ML Pattern Cataloging for Automated Model Recombination
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning (ML) models lack an efficient mechanism for sharing and combining learned patterns across models, limiting their self-learning and growth capabilities.
Innovation Solution
The system and method enable the identification and cataloging of building block pattern functions (bbDNAs) associated with mature ML models, allowing for the automated creation of new ML models by combining these bbDNAs based on their weights and descriptions, without manual intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If machine learning models are trained independently without sharing mechanisms, then each model can learn patterns from data, but the learning time is extended and errors are increased due to redundant learning
Solution Approach 1:
The patent combines multiple machine learning models into an ensemble system where their predictions are aggregated. This merging approach allows the system to leverage patterns learned by individual models while reducing overall error rates through diversity in model perspectives and decision-making processes.
Solution Approach 2:
The patent creates a universal pattern repository that stores learned patterns from multiple models, making these patterns available for reuse across different modeling tasks. This multi-functional approach allows patterns to be applied universally rather than requiring redundant learning for each specific model.
2Adaptability or versatility
If machine learning models are trained from scratch for each task, then models can be highly specialized, but the complexity of the overall system increases and manual intervention is required
Solution Approach 1:
The patent segments the machine learning system into independent model components that can be developed, trained, and maintained separately. Each model can be specialized for specific tasks while the overall system manages these segments through automated ensemble mechanisms, reducing the complexity burden on any single component.
Solution Approach 2:
The patent implements self-service mechanisms where the system automatically selects, combines, and evaluates model ensembles without manual intervention. The automated framework handles model registration, pattern extraction, and ensemble construction, reducing operational complexity while maintaining adaptability.
3Reliability
If machine learning models continuously evolve and learn from data, then model effectiveness improves, but the mechanism for sharing and combining learned patterns is lacking
Solution Approach 1:
The patent implements feedback loops where model predictions and performance metrics are continuously monitored. Learned patterns are extracted from successful predictions and fed back into the pattern repository, which then informs future model training and ensemble construction, creating an automated evolution mechanism.
Solution Approach 2:
The patent introduces a pattern repository as an intermediary between individual machine learning models. This mediator captures, stores, and manages learned patterns, enabling automatic sharing and combination of knowledge across models without direct model-to-model communication.
Data Source
AI summary
Aspects of the subject disclosure may include, for example, a device having a processing system including a processor; and a memory that stores executable instructions that, when executed by the processing system, facilitate performance of operations, including: selecting a first building block pattern function (bbDNA) having a first weight from a catalog, wherein the first bbDNA is based on a first mature pattern reaching a first bbDNA frequency threshold, wherein the first bbDNA is associated with at least one machine learning (ML) model of a plurality of ML models; selecting a second bbDNA having a second weight from the catalog, wherein the second bbDNA is associated with at least one ML model in the plurality of ML models; and creating a new ML model based on a combination of the first bbDNA and the second bbDNA. Other embodiments are disclosed.