3D Point Descriptors Using Moment Matrices for Rotation Invariance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models struggle to generate rotationally invariant or covariant descriptors of three-dimensional configurations of points efficiently, leading to reduced performance in predicting properties of atomic or molecular systems due to insufficient incorporation of high-order geometric information.
Innovation Solution
A method is employed to generate rotationally invariant or covariant descriptors by transforming feature vectors into moment matrices, which are then combined to encode higher-order geometric information without the need for tensor products, utilizing Clebsch-Gordan operations, and optimized using hardware accelerators like GPUs or TPUs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Clebsch-Gordan operations are used to generate rotationally invariant descriptors, then the descriptors can capture high-order geometric information, but the computational complexity increases to O(l^6)
Solution Approach 1:
The patent segments the descriptor generation process into three distinct stages: (1) computing spherical harmonic coefficients from point coordinates, (2) forming moment matrices from these coefficients, and (3) extracting rotational invariants from the moment matrices. This segmentation allows each stage to be optimized independently and reduces the overall computational complexity from O(l^6) to O(l^3).
Solution Approach 2:
The patent introduces moment matrices as an intermediary structure between the spherical harmonic coefficients and the final rotational invariants. The moment matrices serve as a bridge that organizes the geometric information in a way that enables efficient extraction of invariants through trace operations, avoiding the need for direct complex Clebsch-Gordan coefficient calculations.
2Measurement precision
If tensor products are used to encode higher-order geometric information, then the descriptors can characterize configurations uniquely, but the device complexity increases
Solution Approach 1:
Instead of using tensor products to directly compute rotational invariants (the conventional approach), the patent inverts the approach by first computing moment matrices from spherical harmonic coefficients and then extracting invariants through trace operations. This inversion simplifies the computational structure while maintaining the ability to characterize configurations uniquely.
Solution Approach 2:
The patent changes the mathematical parameters used to represent geometric information. Instead of working directly with tensor products and Clebsch-Gordan coefficients, the patent uses moment matrices with specific symmetry properties that allow for simpler parameter transformations and reduce the overall structural complexity of the computation.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium for generating rotationally invariant or covariant descriptors of a three-dimensional configuration of points. In one aspect, a method comprises: using coordinates of the points to determine a plurality of feature vectors, each having a corresponding degree and comprising a respective one or more features, each feature being determined using a spherical harmonic function of the degree of the feature vector and a respective order by combining values of the spherical harmonic function evaluated at the respective coordinates; transforming each of the plurality of the feature vectors into a corresponding moment matrix, wherein each moment matrix corresponds to a respective irreducible representation of the 3D rotation group in a direct sum representation of a tensor product of irreducible representations of the 3D rotation group; and using the moment matrices to determine one or more invariant or covariant descriptors.