Machine Learning Model Integration via Automatic Feature Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The integration of machine learning (ML) models into applications is complex and requires technical knowledge, custom coding, and assistance from data scientists, making it inaccessible to non-technical users and prolonging deployment cycles.
Innovation Solution
An integration system that allows non-technical users to import ML models directly into a computing platform using a user interface, automatically mapping input features and output predictions to dimensions in the target application, eliminating the need for custom coding.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If custom coding is used to integrate ML models into applications, then the integration can be tailored to specific model and application requirements, but the deployment cycle becomes longer and more costly
Solution Approach 1:
The system performs preliminary actions by pre-defining integration templates and configurations for common ML model types and application platforms. When a user wants to integrate an ML model, the system automatically selects and applies the appropriate pre-configured template, eliminating the need for custom coding and significantly reducing deployment time while maintaining adaptability through template selection.
Solution Approach 2:
The patent introduces an intermediary integration layer that sits between the ML model and the application. This intermediary handles the complexity of integration through standardized interfaces and automatic configuration, allowing users to integrate models without needing to write custom code. The intermediary translates between different model formats and application requirements automatically.
2Reliability
If data scientists with technical knowledge implement ML models, then the models can be properly configured and integrated, but non-technical users cannot implement models without assistance
Solution Approach 1:
The system enables self-service by allowing non-technical users to independently integrate and deploy ML models through a simplified interface. The automatic model detection, format identification, and integration configuration eliminate the need for data scientist assistance. Users can upload models, select target applications, and complete integration through guided workflows without requiring technical expertise.
Solution Approach 2:
The patent changes the parameters of the integration process from requiring detailed technical knowledge to using simple user-friendly selections. The system transforms complex integration parameters into predefined options that non-technical users can easily select, such as choosing from common model types and target applications, while the system handles the complex parameter mapping and configuration automatically.
3Ease of manufacture
If ML models are built in external computing systems, then specialized ML platforms can be used, but the complexity of using ML models increases for application users
Solution Approach 1:
The patent implements universality by creating a platform that can integrate multiple types of ML models from various external computing systems into multiple different application types through a single unified interface. The system supports various model formats (PMML, CSV, Excel, etc.) and can deploy to different target applications, eliminating the need for separate integration processes for each model-application pair and reducing overall system complexity.
Solution Approach 2:
The system segments the integration process into distinct, manageable steps: model upload, format detection, template selection, configuration, and deployment. Each step is handled automatically or through simple user input, breaking down the complex integration task into manageable segments that reduce the perceived complexity for users while maintaining the capability to handle diverse model types.
Data Source
AI summary
Systems, methods, and other embodiments associated with integrating a machine learning model into a target application are described. In one embodiment, a method includes parsing a definition file that represents the selected ML model to (1) identify one or more input features defined for the ML model, and (2) identify one or more output predictions that the ML model is configured to generate. The input features and output predictions are mapped to locations within the target application. An executable script is generated and configured to function as a data controller or a data communication interface to obtain the input values from the target application and pass data between the target application and the ML model. In response to obtaining the output predictions, the output predictions are inserted in one or more result locations within the target application.