ML Hyperparameter Selection Using Parallel Instance Evaluation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing hyperparameter tuning methods for machine learning algorithms are limited in their applicability and often require the algorithm to be parameterizable, which restricts their use to specific situations, and struggle to find the global optimum, especially in complex operational scenarios like automated traffic planning.
Innovation Solution
A computer-implemented method that provides multiple instances of a machine learning algorithm with different hyperparameter values, maps input states to outputs, evaluates a quality metric for each, and computes statistics to select the best-performing instance for continued use, allowing for broader applicability and increased chances of finding the global optimum without requiring parameterization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If existing hyperparameter tuning methods are used, then the algorithm can be tuned in specific situations, but the applicability is limited to parameterizable algorithms only
Solution Approach 1:
The patent applies universality by creating a hyperparameter tuning method that works with any machine learning algorithm regardless of whether it is parameterizable. The method uses multiple instances of the algorithm with different hyperparameter values and selects the best performing instance, making the tuning approach universally applicable across different algorithm types and contexts.
2Measurement precision
If existing hyperparameter tuning methods are used, then tuning can be performed, but the ability to find global optimum is reduced
Solution Approach 1:
The patent applies preliminary action by evaluating multiple hyperparameter configurations in parallel through multiple algorithm instances before deployment. This advance evaluation and comparison of different hyperparameter sets allows the system to identify the global optimum more reliably before actual use, rather than performing sequential tuning during operation.
3Measurement precision
If multiple instances of ML algorithm are evaluated, then the global optimum can be found, but the computational resources increase
Solution Approach 1:
The patent applies partial action by evaluating a finite set of multiple hyperparameter configurations rather than exhaustively searching the entire hyperparameter space. This selective evaluation of a manageable number of instances provides sufficient optimization accuracy while controlling computational resource consumption, avoiding the need to test all possible hyperparameter combinations.
Data Source
AI summary
A computer-implemented method of managing a machine-learning, ML, algorithm which is dependent on one or more hyperparameters is described. The method comprises: providing a plurality of instances of the ML algorithm using different values of the hyperparameters; obtaining a plurality of input states; mapping each predefined input to a plurality of outputs using the instances of the ML algorithm; evaluating a predefined quality metric for the outputs; and on the basis of statistics of the quality metric for the outputs, selecting at least one instance of the ML algorithm for continued use. In some embodiments, a technical system is controlled using one primary instance of the ML algorithm. If the selected instance of the ML algorithm is not the primary instance, the primary instance may optionally be replaced by the selected instance.