Unsupervised Feature Vector Comparison for Interpretable Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In unsupervised learning, subject data with different characteristics often get classified into the same cluster, making it difficult to interpret the clustering results.
Innovation Solution
A data analysis apparatus that uses a deep neural network (DNN) for unsupervised learning, employing data augmentation conditions to generate and compare first and second feature vectors, and analyze clustering results using K-Means clustering and dispersion degrees to improve cluster separation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If unsupervised learning is used to cluster subject data, then the data can be automatically classified into clusters, but subject data with different characteristics tend to be classified into the same cluster, making it difficult to interpret the clustering result
Solution Approach 1:
The patent introduces a feedback mechanism where the clustering result is used to select representative data, which then serves as training data for a classification model. The classification model's output is fed back to refine the clustering, creating an iterative process that improves interpretability while maintaining automation.
Solution Approach 2:
The patent introduces a classification model as an intermediary between the unsupervised clustering process and the final interpretation. This intermediary component translates the abstract clustering results into more meaningful classifications by using the representative data selected from each cluster.
2Manufacturing precision
If data augmentation is applied to improve clustering, then cluster separation may be enhanced, but the complexity of the data processing increases
Solution Approach 1:
The patent applies data augmentation partially by using it only for the representative data selection process rather than for all input data. This selective application improves cluster separation for the most critical data points while avoiding the computational overhead of augmenting the entire dataset.
Solution Approach 2:
The patent performs data augmentation as a preliminary action during the representative data selection phase, before the main clustering and classification processes. This timing optimizes the benefits of data augmentation while minimizing its impact on overall processing complexity.
Data Source
AI summary
According to one embodiment, a data analysis apparatus includes processing circuitry. The processing circuitry acquires a plurality of items of subject data, trains a first training model by performing unsupervised learning on the items of subject data using a first data augmentation condition that is a condition related to a data augmentation conversion method, and generates a plurality of first feature vectors corresponding to the items of subject data, generates a first clustering result by clustering the first feature vector, trains a second training model by performing unsupervised learning on the items of subject data using a second data augmentation condition, and generate a plurality of second feature vectors corresponding to the items of subject data, and generates a comparison result by comparing the first feature vectors and the second feature vectors for each of a plurality of clusters based on the first clustering result.