Reusable ML Workflow Execution Through GUI Scheme Editing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The setup and use of data models for machine learning or neural networks are time-intensive and require specialized knowledge, making them inaccessible to non-experts.
Innovation Solution
A method for creating and modifying machine learning schemes through a graphical user interface, allowing users to load, modify, and save data models and tasks, with parallel execution and authorization, enabling iterative workflow formation and query result processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If data models for machine learning or neural networks are set up and used, then data analysis and prediction capabilities are improved, but the process becomes time-intensive and requires specialized knowledge
Solution Approach 1:
The patent segments the complex machine learning workflow into discrete, manageable tasks that can be individually configured and executed. The system divides the overall data model setup into separate operational steps including data loading, feature generation, model training, and evaluation, allowing users to work with one segment at a time rather than facing the entire complex process simultaneously.
Solution Approach 2:
The patent introduces an intermediary automated system that mediates between the user and the complex machine learning infrastructure. This intermediary layer provides pre-configured templates, automated task scheduling, and simplified interfaces that translate user-friendly requests into complex backend operations, shielding users from the underlying complexity while maintaining reliable data analysis capabilities.
2Reliability
If data models for machine learning or neural networks are set up and used, then data analysis and prediction capabilities are improved, but the process requires specialized knowledge and training
Solution Approach 1:
The patent applies preliminary action by providing pre-configured templates and pre-trained models that users can directly apply to their data without needing to understand the underlying complex configurations. The system performs preliminary setup work including data preprocessing pipelines, feature engineering templates, and model parameter configurations, allowing users to bypass the learning curve and directly utilize sophisticated data analysis capabilities.
Solution Approach 2:
The patent enables users to copy and reuse successful data model configurations and workflows across different projects. By providing template copying functionality and workflow reuse capabilities, the system allows users to leverage existing expertise embedded in pre-configured models without needing to develop new complex setups, thereby improving ease of operation while maintaining analysis reliability.
3Adaptability or versatility
If machine learning schemes are created and modified through iterative processes, then workflow flexibility and adaptability are improved, but the time required for setup and execution increases
Solution Approach 1:
The patent implements dynamics by providing a flexible workflow system where task configurations can be dynamically modified and adjusted during execution. The system allows users to add, remove, or modify tasks in the workflow sequence, change parameter values, and adapt data loading configurations on-the-fly without requiring complete workflow reconfiguration, thereby enabling iterative improvement while minimizing time loss through incremental changes rather than complete restarts.
Solution Approach 2:
The patent reduces setup time through preliminary action by providing pre-configured task templates and default configurations that can be quickly adapted to specific needs. Rather than building workflows from scratch during each iterative process, users start with pre-established task structures including data loading templates, feature generation configurations, and model training parameters, significantly reducing the time required for each iteration while maintaining full adaptability.
Data Source
AI summary
A method includes loading a first machine learning scheme, wherein the first scheme includes a set of data models, an assignment of a first set of values to a first feature, and a first set of tasks reliant on the set of data models and the set of features. The method includes implementing the first scheme. The method includes, in response to detecting user inputs, displaying user interface elements for modifying a saved scheme. The method includes, in response to detecting user inputs, modifying the first scheme. Modifying the first scheme includes modifying the first set of tasks and saving the tasks as a second set of tasks. The second set of tasks includes a task for generating an output data set via the set of data models. The method includes saving the modified first scheme as a second scheme, implementing the second scheme, and outputting the output data set.