Polyhedral Classifier Training via Spherical Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for training polyhedral classifiers are inefficient in achieving high accuracy with low computational effort, particularly in devices with limited resources such as embedded systems, and struggle to effectively separate positive and negative data points in high-dimensional spaces.
Innovation Solution
The method involves transforming the original data space into a higher-dimensional space where data points are mapped to a sphere, allowing for the determination of pairs of hyperplanes that minimize the distance between them while separating positive and negative examples, using a Sequential Margin Maximizer approach that iteratively solves a tractable optimization problem to construct a polyhedron that separates the data points efficiently.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional cutting-plane methods are used for training polyhedral classifiers, then the classifier can be trained on embedded systems with lower computational power, but the training time is long and accuracy is limited
Solution Approach 1:
The patent transforms the original data space into a higher-dimensional space where data points are mapped to a sphere. This dimensional transformation enables the use of geometric properties in higher dimensions to achieve better separation between classes, improving classification accuracy while maintaining training efficiency through the structured approach of mapping to spherical surfaces.
Solution Approach 2:
The patent changes the parameter representation by transforming data points into a different dimensional space with specific geometric constraints (spherical mapping). This parameter transformation allows the optimization problem to be reformulated in terms of hyperplane orientations and distances, enabling more efficient training algorithms that converge faster while achieving higher accuracy.
2Measurement precision
If the data space is transformed to higher dimensions with spherical mapping, then the separation of positive and negative data points is improved, but the computational complexity increases
Solution Approach 1:
The patent segments the classification problem into two main components: (1) transforming data points to a spherical surface in higher-dimensional space, and (2) finding hyperplanes that separate positive and negative examples. This segmentation allows each sub-problem to be solved independently with optimized algorithms, reducing overall computational complexity while maintaining high separation accuracy.
Solution Approach 2:
By mapping data points to a sphere in higher-dimensional space, the patent exploits the geometric properties of spherical surfaces to improve separability. The higher-dimensional spherical representation allows linear hyperplanes to achieve better classification boundaries, and the structured nature of spherical mapping enables efficient computation through specialized algorithms.
3Measurement precision
If iterative optimization is used to determine hyperplane orientation, then the classification accuracy improves, but the number of iterations required increases computational effort
Solution Approach 1:
The patent employs an iterative optimization process where each iteration uses feedback from the previous iteration to improve hyperplane orientation. The algorithm adjusts hyperplane parameters based on the separation performance observed in prior iterations, gradually improving classification accuracy. This feedback mechanism ensures convergence to high-accuracy solutions while the structured approach controls the number of iterations required.
Solution Approach 2:
The patent performs preliminary transformation of data points to a spherical surface before initiating the iterative hyperplane optimization. This preliminary action prepares the data in a geometrically favorable configuration that accelerates subsequent optimization iterations, reducing the total computational effort required to achieve high classification accuracy.
Data Source
AI summary
A method for training a polyhedral classifier is described including obtaining training data in a data space, the training data including first data points associated with a first label and second data points associated with a second label, determining a pair of hyperplanes by determining an orientation of the pair of hyperplanes based on a minimization of a distance between the pair of hyperplanes such that the first data points lie between the hyperplanes in relation to a distance between the pair of hyperplanes such that both the first data points and the second data points lie between the hyperplanes and determining the position of the pair of hyperplanes such that the first data points lie between the pair of hyperplanes and the second data points are at least partially separated from the first data points by the pair of hyperplanes.