias occurs when one outcome or answer is selected or encouraged over others in sampling or testing. Bias can occur during any research phase, including study design, data collection, analysis, and publication.
In statistics, a sampling bias is created when a sample is collected from a population, and some members of the population are not as likely to be chosen as others (remember, each member...
In statistics, a sampling bias is created when a sample is collected from a population, and some members of the population are not as likely to be chosen as others (remember, each member...
7.2K
Gauss's Law: Planar Symmetry
9.3K
A planar symmetry of charge density is obtained when charges are uniformly spread over a large flat surface. In planar symmetry, all points in a plane parallel to the plane of charge are identical with respect to the charges. Suppose the plane of the charge distribution is the xy-plane, and the electric field at a space point P with coordinates (x, y, z) is to be determined. Since the charge density is the same at all (x, y) - coordinates in the z = 0 plane, by symmetry, the electric field at P...
9.3K
Graphical Representation of Inequalities
175
The graph of the equation where y equals x squared forms a curve known as a parabola. This curve acts as a boundary in the coordinate plane, dividing it into distinct regions based on the relative position of points.When the equality sign in the equation is replaced with an inequality—such as greater than, less than, greater than or equal to, or less than or equal to—the graphical representation changes from a single curve into a broader shaded area that signifies the set of all...
175
Gaussian Elimination: Problem Solving
162
Systems of linear equations in several variables are pivotal in modeling complex scenarios involving multiple unknowns and constraints. Such systems are widely used in various fields to represent relationships where several conditions must be simultaneously satisfied. Each variable in the system corresponds to an unknown quantity, while each equation imposes a linear constraint, leading to a structured approach for analyzing and solving real-world problems.A system of three equations with three...
162
Unsymmetric Loading of Thin-Walled Members: Problem Solving
499
The shear center of a channel section with uniform thickness, height, and width, is determined by computing the shear force in the member and calculating the moments of inertia of the sections.
To compute the shear forces, find the shear flow at a specific distance from the endpoint using the vertical shear and the moment of inertia values. The total shear force on the flange is calculated by integrating the shear flow from one end of the flange to the other.
Next, calculate the moments of...
To compute the shear forces, find the shear flow at a specific distance from the endpoint using the vertical shear and the moment of inertia values. The total shear force on the flange is calculated by integrating the shear flow from one end of the flange to the other.
Next, calculate the moments of...
499
Application of Nonlinear Inequalities
214
A nonlinear inequality describes a comparison involving an expression that curves or behaves more complexly than a straight line. These inequalities often appear in forms that include squares, products, or variables in the denominator.To solve such an inequality, one starts by rewriting it so that zero appears on one side. For example, the inequality: can be factored as: This form makes it easier to identify the values that cause the expression to equal zero. In this case, the...
214


