Machine Learning Model Training for Human-Animal Misclassification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models erroneously recognize animals as humans due to shared feature amounts, necessitating a technology for creating highly accurate models that differentiate between humans and animals accurately.
Innovation Solution
A machine learning device and program that exclude image data of animals with feature amounts similar to humans during model learning, employing relearning to adjust feature amounts and data until erroneous recognition is minimized.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models are trained with all available image data including animals, then the model learns more features and improves general detection capability, but the model erroneously recognizes animals as humans due to shared feature amounts
Solution Approach 1:
The patent extracts and removes image data of animals from the training dataset that have feature amounts similar to humans. By separating these problematic samples from the general training data, the model is prevented from learning incorrect associations between animal features and human classification, thereby eliminating erroneous recognition while preserving useful training data for accurate detection.
Solution Approach 2:
Instead of trying to make the model distinguish between humans and animals during training, the patent inverts the approach by excluding animal images with human-like features from the training data entirely. This prevents the model from learning ambiguous patterns that would lead to misclassification, effectively solving the problem by what is not done rather than by active differentiation.
2Reliability
If the model is retrained multiple times with adjusted feature amounts, then the classification accuracy improves and erroneous recognition is reduced, but the training time and computational resources increase
Solution Approach 1:
The patent performs preliminary filtering of the training dataset before model training by identifying and removing animal images with feature amounts similar to humans. This pre-processing step prevents the model from learning incorrect patterns in the first place, eliminating the need for multiple iterative retrains and reducing overall training time while achieving high classification accuracy from the outset.
Solution Approach 2:
The patent implements a feedback mechanism where the model's classification results are evaluated, and if erroneous recognition of animals as humans occurs, the training process is adjusted by excluding the problematic samples and retraining. This iterative feedback loop continues until classification accuracy is satisfied, ensuring high reliability while systematically reducing training iterations through learned insights.
Data Source
AI summary
A machine learning device includes a processor which performs, when performing learning of a model with image data, the learning of the model while excluding the image data of an animal having a feature amount common with a feature amount of a human.