ML Workflow Engine for Prerequisite-Aware Model Deployment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The deployment of machine learning models requires significant time and resources from data scientists, data engineers, and software engineers, averaging up to 160 man-hours, due to the non-deterministic nature of these models and the need for real-time monitoring and updates.
Innovation Solution
A model execution workflow engine that allows data scientists to define and deploy machine learning models independently, with automated pre- and post-processing, and selects optimal cloud server systems based on predefined rules and user input, reducing the need for additional roles and enhancing deployment efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a traditional multi-role pipeline (data scientist, data engineer, software engineer) is used to deploy machine learning models, then the deployment process is thorough and reliable, but the deployment time and resource consumption increase significantly (up to 160 man-hours)
Solution Approach 1:
The patent merges the functions of data engineers and software engineers into an automated workflow engine that handles pre-processing, model execution, and post-processing. This consolidation eliminates the need for multiple human roles while maintaining deployment reliability through systematic automated checks and execution protocols.
Solution Approach 2:
The workflow engine enables data scientists to independently deploy models by providing self-service capabilities. The system automatically manages the deployment pipeline, including prerequisite verification, resource allocation, and model execution, allowing data scientists to complete deployments without requiring assistance from other specialized roles.
2Ease of operation
If data scientists independently define and deploy models using scripts, then deployment autonomy improves, but the complexity of managing pre-processing and post-processing prerequisites increases
Solution Approach 1:
The workflow engine acts as an intermediary between data scientists and the complex deployment infrastructure. It abstracts away the complexity of managing pre-processing and post-processing prerequisites by automatically verifying conditions, allocating resources, and coordinating execution steps, while allowing data scientists to maintain simple model definition scripts.
3Manufacturing precision
If manual model deployment processes are used with multiple roles involved, then comprehensive code review and refinement can be performed, but the number of man-hours required increases to an average of 160 hours
Solution Approach 1:
The patent replaces the mechanical system of manual code review and refinement by multiple human roles with an automated workflow engine that systematically manages the deployment process. The engine incorporates built-in validation, prerequisite checking, and execution monitoring that maintains code quality while dramatically reducing the time required from 160 man-hours to minutes.
4Reliability
If traditional deployment pipelines are used, then models can be properly monitored and updated in real-time, but the requirement for multiple specialized roles increases resource consumption
Solution Approach 1:
The workflow engine provides self-service monitoring and update capabilities that automatically track model performance and manage updates in real-time. This eliminates the need for dedicated data engineers and software engineers to manually monitor and maintain models, as the system performs these functions autonomously while maintaining high reliability.
Data Source
AI summary
A method for executing a machine learning model using a workflow engine includes receiving a model configuration including data related to the machine learning model, pre-processing steps having first prerequisites, and post-processing steps having second prerequisites; in response to a determination that the first prerequisites are not met, executing first operations; in response to a determination that the second prerequisites are not met, executing second operations; executing the pre-processing steps to provide first data, the first data including model inputs; causing transmission of the first data from the computer system to the cloud server system; causing execution of the machine learning model on the cloud server system; causing transmission of second data from the cloud server system to the computer system, the second data including an output of the machine learning model; executing the post-processing steps.