AI Apparatus Data Suitability Filtering for Model Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial intelligence models face issues when updated based on incorrect user feedback, leading to incorrect learning and reduced model performance, as they rely on assumptions of normal user feedback without validating the suitability of usage logs.
Innovation Solution
An AI apparatus that determines the suitability of input data and feedback for updating the model using multiple methods, including outlier detection, sensitivity analysis, ensemble inference comparison, and kernel score calculation, to generate suitable training data and prevent incorrect updates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the artificial intelligence model is updated based on user usage logs, then the model can learn from real user behavior and improve realism, but the model may be incorrectly learned if users provide wrong feedback
Solution Approach 1:
The patent applies preliminary action by performing suitability determination on usage logs before they are used to update the AI model. The processor determines whether collected usage logs are suitable for model updates by comparing them against predetermined criteria, filtering out inappropriate data before the learning process begins. This prevents incorrect feedback from corrupting the model while still allowing comprehensive user behavior learning.
2Reliability
If multiple suitability determination methods are used to filter data, then the reliability of training data improves, but the processing time and computational complexity increase
Solution Approach 1:
The patent segments the data validation process into multiple independent suitability determination methods that can be applied systematically. Each method (comparing against predetermined suitability criteria, outlier detection, sensitivity analysis) operates as a separate filtering stage, allowing the system to process data through multiple checks without creating a monolithic complex procedure. This modular approach improves data quality while managing processing complexity.
3Reliability
If usage logs are filtered to remove outliers, then the model learns from normal user behavior, but useful information about edge cases may be lost
Solution Approach 1:
The patent applies parameter changes by adjusting the predetermined suitability criteria dynamically. The system can modify thresholds and parameters in the suitability determination based on different contexts, allowing it to distinguish between genuine outliers representing new user scenarios and actual noise. This enables the model to maintain consistency from normal behavior while still adapting to legitimate edge cases that meet updated criteria.
Data Source
AI summary
An embodiment of the present disclosure provides an artificial intelligence apparatus for generating training data including a memory configured to store an artificial intelligence model, an input interface including a microphone or a camera, and a processor configured to receive, via the input interface, input data, generate an inference result corresponding to the input data by using the artificial intelligence model, receive feedback corresponding to the inference result, determine suitability of the input data and the feedback for updating the artificial intelligence model, and generate training data based on the input data and the feedback if the input data and the feedback are determined as data suitable for updating of the artificial intelligence model.