在受限制的博尔兹曼机器中使用目标能量的比率差异学习:超越库尔巴克-莱布勒差异学习
Yuichi Ishida1, Yuma Ichikawa1,2, Aki Dote1
1Fujitsu Ltd., Kawasaki, Japan.
Physical review. E
|November 18, 2025
概括
比率分歧 (RD) 学习通过结合前进和反向Kullback-Leibler分歧 (KLD) 学习来增强离散的基于能源的模型. 这种新的方法改善了能量功能的适配和学习稳定性,特别是在复杂的模型中.
科学领域:
- 机器学习 机器学习
- 人工智能的人工智能
- 计算统计学 计算统计学
背景情况:
- 离散的基于能源的模型 (EBM) 是模拟复杂数据分布的强大工具.
- 限制波尔兹曼机器 (RBM) 是一种满足普遍近似定理的基本离散EBM.
- 现有的学习方法,如前进和反向Kullback-Leibler分歧 (KLD) 学习,具有局限性,包括不足和模式崩.
研究的目的:
- 引入Ratio Divergence (RD) 学习,这是一种用于离散基于能量模型的新方法.
- 在离散的EBM中解决传统KLD学习方法的局限性.
- 评估与现有方法相比,RD学习的有效性.
主要方法:
- 研发开发学习利用了训练数据和可处理的目标能量函数.
- RD学习结合了前进和反向KLD学习的优势.
- 使用各种离散的EBM进行了数值实验,包括RBM,以RD学习为主要方法.
主要成果:
- 在能源功能的配件,模式覆盖和学习稳定性方面,RD学习显著优于现有的方法.
- 研究与开发学习的有效性随着目标模型的维度增加而增加.
- 一种基线方法总结了前进和反向KLD,并与RD学习进行了评估.
结论:
- 研发与开发学习为训练离散的基于能源的模型提供了一种强大而有效的方法.
- RD学习成功地减轻了KLD学习固有的不足和模式崩问题.
- 拟议的RD学习方法对复杂的,高维的离散分布建模具有前景.
相关概念视频
Maxwell-Boltzmann Distribution: Problem Solving
2.8K
Individual molecules in a gas move in random directions, but a gas containing numerous molecules has a predictable distribution of molecular speeds, which is known as the Maxwell-Boltzmann distribution, f(v).
This distribution function f(v) is defined by saying that the expected number N (v1,v2) of particles with speeds between v1 and v2 is given by
This distribution function f(v) is defined by saying that the expected number N (v1,v2) of particles with speeds between v1 and v2 is given by
2.8K
Divergence and Stokes' Theorems
3.5K
The divergence and Stokes' theorems are a variation of Green's theorem in a higher dimension. They are also a generalization of the fundamental theorem of calculus. The divergence theorem and Stokes' theorem are in a way similar to each other; The divergence theorem relates to the dot product of a vector, while Stokes' theorem relates to the curl of a vector. Many applications in physics and engineering make use of the divergence and Stokes' theorems, enabling us to write...
3.5K
Energy Conservation and Bernoulli's Equation
10.5K
Applying the conservation of energy principle or the work-energy theorem to an incompressible, inviscid fluid in laminar, steady, irrotational flow leads to Bernoulli's equation. It states that the sum of the fluid pressure, potential, and kinetic energy per unit volume is constant along a streamline.
All the terms in the equation have the dimension of energy per unit volume. The kinetic energy per unit volume is called the kinetic energy density, and the potential energy per unit volume is...
All the terms in the equation have the dimension of energy per unit volume. The kinetic energy per unit volume is called the kinetic energy density, and the potential energy per unit volume is...
10.5K
Conservation of Energy in Control Volume
1.1K
Consider a turbine operating under steady-flow conditions. The control volume is drawn around the turbine, with fluid entering at one point and exiting at another. The turbine extracts energy from the fluid, which performs mechanical work (shaft work).
For steady flow systems, the time derivative of the stored energy becomes zero since there is no energy accumulation within the control volume. This simplifies the energy equation to:
For steady flow systems, the time derivative of the stored energy becomes zero since there is no energy accumulation within the control volume. This simplifies the energy equation to:
1.1K
Difference from Background: Limit of Detection
8.0K
The limit of detection (LOD) is the smallest amount of analyte that can be distinguished from the background noise. The LOD value corresponds to the concentration at which the analyte signal is three times larger than the standard deviation of the blank signal. Below this value, the analyte signal cannot be differentiated from the background noise. It is calculated by dividing the calibration slope by 3 times the standard deviation of the blank signals.
The LOD indicates the presence or absence...
The LOD indicates the presence or absence...
8.0K
Conservation of Energy: Application
8.0K
When solving problems using the energy conservation law, the object (system) to be studied should first be identified. Often, in applications of energy conservation, we study more than one body at the same time. Second, identify all forces acting on the object and determine whether each force doing work is conservative. If a non-conservative force (e.g., friction) is doing work, then mechanical energy is not conserved. The system must then be analyzed with non-conservative work. Third, for...
8.0K
