Support Vector Model Size Reduction via Kernel Space Precomputation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current prediction models, particularly Support Vector Machines (SVMs), face challenges in efficiently encoding models for use in both linear and nonlinear systems, leading to large model sizes and increased prediction times, which are prohibitive in resource-constrained environments, and often sacrifice generalization accuracy to reduce model size.
Innovation Solution
The technique involves precomputing weight vector coefficients in the kernel space, eliminating the need to store support vectors, resulting in a compact model that maintains generalization performance and accelerates prediction, applicable to various kernel functions including linear, polynomial, and radial basis function kernels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If Support Vector Machines (SVMs) are used for prediction in both linear and nonlinear systems, then generalization performance is improved, but model size increases and prediction time increases
Solution Approach 1:
The patent extracts only the essential components needed for prediction by eliminating the need to store support vectors. The method precomputes and stores only the weight vector coefficients in the kernel space, which are sufficient for making predictions. This extraction of necessary elements reduces model size while preserving generalization performance.
Solution Approach 2:
The patent applies preliminary action by precomputing the weight vector coefficients in the kernel space during the training phase. This precomputation allows the model to be compact during the prediction phase, as all necessary computations have been prepared in advance. The preprocessing step transforms the complex SVM model into a compact form suitable for resource-constrained environments.
2Reliability
If Support Vector Machines (SVMs) are used for prediction in both linear and nonlinear systems, then generalization performance is improved, but prediction time increases
Solution Approach 1:
The patent applies preliminary action by precomputing the weight vector coefficients in the kernel space during the training phase. This precomputation allows the model to be compact during the prediction phase, as all necessary computations have been prepared in advance. The preprocessing step transforms the complex SVM model into a compact form suitable for resource-constrained environments.
3Adaptability or versatility
If model size is reduced to fit resource-constrained environments, then device compatibility is improved, but generalization accuracy is sacrificed
Solution Approach 1:
The patent extracts only the essential components needed for prediction by eliminating the need to store support vectors. The method precomputes and stores only the weight vector coefficients in the kernel space, which are sufficient for making predictions. This extraction of necessary elements reduces model size while preserving generalization performance.
Solution Approach 2:
The patent changes the representation parameters of the SVM model by storing weight vector coefficients in the kernel space instead of storing support vectors. This parameter transformation enables the model to be compact while maintaining its predictive capabilities. The change in representation allows the model to fit resource-constrained environments without sacrificing accuracy.
4Ease of manufacture
If a compact model is created to reduce storage requirements, then ease of deployment is improved, but prediction capability may be compromised
Solution Approach 1:
The patent extracts only the essential components needed for prediction by eliminating the need to store support vectors. The method precomputes and stores only the weight vector coefficients in the kernel space, which are sufficient for making predictions. This extraction of necessary elements reduces model size while preserving generalization performance.
Data Source
AI summary
Disclosed are a system and method for constructing and using a predictive model to generate a prediction signal, also referred to as a classification signal when the signal indicates one of a plurality of distinct classes. In various embodiments, the disclosed technique reduces a size of a predictive Support Vector Model by extracting certain values beforehand and storing only weighting values. The technique does not sacrifice generalization performance but does significantly reduce the model size and accelerate prediction performance. The described system applies to most kernel functions, whether linear or nonlinear.