Object Detection Model Training Limit via Anomaly Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection techniques using machine learning face challenges in determining the training limit for eliminating anomalies such as false positive and false negative detections, which can vary based on the technical expertise of the person in charge, posing risks in applications requiring high reliability like autonomous driving.
Innovation Solution
An information processing method and system that utilize a series of machine learning-based trainings and determinations to identify the training limit by analyzing prediction results and their anomalies, using evaluation data and reference information to determine if further training is necessary.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning-based model training is conducted using training data including extra data to be detected with anomalies, then detection anomalies such as false positives and false negatives can be reduced, but it becomes difficult to determine whether the training limit has been reached
Solution Approach 1:
The system implements feedback by comparing prediction results across multiple training iterations. The determination unit analyzes whether errors and omissions in prediction results have stabilized by comparing current training results with previous training results, providing feedback on whether further training will yield improvements or if the training limit has been reached.
Solution Approach 2:
The patent introduces an intermediary evaluation mechanism that uses reference information and comparison processes to assess training status. The determination unit acts as an intermediary between the training process and the trainer, objectively evaluating whether the training limit has been reached based on stabilization of prediction errors across iterations.
2Measurement precision
If multiple training iterations are performed to eliminate detection anomalies, then detection precision improves, but training time and computational resources increase
Solution Approach 1:
The system performs training iterations beyond what might be considered sufficient, continuing until the stabilization criterion is met. This partial/excessive action ensures that the training limit is definitively identified, preventing both premature termination and unnecessary extended training, thus optimizing the balance between precision improvement and time consumption.
Solution Approach 2:
The feedback mechanism compares prediction results across training iterations to detect stabilization. When errors and omissions stop improving between iterations, the system automatically determines that the training limit has been reached, providing an objective stopping criterion that prevents wasteful continuation of training while ensuring optimal precision is achieved.
3Reliability
If training continues beyond the training limit, then no further improvement in detection accuracy is achieved, but computational resources are wasted
Solution Approach 1:
The determination unit continuously monitors prediction results and compares them across training iterations. When the feedback indicates that errors and omissions have stabilized and no further improvement is occurring, the system automatically terminates training, preventing waste of computational energy while ensuring detection reliability is optimized.
Solution Approach 2:
The system performs self-evaluation of its own training status through the determination unit, which autonomously assesses whether the training limit has been reached based on stabilization of prediction results. This self-service mechanism eliminates the need for external judgment and automatically prevents unnecessary continuation of training, conserving computational resources.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An information processing method includes acquiring a first prediction result by inputting evaluation data to a first model; determining an anomaly in the first prediction result based on the first prediction result and reference information; acquiring a second model based on the determination result; acquiring a second prediction result by inputting the evaluation data to the second model; determining an anomaly in the second prediction result based on the second prediction result and the reference information; acquiring a third model based on the determination result; acquiring a third prediction result by inputting the evaluation data to the third model; determining an anomaly in the third prediction result based on the third prediction result and the reference information; and if the anomaly in the third prediction result is recognized as being identical to the anomaly in the first prediction result, outputting information about a training limit of the first model.