Feature Recommendation Graph for ML Model Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing feature management systems in machine learning struggle to recommend relevant features for new models, as popularity-based recommendations are limited and often irrelevant across different models, leading to inefficient training processes and less accurate models due to the inclusion of irrelevant features.
Innovation Solution
A feature management system that generates a graph of ML models and features, using methods like Personalized PageRank, Matrix Factorization, or Random Walk to determine relevancy scores and recommend candidate features based on shared usage across multiple models, allowing users to propose features and select relevant ones for training new models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If popularity-based feature recommendation is used, then feature selection is automated, but recommendation relevance deteriorates across different models
Solution Approach 1:
The patent segments the monolithic feature recommendation problem into model-specific sub-problems. By building separate recommendation graphs for different model types (e.g., classification, regression, clustering), the system provides automated feature selection tailored to each model's specific requirements, thereby maintaining both automation and relevance.
Solution Approach 2:
The patent applies local quality by customizing feature recommendation strategies for different model types. Each model type receives feature recommendations optimized for its specific characteristics and requirements, rather than applying a uniform popularity-based approach across all models, thus improving recommendation relevance locally for each model category.
2Quantity of substance
If all available features are used for training, then feature completeness is maximized, but computation cost increases
Solution Approach 1:
The patent extracts and removes irrelevant features from the training process by using the recommendation graph to identify and exclude features that do not contribute to the specific model type. This extraction of unnecessary features reduces computation power consumption while maintaining feature completeness for relevant attributes.
Solution Approach 2:
The patent applies partial action by selecting only the necessary subset of features required for effective model training, rather than using all available features. This partial feature selection approach reduces computational overhead while maintaining sufficient feature completeness for achieving accurate model performance.
3Reliability
If domain expert intuition is used for feature selection, then feature relevance improves, but selection time increases
Solution Approach 1:
The patent implements self-service by enabling the system to automatically generate model-specific feature recommendations without requiring domain expert intervention. The recommendation graph autonomously identifies relevant features based on model type characteristics, maintaining high feature selection accuracy while eliminating the time loss associated with expert consultation.
Solution Approach 2:
The patent applies preliminary action by pre-computing and storing feature recommendation graphs for different model types in advance. When a new model is created, the system can quickly retrieve and apply pre-established feature recommendations, achieving both high accuracy and fast selection without requiring real-time expert analysis.
4Reliability
If model-specific feature recommendations are implemented, then feature relevance improves, but system complexity increases
Solution Approach 1:
The patent applies universality by creating a multi-functional feature management system that handles multiple model types through a unified graph-based framework. The same core infrastructure supports different model types (classification, regression, clustering, etc.), reducing system complexity compared to implementing separate systems for each model type while maintaining model-specific recommendation relevance.
Data Source
AI summary
A system maintains a data store for managing machine-learning (ML) models and features that are used by the models. The system generates a graph including nodes for each model and a node for each feature, and edges linking models and features that are used by the models. For a new model to be trained, the system receives a proposed feature corresponding to a node in the graph, and identifies one or more candidate features corresponding to nodes in the graph based in part on relevancy scores between the proposed feature with other features corresponding to nodes in the graph. The system presents in a user interface a suggestion to use one or more candidate features with the new model. Responsive to receiving a user selection of at least one candidate feature, the system causes the new model to be trained using the selected candidate feature and the proposed feature.