Compressed Classification Model Using Optimal Image Reward
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional classification systems using neural networks face inefficiencies due to iterative processes in adjusting weights and biases, leading to inadequate training and reduced accuracy, particularly in identifying relevant features for image classification.
Innovation Solution
A method and system for performing classification using a compressed classification model, where relevant neurons are identified, and a classification error is determined to generate a reward value, resulting in a compressed model that eliminates inert neurons and reduces the number of neurons required for classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If iterative processes are used to adjust weights and biases in neural networks, then classification accuracy may be improved, but training time and computational complexity increase significantly
Solution Approach 1:
The patent applies preliminary action by pre-processing input samples to extract dominant attributes and generating optimal images that highlight key features before classification. This preparation work is done in advance to reduce the complexity of the subsequent classification process, allowing faster convergence without sacrificing accuracy.
Solution Approach 2:
The patent changes parameters by transforming input images into optimal images with modified parameters that emphasize dominant attributes. This transformation alters the parameter space in which classification occurs, making the problem easier to solve with fewer iterations.
2Measurement precision
If more neurons are included in the classification model, then classification accuracy improves, but device complexity and computational load increase
Solution Approach 1:
The patent applies the taking out principle by extracting and removing inert neurons from the classification model. By identifying and eliminating neurons that do not contribute to classification performance, the model complexity is reduced while maintaining accuracy through the use of optimal images that enhance the effectiveness of remaining neurons.
Solution Approach 2:
The patent applies partial action by using only the necessary subset of neurons for classification after removing inert ones. Rather than using all available neurons, the system employs just enough computational resources to achieve the required accuracy level.
3Loss of information
If traditional feature classification methods are used, then comprehensive feature analysis is achieved, but identification of relevant features for real-time classification becomes inefficient
Solution Approach 1:
The patent applies preliminary action by pre-identifying dominant attributes in input samples and generating optimal images that highlight these features before classification. This advance preparation ensures that relevant features are readily identifiable during real-time classification without sacrificing comprehensive analysis.
Solution Approach 2:
The patent applies segmentation by separating the feature analysis process into distinct stages: identifying dominant attributes, generating optimal images, and performing classification. This segmentation allows comprehensive feature analysis to be completed efficiently by breaking it down into manageable steps that can be optimized independently.
Data Source
AI summary
The present disclosure relates to method and system for performing classification of real-time input sample using compressed classification model. Classification system receives classification model configured to classify training input sample. Relevant neurons are identified from neurons of the classification model. Classification error is identified for each class. Reward value is determined for the relevant neurons based on relevance score of each neuron and the classification error. Optimal image is generated for each class based on the reward value of the relevant neurons. The optimal image is provided to the classification model for generating classification error vector for each class. The classification error vector is used for identifying pure neurons from the relevant neurons. A compressed classification model comprising the pure neurons is generated. The generated compressed classification model is used for performing the classification of real-time input sample.