Machine Learning Parameter Optimization via Discontinuity Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In deep learning-based information processing systems, determining optimal parameter values for model generation is inefficient when the acquisition function is discontinuous, leading to difficulties in specifying appropriate parameter values for enhancing the percentage of correct answers.
Innovation Solution
A method that identifies discontinuity points in learning time relative to parameter variations, specifies continuous ranges for parameter values, calculates estimated performance improvements, and selects optimal parameter values for efficient machine learning model generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning is performed with varied learning parameters to optimize model performance, then the percentage of correct answers is improved, but the learning time becomes discontinuous and difficult to predict
Solution Approach 1:
The patent segments the learning parameter space into multiple ranges based on detected discontinuity points. Each range is evaluated separately to determine which range provides the best performance-to-time ratio, avoiding the need to search through all parameter values uniformly.
Solution Approach 2:
The patent changes the approach from directly optimizing for highest accuracy to optimizing for the ratio of accuracy improvement to learning time. This parameter transformation allows the system to find optimal parameters more efficiently by considering both performance and time together.
2Measurement precision
If the learning parameter is adjusted to improve model accuracy, then the percentage of correct answers increases, but the acquisition function becomes discontinuous making parameter specification difficult
Solution Approach 1:
The patent divides the continuous learning parameter space into discrete ranges separated by discontinuity points. This segmentation makes parameter specification easier by providing clear boundaries and guidance on which ranges to explore, rather than searching through all possible values.
Solution Approach 2:
The patent introduces discontinuity points as intermediary elements that mediate between the learning parameter and the acquisition function. These points serve as markers that simplify the specification process by indicating where significant changes in learning behavior occur.
3Measurement precision
If exhaustive search of learning parameters is performed to find optimal values, then the model performance is maximized, but the computational efficiency and productivity decrease
Solution Approach 1:
The patent segments the parameter search space into meaningful ranges based on discontinuity detection, allowing the system to focus computational resources on evaluating representative parameters from each range rather than exhaustively searching all possible values.
Solution Approach 2:
The patent performs partial action by evaluating only the most promising parameter ranges identified through discontinuity analysis, rather than performing exhaustive search. This partial evaluation achieves sufficient model performance while significantly improving computational efficiency.
Data Source
AI summary
A non-transitory computer-readable storage medium storing therein a learning program that causes a computer to execute a process includes: determining whether or not there is a discontinuity point at which a variation in a learning time relative to a variation in a learning parameter is discontinuous; specifying, when the discontinuity point is present, ranges of the learning parameter in which the variation in the learning time relative to the variation in the learning parameter is continuous, based on the discontinuity point; calculating, for each of the specified ranges, an estimated value of performance of trials using a trial parameter learned by machine learning per a learning time of machine learning using a learning parameter included in the range; and specifying a learning parameter which enables any of the estimated values selected in accordance with a magnitude of the estimated value among the calculated estimated values.