Kohonen Network Ensemble for High-Dimensional Data Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning systems face challenges in handling large-scale, high-dimensional data, particularly in efficiently processing streaming data and performing dimensionality reduction, while also being deployable on massively parallel hardware.
Innovation Solution
The method employs Kohonen networks for feature selection and classifier construction, utilizing parallel processing to train multiple Kohonen nets, retain only active neurons, and construct an ensemble of classifiers, enabling efficient handling of high-dimensional data and imbalanced datasets.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional machine learning systems process large-scale high-dimensional data, then classification accuracy can be maintained, but processing speed and scalability deteriorate
Solution Approach 1:
The patent divides the high-dimensional data processing into multiple stages: (1) dimensionality reduction stage using Kohonen networks to project high-dimensional data to lower dimensions, (2) feature selection stage to identify relevant features, and (3) classification stage using ensemble methods. This segmentation allows the system to maintain classification accuracy while significantly improving processing speed by reducing the computational complexity at each stage.
Solution Approach 2:
The patent extracts only the most relevant features from high-dimensional data using Kohonen networks and feature selection algorithms, discarding redundant dimensions. This extraction process reduces the data volume that needs to be processed in subsequent classification stages, thereby improving processing speed without sacrificing classification accuracy.
2Productivity
If dimensionality reduction is performed on high-dimensional data, then processing efficiency improves, but information loss increases
Solution Approach 1:
The patent employs feedback mechanisms in the Kohonen network training process and iterative feature selection to continuously evaluate the quality of dimensionality reduction. The system monitors classification performance and adjusts the reduction process to retain features that are most informative for classification, thereby minimizing information loss while maintaining processing efficiency.
Solution Approach 2:
The patent dynamically adjusts parameters in the Kohonen network (such as learning rate, neighborhood function, and grid size) during the dimensionality reduction process to optimize the balance between reducing dimensionality and preserving important information. This parameter tuning ensures that processing efficiency is improved without excessive information loss.
3Measurement precision
If multiple Kohonen networks are trained in parallel, then feature selection quality improves, but computational complexity increases
Solution Approach 1:
The patent combines multiple Kohonen networks into an ensemble system where each network processes different aspects of the data. The results from multiple networks are merged through voting or aggregation mechanisms to produce the final feature selection and classification. This merging approach improves feature selection quality by leveraging diverse perspectives while distributing the computational load.
Solution Approach 2:
The patent trains multiple Kohonen networks in parallel, which represents an excessive action compared to using a single network. This approach improves feature selection quality by capturing different patterns in the data, but the patent manages the increased computational complexity through parallel processing and efficient resource allocation.
4Speed
If streaming data is processed in real-time, then responsiveness improves, but data volume handling capability deteriorates
Solution Approach 1:
The patent performs preliminary dimensionality reduction and feature extraction on streaming data using Kohonen networks before the data enters the classification stage. This preliminary action reduces the volume of data that needs to be processed in real-time, enabling the system to maintain responsiveness while handling large volumes of streaming data efficiently.
Data Source
AI summary
Analyzing patterns in a volume of data and taking an action based on the analysis involves receiving data and training the data to create training examples, and then selecting features that are predictive of different classes of patterns in the data stream, using the training examples. The process further involves training in parallel a set of artificial neural networks (“ANNs”), using the data, based on the selected features, and extracting only active nodes that are representative of a class of patterns in the data stream from the set of ANNs. The process continues with adding class labels to each extracted active node, classifying patterns in the data based on the class-labeled active nodes, and taking an action based on the classifying patterns in the data.