用布雷格曼距离对牛顿方法的梯度规范化
Nikita Doikov1, Yurii Nesterov2
1Institute of Information and Communication Technologies, Electronics and Applied Mathematics (ICTEAM), Catholic University of Louvain (UCLouvain), Louvain-la-Neuve, Belgium.
这项研究引入了一种使用布雷格曼距离的新型二阶优化方法,实现形问题的汇率. 它提供了一个更简单,但有效的,替代立方牛顿正规化.
科学领域:
- 优化理论 优化理论
- 数字分析 数字分析
- 机器学习 机器学习
背景情况:
- 凸的优化问题是各种科学领域的基础.
- 现有的二阶方法,如立方牛顿正规化,提供了强大的趋同保证,但可能是计算密集的.
- 需要有效的优化方案,以平衡融合率与计算简单性.
研究的目的:
- 提出一种新的二次优化方案,利用任意的非欧几里德规范和布雷格曼距离.
- 分析复合凸优化的拟议方法的收性质.
- 为现有方法提供一个计算更简单的替代方案,如立方牛顿正规化.
主要方法:
- 开发了第二阶牛顿式代,结合了布雷格曼距离.
- 调整参数的设置与梯度规范的平方根成正比.
- 根据Hessian Lipschitz连续性和各种凸度类型 (均,强) 的假设,对收率的分析.
主要成果:
- 建立了基本方案的
- 证明了三度均凸函数的全局线性收和强凸函数的局部超线性收.
- 提出了一个加速方案,其汇率为
.
结论:
- 拟议的方法提供了立方牛顿规则化的放松,保留了收性质,同时简化了子问题.
- 该方法在不同的凸度假设中显示出强大的收率.
- 调整参数的自适应搜索程序提高了实际应用性.
更多相关视频
06:45Design and Application of a Fault Detection Method Based on Adaptive Filters and Rotational Speed Estimation for an Electro-Hydrostatic Actuator
Published on: October 28, 2022
08:27Author Spotlight: Efficient Image Recognition Using Directional Gradient Histogram Technique and Support Vector Machines
Published on: January 5, 2024
相关概念视频
Newton's Law of Gravitational Attraction
Newton's law of gravitational attraction is a fundamental law of physics that governs the attraction between objects. It states that the magnitude of the gravitational force between any two objects is proportional to their masses and inversely...
Gradient and Del Operator
Residuals and Least-Squares Property
If the observed data point lies above the line, the residual is positive, and the line underestimates the actual data value for y. If the observed data point lies below the line, the residual is negative, and the line overestimates the actual data value for y.
The process of fitting the best-fit...
Newtonian Fluid: Problem Solving
A velocity gradient forms within the fluid when a Newtonian fluid is placed between two parallel plates, with...
Influence of Earth's Curvature and Atmospheric Refraction on Leveling
Gauss's Law: Problem-Solving
