ML Feedback Filtering for Reliable Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models face challenges in training due to the quality of training data and user feedback, necessitating improved methods for filtering and improving user feedback and output.
Innovation Solution
A method involving a filter system, including a rule-based and machine learning-based approach, to detect and correct errors in user feedback and model output, utilizing autoencoders and masked language models to generate corrected results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning models are trained with user feedback and training data, then the model's learning capability is improved, but the training quality becomes challenging and unreliable due to poor quality feedback
Solution Approach 1:
The filter is applied before training data is used to pre-process and validate user feedback, detecting errors in class type information, relation type information, and entity consistency. This preliminary filtering ensures that only high-quality, error-free feedback reaches the training process, improving training reliability without adding complex training procedures
Solution Approach 2:
A filter system acts as an intermediary component between user feedback collection and machine learning model training. This filter uses trained models (autoencoder, anomaly detection) to automatically assess and clean feedback quality, serving as a mediator that protects the training process from poor quality input without requiring direct intervention in the training algorithm itself
2Measurement precision
If a filter system is implemented to detect errors in user feedback, then the training data quality is improved, but the system complexity increases
Solution Approach 1:
The filter system is designed to perform multiple functions simultaneously: it detects errors in class type information, validates relation type information, checks entity consistency, and prepares cleaned feedback for training. By consolidating these multiple quality assessment tasks into a single multi-functional filter component, the system achieves comprehensive feedback validation without proportionally increasing system complexity
Solution Approach 2:
The filter incorporates trained models (autoencoder model, anomaly detection model) that automatically assess and filter feedback quality without requiring manual review. The system serves itself by using machine learning models to evaluate feedback, reducing the need for human intervention in quality assessment while maintaining high measurement precision
Data Source
Figure 1~2
AI summary
Computer implemented method for filtering user feedback and/or output of a machine learning model, comprising: providing a filter for filtering user feedback and/or output of a machine learning model; receiving user feedback and/or output of the machine learning model; filtering the user feedback and/or the output with the filter and determining a filtering result, wherein the filtering result comprises at least a detected error; providing the filtering result for further processing.