Iterative Fisher Discriminant Classifier for Multi-Modal Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing classification methods, such as linear Fisher discriminant analysis, face challenges in achieving high accuracy for data with complex multi-modal attribute distributions while maintaining minimal impact on classification time, especially when data distributions deviate from Gaussian assumptions.
Innovation Solution
An iterative classifier based on linear Fisher discriminant analysis is proposed, where property distances are projected onto discriminant and principal component vectors to form histograms, allowing for nesting analysis to identify and eliminate correctly classified data, with subsequent reclassification using either analysis until full classification is achieved.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If linear Fisher discriminant analysis is used for classification, then classification speed is fast and training time is low, but classification accuracy degrades when data distributions are multi-modal or deviate from Gaussian assumptions
Solution Approach 1:
The patent segments the classification process into multiple iterations. In each iteration, the algorithm identifies and removes correctly classified data points, then re-trains the discriminant analysis on the remaining misclassified data. This segmentation allows the simple fast linear Fisher discriminant to be applied repeatedly to progressively harder cases, maintaining speed while improving overall accuracy for complex multi-modal distributions.
Solution Approach 2:
The patent introduces dynamic adaptivity by iteratively updating the training set. The classification system dynamically adjusts which data points are included in training by removing correctly classified points from the training set in each iteration. This dynamic approach allows the model to adapt to complex data distributions without requiring a completely complex initial model, balancing speed and accuracy.
2Measurement precision
If more complex classification methods are used to handle multi-modal distributions, then classification accuracy improves, but training time and computational complexity increase significantly
Solution Approach 1:
The patent applies partial action by using the simple linear Fisher discriminant analysis for the portion of data that is easily classifiable (Gaussian-distributed), and only applying the full iterative process to the remaining difficult cases. This partial application of a more complex approach (iterative re-training) only where necessary maintains accuracy for multi-modal data while avoiding the full computational cost for all data points.
Solution Approach 2:
The iterative process is self-adaptive: the algorithm automatically identifies which data points are correctly classified and removes them from further training consideration. This self-service mechanism eliminates the need for manual intervention or complex pre-processing to identify difficult cases, allowing the system to efficiently focus computational resources only where needed.
Data Source
AI summary
An exemplary method includes receiving an image data set that comprises a multidimensional property space and data classifiable into data classes, determining a projection vector for data of the data set wherein the projection vector maximizes a ratio of between-class scatter to within-class scatter, selecting a reference for the vector, projecting at least some of the data onto the vector, measuring distances from the reference to at least some of the data, classifying at least some of the data into data classes based on a nesting analysis of the distances, eliminating the classified data from the image data set to produce a modified image data set and deciding whether to determine another projection vector for the modified image data set. Various other exemplary methods, devices, systems, etc., are also disclosed.