Shape Classification Model Training for Subtle Subject Differences
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques struggle to accurately classify the shape of subjects when it is difficult for a person to determine the appropriate category to which the subject belongs, particularly in cases where subtle differences exist between subjects.
Innovation Solution
A trained model generation device and method that utilizes machine learning to generate a model capable of classifying subject shapes by training the model to maximize the probability of the same shape category for the same individual or species and increase the variance of the probability distribution for different individuals or species, using a loss function that minimizes entropy and cross-entropy to ensure accurate classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a trained model is generated using conventional machine learning methods, then the model can classify shapes when categories are clearly distinguishable, but the model fails to accurately classify shapes when subtle differences exist and human determination is difficult
Solution Approach 1:
The patent changes the training parameters by introducing a dual-objective loss function that simultaneously optimizes for classification accuracy and probability distribution variance. This allows the model to adapt to subtle shape differences by adjusting the optimization criteria rather than just the model architecture or input data
Solution Approach 2:
The patent makes the training process dynamic by using a loss function that adapts to different training scenarios. The loss function dynamically adjusts based on whether samples belong to the same or different categories, allowing the model to learn both fine-grained distinctions and general category characteristics
2Reliability
If the learning model is trained to maximize probability for correct categories, then classification confidence increases, but the model becomes overconfident and fails to distinguish subtle differences between categories
Solution Approach 1:
The patent implements feedback by using the probability distribution variance as a metric to guide training. The model receives feedback about the spread of its probability predictions and adjusts its parameters to maintain appropriate confidence levels while improving distinction accuracy
Solution Approach 2:
The patent performs preliminary action by pre-defining the dual-objective loss function structure before training begins. This loss function structure is designed in advance to prevent overconfidence by incorporating variance maximization as a constraint on the probability distributions
3Productivity
If conventional loss functions are used for training, then training convergence is achieved, but the model cannot handle cases where human determination of shape categories is difficult
Solution Approach 1:
The patent creates a composite loss function by combining classification loss and variance maximization loss. This composite approach integrates multiple objectives into a single training framework, allowing the model to converge while handling subtle shape differences that conventional single-objective loss functions cannot address
Data Source
AI summary
An information processing device acquires plural learning images in which a subject that is a part of each of individuals or species appears, the plural learning images being captured for each of the individual or species. The information processing device trains a learning model such that a probability of the same shape category is the highest in a case in which learning images in which the subjects of the same individual or species appear are input to the learning model. The information processing device generates the trained model by training the learning model so as to increase a variance of a probability distribution output from the learning model in a case in which each of learning images in which the subjects of plural different individuals or species appear is input to the learning model.