Mini-ML Variants for Low-Cost Algorithm Selection Accuracy

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing machine learning algorithms face challenges in selecting the best algorithm for a given task due to high computational overhead and low accuracy in predicting performance, especially when considering variants of the same algorithm, leading to inefficient resource utilization.

Innovation Solution

The development of mini-machine learning (mini-ML) algorithms that generate computationally less costly variants while maintaining performance by iteratively modifying hyper-parameters of a reference variant, using cross-validation and meta-learning techniques to reduce computational resources and improve accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If full variety of machine learning techniques are automatically explored, then algorithm selection accuracy is improved, but computational overhead increases significantly

Engineering Contradiction:
Improvealgorithm selection accuracyVSAvoidcomputational overhead
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent segments the machine learning algorithm selection process into two distinct phases: (1) a lightweight filtering phase using landmark algorithms to eliminate clearly suboptimal algorithms, and (2) a detailed evaluation phase using mini-ML models only for the remaining candidate algorithms. This segmentation reduces computational overhead by avoiding exhaustive exploration of all algorithms while maintaining selection accuracy through targeted detailed evaluation of promising candidates.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent creates simplified copies (mini-ML models) of complex machine learning algorithms that replicate their essential behavior at a fraction of the computational cost. These mini-ML models are trained on small datasets and can quickly predict performance metrics, enabling efficient algorithm selection without requiring full training and validation of each candidate algorithm.

Inventive Principle:
Principle #26Copying

2Use of energy by moving object

If landmark algorithms are used to predict performance, then computational overhead is reduced, but prediction accuracy deteriorates

Engineering Contradiction:
Improvecomputational overheadVSAvoidprediction accuracy
Core Design Contradiction:
Use of energy by moving objectVSMeasurement precision

Solution Approach 1:

The patent introduces mini-ML models as intermediary representations between landmark algorithms and full ML algorithms. The landmark algorithms provide initial performance estimates that guide selection of which mini-ML models to train and evaluate. The mini-ML models then serve as more accurate predictors for final algorithm selection, bridging the gap between computational efficiency and prediction accuracy.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent replaces the mechanical system of directly training and validating full ML algorithms (computationally expensive) with a substituted system using mini-ML models (computationally efficient). The mini-ML models are designed to mimic the behavior of full algorithms but with significantly reduced computational requirements, enabling accurate performance prediction without the overhead of complete algorithm training.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

3Measurement precision

If variants of the same algorithm are considered, then selection accuracy is improved, but device complexity increases

Engineering Contradiction:
Improveselection accuracyVSAvoidalgorithm variant complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies local quality by creating mini-ML models with different levels of sophistication tailored to specific algorithm variants and problem characteristics. Rather than uniformly complex models for all variants, the system selectively applies appropriate mini-ML model complexities based on the specific algorithm being evaluated and the available computational resources, optimizing the balance between selection accuracy and system complexity.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP3867827B1MINI-machine learning
Publication Date: 2025.11.26 ORACLE INT CORP
  • EP3867827B1 patent drawingFigure 1
  • EP3867827B1 patent drawingFigure 2
  • EP3867827B1 patent drawingFigure 3A~3B

AI summary

Techniques are described for generating and applying mini-machine learning variants of machine learning algorithms to save computational resources in tuning and selection of machine learning algorithms. In an embodiment, at least one of the hyper-parameter values for a reference variant is modified to a new hyper-parameter value thereby generating a new variant of machine learning algorithm from the reference variant of machine learning algorithm. Based on the cost metric of the new variant of machine learning algorithm and comparing the performance score for the new and reference variants, the system determines whether the modified reference machine algorithm is the mini-machine learning algorithm that is computationally less costly than the reference variant of machine learning algorithm but closely tracks the accuracy thereof