Learning Device Loss Function Log-Likelihood Ratio
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning techniques for class classification lack an effective method to incorporate probability density estimation, which is crucial for achieving high accuracy in classification tasks.
Innovation Solution
A learning device and method that utilize a loss function to calculate the difference between log-likelihood ratios of class probabilities, ensuring the magnitude of this difference is minimized, thereby improving class classification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional loss functions are used for class classification, then the classification process is simple, but the classification accuracy is insufficient due to lack of probability density estimation
Solution Approach 1:
The patent transforms the loss function by applying a monotonic transformation function to the log-likelihood ratio, changing the parameter space from unbounded to a finite range. This allows the incorporation of probability density estimation while maintaining the optimization framework, thereby improving classification accuracy without excessive complexity increase
Solution Approach 2:
The patent introduces a monotonic transformation function as an intermediary between the likelihood ratio and the loss calculation. This intermediary function (such as logistic sigmoid or softmax) maps the unbounded log-likelihood ratio to a bounded range, enabling probability density estimation to be integrated into the classification process while maintaining computational tractability
2Reliability
If the likelihood ratio is used directly in the loss function, then the probability density estimation is incorporated, but the function value range becomes unbounded causing optimization difficulties
Solution Approach 1:
The patent applies a monotonic transformation function that maps the unbounded log-likelihood ratio to a bounded range (e.g., (0,1) or (-1,1)). This parameter transformation maintains the ordering information needed for reliable probability density estimation while eliminating the unbounded nature that causes optimization convergence issues
Solution Approach 2:
The patent transforms the loss function from operating in the unbounded real number dimension to operating in a bounded dimension through the monotonic transformation. This dimensional transformation allows the use of standard optimization algorithms while preserving the essential probabilistic information needed for reliable classification
Data Source
AI summary
A learning device includes a class classification learning unit that learns class classification of a classification target by using a loss function in which a loss is calculated to become smaller as a magnitude of a difference between a function value obtained by inputting a log-likelihood ratio to a function having a finite value range and a constant associated with a correct answer to the class classification of the classification target becomes smaller, the log-likelihood ratio being the logarithm of a ratio between the likelihood that the classification target belongs to a first class and the likelihood that the classification target belongs to a second class.