AI Reconstruction Classification Using Compression Error Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional probabilistic classification methods face accuracy issues due to the skewed normalization problem, leading to false positives and false negatives in distinguishing authentic products from counterfeit ones, especially when counterfeit products have a high degree of similarity to one authentic class while being dissimilar to others.
Innovation Solution
The use of AI models that transform and reconstruct input data through multiple stages of compression and decompression, allowing accurate classification based on reconstruction error rather than probabilities, trained solely on authentic samples without needing samples from other classes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional probabilistic classification methods are used, then classification can be performed, but accuracy deteriorates due to skewed normalization problem causing false positives and false negatives
Solution Approach 1:
Instead of using probabilistic classification that assigns samples to classes based on normalized probabilities (which causes skewed normalization problems), the patent inverts the approach by using autoencoders to reconstruct input data and measuring reconstruction error. Samples from the same class have low reconstruction error while samples from different classes have high reconstruction error, eliminating the need for normalization and avoiding false positives.
Solution Approach 2:
The patent replaces the probabilistic classification mechanism with an autoencoder-based reconstruction mechanism. Rather than using statistical probability models that are sensitive to normalization skew, the system uses neural network-based data reconstruction where the degree of reconstruction accuracy directly indicates class membership, substituting a more robust mechanical classification approach.
2Adaptability or versatility
If AI models are trained using samples from multiple classes, then classification coverage is improved, but training complexity and data requirements increase
Solution Approach 1:
The patent segments the classification task by training separate autoencoder models for each class rather than using a single multi-class classification model. Each autoencoder is trained exclusively on samples from its corresponding class, allowing the system to handle multiple classes with simplified, specialized models that are easier to train and less computationally complex.
Solution Approach 2:
Instead of training a single comprehensive model to handle all classes, the patent creates multiple simplified models (copies) each specialized for a specific class. Each autoencoder is a copy trained on its target class data, and the classification decision is made by evaluating which model produces the lowest reconstruction error, achieving multi-class coverage through replicated specialized models.
Data Source
AI summary
The present invention provides AI strategies that can be used to classify samples. The strategies use AI models to transform and reconstruct an input dataset for a sample into a reconstructed dataset. An aspect of the transformation includes at least one compression of data and/or at least one decompression (or expansion) of data. Preferably the transformation involves compressing the data in a plurality of data compression stages and decompressing or expanding the data in a plurality of data decompressing or expansion stages. The advantage of compressing and decompressing the data is that the transformation becomes so complex and uniquely tailored to the trained, authentic samples such that only authentic samples of the associated class or classes are able to be reconstructed with sufficient accuracy to meet a reconstruction error threshold with high classification accuracy. The reconstruction error of other samples outside the associated class or classes generally would not reconstruct accurately enough to meet the reconstruction error threshold.