Unsupervised Learning Model Cluster Number Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning technologies lack a training apparatus capable of training models suitable for classification into a target cluster number, leading to unsatisfactory results in unsupervised learning where the number of clusters exceeds user expectations.
Innovation Solution
A training apparatus that acquires subject data and a target cluster number, iteratively trains a learning model using unsupervised learning, estimates a feature cluster number based on feature vectors, and updates learning conditions to align the feature cluster number with the target cluster number.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If unsupervised learning is performed without classification labels, then the learning model can process untagged data, but the number of clusters exceeds user expectations and results are not preferable
Solution Approach 1:
The patent implements feedback by calculating a loss function that measures the difference between the actual number of clusters formed and the target cluster number, then using this loss to update learning conditions in subsequent training iterations. This closed-loop feedback mechanism enables the system to progressively adjust and converge toward the desired cluster number while maintaining the ability to process untagged data.
Solution Approach 2:
The patent dynamically adjusts learning conditions (such as temperature parameters, clustering coefficients, or network architecture parameters) based on the measured deviation between actual and target cluster numbers. This dynamic adaptation allows the system to transition from initial exploration to precise cluster formation, resolving the contradiction between versatility and precision.
2Measurement precision
If the learning model is trained to classify into a target cluster number, then classification accuracy improves, but the training process becomes complex and requires iterative adjustment of learning conditions
Solution Approach 1:
The patent enables self-service by automatically calculating the actual cluster number from feature vectors, computing the loss based on the difference from the target cluster number, and autonomously updating learning conditions without requiring manual intervention. This self-adjusting mechanism simplifies the overall process despite the iterative nature, as the system serves itself through automated feedback loops.
Solution Approach 2:
The patent systematically changes key parameters (such as temperature in softmax functions, clustering coefficients, or regularization terms) based on the measured cluster number deviation. By controlling and adjusting these parameters iteratively, the system achieves accurate classification while managing training complexity through structured parameter optimization rather than uncontrolled search.
3Quantity of substance
If feature vectors are extracted from subject data, then the data can be clustered, but the feature cluster number may not match the target cluster number
Solution Approach 1:
The patent uses feedback by measuring the actual cluster number formed from extracted feature vectors and comparing it with the target cluster number. This measurement feeds back into the learning condition updates, allowing the system to adjust feature extraction parameters or clustering coefficients to achieve better cluster number matching while preserving the capability to extract meaningful features.
Solution Approach 2:
The patent adjusts parameters related to feature extraction and clustering (such as dimensionality reduction parameters, feature transformation coefficients, or clustering hyperparameters) based on the cluster number mismatch. This parameter optimization enables the system to maintain effective feature extraction while improving cluster number accuracy through iterative refinement.
Data Source
AI summary
According to one embodiment, a training apparatus includes processing circuitry. The processing circuitry acquires a plurality of items of subject data and a target cluster number, iteratively trains a learning model on the plurality of items of subject data by unsupervised learning based on learning conditions, estimates a feature cluster number based on a plurality of feature vectors corresponding to the plurality of items of subject data, and updates the learning conditions based on the feature cluster number and the target cluster number.