Supervised Learning Model Weighting via Clustering Discrepancy Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Supervised machine learning models often produce inaccurate outputs due to unreliability in data labeling, leading to operational inefficiencies, as all models are relied upon equally despite varying accuracy based on their training data and characteristics.
Innovation Solution
A computing platform applies unsupervised learning algorithms to identify discrepancies between clustering information and labeling information, flags data labeling errors, and trains supervised learning models using only accurately labeled data, while grading data sources based on error rates and confidence levels to weight their contributions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If all models are relied upon equally in making predictions, then the system is simple to operate, but inaccurate outputs are produced
Solution Approach 1:
The patent applies local quality by assigning different weights to different models based on their individual accuracy characteristics. Instead of treating all models equally, the system identifies which models are more accurate for specific tasks or data types and assigns them higher weights, thereby improving overall prediction accuracy while maintaining ease of operation through automated weight assignment.
2Productivity
If data labeling errors are present in the training set, then the training process is faster and simpler, but the model accuracy deteriorates
Solution Approach 1:
The patent applies preliminary action by performing data cleaning and error detection before the main training process. The system identifies and removes or corrects labeling errors in the training data prior to model training, ensuring that the models are trained on accurate data while maintaining efficient training processes through automated error detection methods.
3Reliability
If unsupervised learning is applied to detect labeling errors, then the reliability of training data improves, but the device complexity increases
Solution Approach 1:
The patent applies the intermediary principle by introducing unsupervised learning algorithms as a mediator between the raw labeled data and the supervised learning training process. This intermediary layer detects and filters labeling errors without requiring complex manual intervention, improving data reliability while keeping the overall system complexity manageable through automated error detection mechanisms.
Data Source
AI summary
Aspects of the disclosure relate to evaluating supervised learning models. A computing platform may receive initial training data, train supervised learning models using the initial training data, and form a composite model based on the supervised learning models. The computing platform may receive additional training data and corresponding prediction parameters, indicating actual outcomes. The computing platform may input the additional training data into the composite model to generate model-predicted outcome data, and may compare the model-predicted outcome data to the actual outcomes. Based on results of the comparison of the model-predicted outcome data to the actual outcomes, the computing platform may score each of the supervised learning models to reflect corresponding reliability levels. The computing platform may store a matrix relating the scores to their corresponding supervised learning models, which may cause the computing platform to weight results obtained from each supervised learning model when applying the composite model.