Dynamic Dataset Sizing for Machine Learning Convergence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning methods face inefficiencies when dealing with multiple machine learning algorithms and varying training dataset sizes, as they struggle to achieve high prediction performance efficiently, often requiring excessive computational resources and time.
Innovation Solution
A computer program that identifies the maximum prediction performance score across multiple machine learning algorithms and dataset sizes, dynamically determines the optimal dataset size for each algorithm to maximize prediction performance and minimize runtime, using regression analysis to estimate future performance and runtime.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a larger-sized training dataset is used to achieve higher prediction performance, then the prediction accuracy improves, but the model learning time increases significantly
Solution Approach 1:
The patent applies dynamics by making the training dataset size adjustable and variable throughout the learning process. The system dynamically changes the dataset size from small to large based on the current prediction performance and improvement rate, rather than using a fixed size. This allows the system to adapt the resource allocation to the actual learning needs at each stage.
Solution Approach 2:
The patent changes the parameter of training dataset size based on measured performance metrics. By monitoring the improvement rate of prediction performance and comparing it against thresholds, the system adjusts the dataset size parameter to optimize the balance between learning time and prediction performance, preventing both premature termination and excessive learning time.
2Measurement precision
If multiple machine learning algorithms are executed with various training dataset sizes to find the optimal model, then the prediction performance improves, but the computational complexity and execution time increase
Solution Approach 1:
The patent segments the search space of multiple algorithms and dataset sizes into manageable groups. By dividing the comprehensive search into structured segments (different algorithm categories, different dataset size ranges), the system can systematically explore options without being overwhelmed by the full combinatorial complexity of all possible combinations.
Solution Approach 2:
The patent applies partial action by not exhaustively testing all possible algorithm-dataset combinations. Instead, it performs a controlled subset of executions based on preliminary assessments and performance thresholds, achieving sufficient prediction performance without the excessive computational cost of complete enumeration.
3Measurement precision
If the system executes machine learning algorithms iteratively with increasing dataset sizes, then the prediction performance converges to optimal levels, but the total execution time increases
Solution Approach 1:
The patent implements feedback mechanisms by continuously measuring the improvement rate of prediction performance after each learning execution. This feedback is used to adjust subsequent dataset sizes and determine when to terminate the iterative process, creating a closed-loop system that accelerates convergence by responding to actual performance gains rather than following a predetermined schedule.
Solution Approach 2:
The patent uses periodic evaluation of prediction performance at structured intervals during the iterative process. By performing assessments at regular cycles and comparing improvement rates against thresholds, the system creates a rhythmic pattern of learning and evaluation that prevents both premature termination and unnecessarily prolonged execution, thereby improving convergence speed.
Data Source
AI summary
A machine learning management apparatus identifies a maximum prediction performance score amongst a plurality of prediction performance scores corresponding to a plurality of models generated by executing each of a plurality of machine learning algorithms. As for a first machine learning algorithm having generated a model corresponding to the maximum prediction performance score, the machine learning management apparatus determines a first training dataset size to be used when the first machine learning algorithm is executed next time based on the maximum prediction performance score, first estimated prediction performance scores, and first estimated runtimes. As for a second machine learning algorithm different from the first machine learning algorithm, the machine learning management apparatus determines a second training dataset size to be used when the second machine learning algorithm is executed next time based on the maximum prediction performance score, second estimated prediction performance scores, and second estimated runtimes.