Historical Rule-Model Matching for Task-Specific ML Generation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing methods for designing machine learning models are manual, resource-intensive, and time-consuming, often resulting in sub-optimal results due to the lack of refinement of feature-extraction rules and ML models to specific applications, tasks, or data.

Innovation Solution

A computer system automatically searches for candidate ML models and feature-extraction rules by matching historical interaction datasets to a test dataset, iteratively evaluating combinations to generate a model-and-rule combination that meets a specified performance metric, reducing manual effort and computational overhead.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If manual design methods are used for machine learning models, then model performance can be optimized through expert knowledge, but the process becomes resource-intensive and time-consuming

Engineering Contradiction:

Improvemodel performanceVSAvoiddevelopment time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The system performs preliminary actions by pre-processing the test dataset to identify its characteristics (data types, metadata, distributions) before model selection. Historical datasets are pre-analyzed and stored with their associated models and feature-extraction rules, enabling rapid matching and retrieval during the model generation process, thus reducing development time while maintaining performance quality

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system creates copies of historical datasets that share similar characteristics with the test dataset. Instead of manually designing models from scratch, it replicates successful model-and-rule combinations from historical contexts and adapts them to the current test dataset, significantly reducing both time and resource requirements while preserving model effectiveness

Inventive Principle:

Principle #26Copying

2Loss of time

If existing automation techniques are used to generate model combinations, then development time is reduced, but the models do not actually generate new configurations and produce sub-optimal results

Engineering Contradiction:

Improvedevelopment timeVSAvoidmodel performance

Core Design Contradiction:

Loss of timeVSReliability

Solution Approach 1:

The system automatically modifies parameters of both the machine learning models and feature-extraction rules based on the characteristics of the test dataset. It adjusts model parameters (such as learning rates, tree depths, network architectures) and feature-extraction parameters (such as feature selection criteria, transformation parameters) to optimize performance for the specific application context, thereby generating truly new and optimized model configurations rather than simple combinations

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The system implements feedback mechanisms by evaluating the performance of generated model combinations on the test dataset and using this information to iteratively refine both model parameters and feature-extraction rules. This closed-loop approach ensures that the automation process continuously improves model performance based on actual results, preventing sub-optimal outcomes

Inventive Principle:

Principle #23Feedback

3Reliability

If comprehensive evaluation of all model and rule combinations is performed, then optimal model performance is achieved, but computational overhead increases significantly

Engineering Contradiction:

Improvemodel performanceVSAvoidcomputational overhead

Core Design Contradiction:

ReliabilityVSUse of energy by moving object

Solution Approach 1:

The system applies partial action by evaluating only the most promising model-and-rule combinations rather than exhaustively testing all possible combinations. It uses characteristic matching between historical and test datasets to filter and prioritize candidate models, evaluating only those combinations that are most likely to perform well based on dataset similarity, thus reducing computational overhead while maintaining the ability to find optimal solutions

Inventive Principle:

Principle #16Partial or excessive action

Solution Approach 2:

The system performs preliminary filtering by analyzing dataset characteristics and matching them to historical patterns before conducting full model evaluation. This pre-screening process eliminates obviously sub-optimal combinations early in the process, reducing the search space and computational requirements for the subsequent detailed evaluation phase

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS12462151B2Generating new machine learning models based on combinations of historical feature-extraction rules and historical machine-learning models

Publication Date: 2025.11.04 ADOBE INC

AI summary

Techniques for generating new machine learning (ML) systems are described. In an example, a computer system receives a request specifying a task and a performance metric for the new ML model via a user interface. In response, the computer system dynamically generates new feature-extraction rules and new machine learning models based on a rule-model combination that would perform the specified task at a level meeting or exceeding the performance metric.