原子中心神经网络潜力的自动特征选择使用梯度增强决策算法
Renzhe Li1, Jiaqi Wang1, Akksay Singh1,2,3
1Department of Materials Science and Engineering, Southern University of Science and Technology, Shenzhen 518055, People's Republic of China.
Journal of chemical theory and computation
|November 19, 2024
概括
本研究介绍了一种使用梯度增强决策树 (GBDT) 选择最佳原子中心对称函数 (ACSFs) 准确的原子中心神经网络 (ANN) 潜力的自动化方法. 该方法通过提高准确性和效率来增强机器学习潜力的发展.
科学领域:
- 计算材料科学 计算材料科学
- 在化学和物理领域的机器学习.
- 用于材料建模的方法开发.
背景情况:
- 原子中心神经网络 (ANN) 潜力为原子系统模拟提供了高精度和效率.
- ANN潜力的性能严重取决于适当选择描述原子环境的以原子为中心的对称函数 (ACSFs).
- 低于最佳的ACSF选择可以显著降低ANN潜力的质量和预测能力.
研究的目的:
- 为原子中心神经网络 (ANN) 潜能选择最佳的原子中心对称函数 (ACSF) 开发一个自动化框架.
- 通过系统的特征选择,提高ANN潜力的精度和计算效率.
- 为机器学习潜在发展提供一种强大的方法来识别最相关的原子特征.
主要方法:
- 实现基于梯度增强决策树 (GBDT) 的框架,用于自动选择 ACSF.
- 评估一个全面的一组均分布的ACSF的相对重要性.
- 培训和验证ANN潜力,使用选择的最佳ACSF,与网格搜索和其他特征选择算法进行基准测试.
主要成果:
- 基于GBDT的框架成功地为 (Ge) 系统确定了一组最小的18个ACSF.
- 实现了高精度,平均平方根误差 (RMSE) 为10.2meV/原子的能量和84.8meV/Å的力预测.
- 与常用的特征选择算法相比,证明了卓越的性能,验证了所选 ACSF 的最佳性.
结论:
- 拟议的GBDT框架提供了一种有效和准确的方法,用于在ANN潜在开发中自动选择ACSF.
- 优化的ACSF选择显著提高了机器学习潜力的准确性和计算效率之间的平衡.
- 这种方法促进了对材料模拟的更可靠和高性能机器学习潜力的开发.
相关概念视频
Graded Potential
3.7K
Graded potentials are localized fluctuations in the cell membrane's electrical charge, commonly found in the dendrites of neurons. The magnitude of these potential changes depends on the strength of the initiating stimulus. In a membrane at its resting potential, a graded potential signifies a voltage shift either above -70 mV or below -70 mV.
Graded potentials fall into two categories: depolarizing and hyperpolarizing. Depolarizing graded potentials typically occur when sodium (Na+) or...
Graded potentials fall into two categories: depolarizing and hyperpolarizing. Depolarizing graded potentials typically occur when sodium (Na+) or...
3.7K
Predicting Reaction Outcomes
8.2K
Kinetics describes the rate and path by which a reaction occurs. In contrast, thermodynamics deals with state functions and describes the properties, behavior, and components of a system. It is not concerned with the path taken by the process and cannot address the rate at which a reaction occurs. Although it does provide information about what can happen during a reaction process, it does not describe the detailed steps of what appears on an atomic or a molecular level. On the other hand,...
8.2K
Predicting Molecular Geometry
34.1K
VSEPR Theory for Determination of Electron Pair Geometries
34.1K
Action Potential
7.8K
Neurons communicate by firing action potentials—the electrochemical signal that is propagated along the axon. The signal results in the release of neurotransmitters at axon terminals, thereby transmitting information to the nervous system. An action potential is a specific "all-or-none" change in membrane potential that results in a rapid spike in voltage.
Membrane potential in neurons
Neurons typically have a resting membrane potential of about -70 millivolts (mV). When they...
Membrane potential in neurons
Neurons typically have a resting membrane potential of about -70 millivolts (mV). When they...
7.8K
Nuclear Overhauser Enhancement (NOE)
641
Irradiation of a spin-active nucleus causes an increase or decrease in the signal intensity of neighboring nuclei that are not necessarily chemically bonded or involved in J-coupling. This phenomenon, called the Nuclear Overhauser Enhancement (NOE), results from through-space interactions between the nuclear spins. The NOE effect decreases with increasing internuclear distance and is generally not observed beyond 4 angstroms. In NOE, dipole-dipole interactions between neighboring...
641
Predicting Products: SN1 vs. SN2
13.2K
Nucleophilic substitution reactions of alkyl halides can proceed via an SN1 or an SN2 mechanism. While in SN2 reactions, the nucleophile attacks the substrate simultaneously as the leaving group departs, in SN1 reactions, the substrate first dissociates to give the carbocation intermediate. Various factors such as the structure of the substrate, the strength of the nucleophile, and the nature of the solvent promote one mechanism over the other.
With increased substitution on the alkyl halide,...
With increased substitution on the alkyl halide,...
13.2K


