Machine Learning Decision Engine Real-Time Model Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
There is a need for methods to deploy and automate software-based decisioning systems using machine learning models and artificial intelligence that can be dynamically updated in real-time, allowing for recursive decision-making and improvement over time.
Innovation Solution
A computer-implemented method for generating and deploying machine learning-enabled decision engines (MLDEs) that involves receiving requests to generate MLDEs, identifying influencing data and performance metrics, generating predictive models, setting controls, validating the MLDEs, and retraining models using updated datasets, with the ability to convert MLDEs into microservices accessible via APIs and generate synthetic datasets for new MLDEs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models are continuously retrained with updated datasets in real-time, then decision-making accuracy is improved, but system complexity and computational resources increase
Solution Approach 1:
The system segments the machine learning model into a persistent model component and a recursive decisioning component. The persistent model is trained offline on historical data and stored, while the recursive decisioning component operates online using the stored model and updated datasets, dividing the complexity across different operational layers.
Solution Approach 2:
The system performs preliminary model training offline using historical data before deployment. The pre-trained model is stored and reused for recursive decisioning, avoiding the need to retrain the entire model in real-time and reducing online computational complexity.
2Adaptability or versatility
If machine learning models are dynamically updated in real-time, then adaptability to changing conditions is improved, but loss of time for data processing increases
Solution Approach 1:
The system implements periodic model retraining at predetermined intervals using accumulated updated datasets, rather than continuous retraining. This periodic approach maintains adaptability while reducing computational overhead and processing time compared to continuous updates.
Solution Approach 2:
The system creates a copy of the trained model and stores it in storage for reuse. This stored model copy can be quickly loaded and applied to new data without requiring the full training process again, enabling fast real-time decisioning while maintaining adaptability through periodic updates.
3Reliability
If controls are added to limit decision-making options, then reliability and compliance are improved, but device complexity increases
Solution Approach 1:
The system introduces a control layer as an intermediary between the recursive decisioning engine and the output. This control layer validates decisions against predetermined controls and rules, ensuring reliability and compliance without requiring the core decisioning engine to be modified, thus managing complexity through separation of concerns.
4Productivity
If synthetic datasets are generated for new MLDEs, then productivity in creating new decision engines is improved, but loss of information about real data distributions may occur
Solution Approach 1:
The system generates synthetic datasets that copy the statistical properties and distributions of real data without containing actual sensitive information. This allows rapid creation of training data for new MLDEs while preserving the essential data characteristics needed for model training through carefully constructed synthetic data generation.
Data Source
AI summary
The present disclosure relates generally to the generation and deployment of a machine learning-enabled decision engine (MLDE). The MLDE includes decision options that are composed of a discrete list of selectable options. Further, the MLDE includes data inputs that can be used to influence decisions made by the machine learning models of the MLDE. Controls are applied to the MLDE to overlay and bound the decisioning within guidelines established by an operator of the MLDE. Once the MLDE is established, the MLDE is validated and deployed for use by software applications to make decisions.