Composite Machine Learning Scenario Management System
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current database systems face challenges in managing and maintaining multiple trained machine learning models, particularly in ensuring consistency and accuracy of inferences due to differences in training data, platforms, and protocols, leading to high development and operational costs.
Innovation Solution
The implementation of a composite scenario management system that abstracts multiple machine learning scenarios into a single object, allowing for efficient training and aggregation of inferences without requiring code changes to the application, and optimizing data distribution across ML platforms to reduce processing load and improve consistency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multiple trained machine learning models are used to improve application functionality, then the functionality and accuracy of the application is improved, but the complexity of training and managing these models increases significantly
Solution Approach 1:
The patent combines multiple trained machine learning models into a single composite model object that encapsulates all individual models. This composite model presents a unified interface to applications, merging the functionality of multiple models while simplifying management. The composite model coordinates training and inference across all constituent models through a single object, reducing the operational complexity of handling multiple separate models.
Solution Approach 2:
The composite model acts as an intermediary between applications and multiple individual machine learning models. It provides a standardized interface that applications interact with, while the composite model internally manages the complexity of coordinating multiple underlying models. This mediator pattern isolates applications from the complexity of model management while enabling access to enhanced functionality.
2Adaptability or versatility
If applications directly manage multiple trained models, then model customization and control are maximized, but development and maintenance costs increase significantly
Solution Approach 1:
The composite model provides a universal interface that works with multiple different machine learning models through a single standardized object. This multi-functional design allows the same composite model structure to encapsulate various types of models (e.g., regression, classification, clustering) without requiring application-specific code changes. The universal interface handles model training, inference, and coordination consistently across different model types.
Solution Approach 2:
The composite model serves as an intermediary layer that shields applications from the complexity of managing multiple models. It provides a standardized, simplified interface while handling the intricate coordination of multiple underlying models internally. This reduces development costs by eliminating the need for applications to implement complex model management logic.
3Adaptability or versatility
If different machine learning platforms are used to train models, then model diversity and algorithm selection are improved, but consistency and accuracy of inferences deteriorate
Solution Approach 1:
The composite model applies different training strategies and data distributions to different constituent models based on their specific requirements. Each model within the composite can be trained on appropriately distributed data from the training dataset, with the composite model coordinating the training process to ensure consistency. This local optimization approach maintains inference consistency while allowing diversity in underlying algorithms and platforms.
Solution Approach 2:
The composite model dynamically adjusts training parameters such as data distribution and sampling strategies based on the specific requirements of each constituent model. By changing these parameters during the training process, the composite model ensures that models from different platforms receive appropriate training data distributions, thereby maintaining inference consistency and accuracy across diverse model types.
4Ease of operation
If applications are responsible for model training and inference, then direct control over the process is maximized, but operational complexity and maintenance effort increase
Solution Approach 1:
The patent extracts the complex model training and coordination logic from applications and encapsulates it within the composite model object. The composite model internally manages the training process, data distribution, and coordination of multiple constituent models, removing this operational complexity from applications. Applications retain control through the standardized interface while the composite model handles the intricate operational details.
Solution Approach 2:
The composite model provides self-service capabilities by automatically managing its own training process, data distribution, and coordination of constituent models. It autonomously handles the complexity of multi-model training and inference without requiring direct application intervention. This self-management reduces operational complexity while maintaining application control through the unified interface.
Data Source
AI summary
Systems and methods include reception of a request to train a composite machine learning scenario comprising a plurality of machine learning scenarios, in response to the request, identification of a training dataset associated with each of the plurality of machine learning scenarios and triggering of training of each of the plurality of machine learning scenarios based on an associated training dataset to generate a plurality of trained machine learning models, acquisition of metrics associated with each trained machine learning model, aggregation of the acquired metrics into training metrics associated with the composite machine learning scenario, and return of the training metrics associated with the composite machine learning scenario.