ML Pipeline Visualization Predictor for Non-Expert Customization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current AutoML solutions and open-source software databases are insufficient for non-experts to fully implement new machine learning projects due to the scarcity of ML experts and the difficulty in finding and modifying existing projects for new requirements.
Innovation Solution
A method that utilizes a visualization predictor to automatically generate visualizations for machine learning pipelines by synthesizing functional blocks using a hierarchical approach, extracting code and dataset features, and applying developed rules to predict visualization commands.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If current AutoML solutions and open-source software databases are used, then machine learning projects can be implemented, but the solutions are simplistic and partial, making it difficult for non-experts to find and modify existing ML projects for new requirements
Solution Approach 1:
The patent segments the ML pipeline into functional blocks (data loading, preprocessing, modeling, evaluation) and visualizes each block separately. This segmentation allows non-experts to easily navigate, select, and modify specific parts of the pipeline without understanding the entire complex system, directly addressing the difficulty in modifying existing projects.
Solution Approach 2:
The patent introduces a visualization layer as an intermediary between the ML pipeline and the user. This visualization layer translates complex pipeline structures into intuitive graphical representations, enabling non-experts to interact with and modify the pipeline without needing to understand the underlying code complexity.
2Loss of information
If ML pipelines are made more comprehensive and visualizable, then insights into functional blocks and dataset features are provided, but the system complexity increases
Solution Approach 1:
The patent creates a visual copy or representation of the ML pipeline structure. Instead of directly manipulating complex code, users interact with a simplified visual copy that mirrors the pipeline's functional blocks and data flows, reducing the cognitive load while maintaining comprehensive information about the pipeline.
Solution Approach 2:
The patent adds a visual dimension to the pipeline representation. By transforming the linear or hierarchical code structure into a two-dimensional visual layout with nodes and edges, the system provides intuitive spatial relationships between functional blocks, making complex pipeline information more accessible without increasing operational complexity.
Data Source
AI summary
A method may include obtaining a machine learning (ML) pipeline including a plurality of functional blocks within the ML pipeline. The method may also include using the ML pipeline as an input to a visualization predictor, where the visualization predictor may be trained to output one or more visualization commands based on relationships between the visualization commands and the functional blocks within the pipeline. The method may additionally include invoking the visualization commands to instantiate the ML pipeline with visualizations generated by the one or more visualization commands.