Hierarchical Feature Library for Automated AI Model Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI/ML solutions for managing equipment in buildings are costly and resource-intensive, requiring manual selection of algorithms and feature sets, which is time-consuming and often inefficient due to the need for multiple iterations and deep domain knowledge.
Innovation Solution
A system and method that utilize a feature set library and recommendation engine to create and recommend hierarchical feature sets, using a validation module and labelling module to provide preprocessed and contextual features for AI/ML platforms, allowing for automated feature selection and iterative processing within specified error margins or resource limits.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If manual identification and selection of AI/ML algorithms and feature sets is performed, then the model can be customized to specific problems, but the process becomes time-consuming and resource-intensive
Solution Approach 1:
The patent applies preliminary action by pre-processing and contextualizing feature sets before they are needed for model development. The system creates a library of pre-processed features with associated metadata, schemas, and contextual information in advance, so that when users need to develop models, they can quickly select from pre-prepared options rather than creating features from scratch each time.
Solution Approach 2:
The patent introduces an intermediary layer between raw data and AI/ML models through a feature store that acts as a mediator. This feature store contains pre-processed, contextualized features with metadata and schemas that bridge the gap between raw equipment data and model requirements, eliminating the need for users to manually process raw data each time they develop a model.
2Reliability
If multiple AI/ML models are used to achieve performance objectives, then the system can handle complex equipment management tasks, but the processing requirements and costs increase significantly
Solution Approach 1:
The patent merges multiple feature processing operations into a unified feature store infrastructure. Instead of each AI/ML model independently processing raw data, the system combines feature extraction, preprocessing, contextualization, and storage into a single shared resource that multiple models can utilize, thereby reducing redundant processing and overall computational requirements.
Solution Approach 2:
The feature store serves multiple functions simultaneously: it stores raw features, pre-processed features, contextual information, metadata, and schemas. This universal infrastructure supports multiple AI/ML models and use cases without requiring separate processing pipelines for each model, reducing overall processing requirements while maintaining reliability.
3Measurement precision
If labelled datasets are created manually for AI/ML models, then the training data can be accurate and relevant, but the cost of creation becomes very high
Solution Approach 1:
The patent applies copying by creating reusable feature templates and schemas that can be replicated across multiple use cases. Instead of manually creating labelled datasets from scratch for each problem, the system copies and adapts pre-existing feature definitions, contextual information, and schemas to new scenarios, significantly reducing the cost of data preparation while maintaining accuracy through proven templates.
4Ease of operation
If users lack deep domain knowledge and expertise, then the system is easier to use, but the ability to identify the right AI/ML algorithm and inputs is constrained
Solution Approach 1:
The patent implements self-service through an automated feature recommendation system that provides contextual information, metadata, and schemas to guide users in selecting appropriate features and algorithms. The system automatically suggests relevant features based on the problem type and provides contextualized information about each feature's meaning and usage, enabling users without deep domain knowledge to effectively develop models while maintaining adaptability to different problem types.
Data Source
AI summary
A system and a method for recommending feature sets for a plurality of equipment to a user. The method includes creating a library of contextual and preprocessed feature sets in a hierarchical manner for recommending features sets to a user. The method also includes compiling a plurality of hierarchical feature sets with a last feature set in a hierarchy being generated using an output of a module and incrementally adding the generated feature sets using different modules to the feature sets generated by the module. The method includes validating the generated feature sets to remove errors and using the validated feature sets as labelled data for previous feature sets and using attributes to categorize the labelled data corresponding to the contextual and pre-processed feature sets.