Database-Integrated ML Inference Engine for Real-Time Predictive Queries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Business intelligence applications struggle to accurately predict future events and infer user or product characteristics due to limitations in accessing large datasets for real-time analysis, requiring advanced machine learning techniques that are typically handled by data scientists, thus disconnecting business users from the capabilities they need.
Innovation Solution
A massively parallel real-time database-integrated machine learning inference engine that allows users to query machine learning models using standard query interfaces, enabling predictive analysis within existing business intelligence tools without modification, and allowing software developers to easily add ML capabilities through familiar pathways like SQL.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning techniques are used for predictive analysis, then prediction accuracy is improved, but system complexity increases and requires specialized data scientists
Solution Approach 1:
The patent introduces an intermediary layer that translates standard SQL queries into machine learning inference operations. This mediator component allows business users to interact with complex ML models through familiar SQL syntax, hiding the underlying complexity while maintaining high prediction accuracy through ML techniques.
2Adaptability or versatility
If machine learning models are integrated into database systems, then predictive capabilities are improved, but ease of operation deteriorates due to requiring specialized knowledge
Solution Approach 1:
The patent makes the database system universal by enabling it to perform both traditional data storage/querying functions and machine learning inference functions through a unified SQL interface. This allows the same database infrastructure to serve multiple purposes without requiring users to learn specialized ML operations.
Solution Approach 2:
The system enables business users to perform predictive analysis themselves using standard SQL queries without needing to consult data scientists. The database automatically handles the translation of SQL queries into ML inference operations, allowing users to serve their own predictive analysis needs directly.
3Measurement precision
If real-time predictive analysis is performed on large datasets, then analysis accuracy is improved, but processing time increases
Solution Approach 1:
The patent pre-loads and caches machine learning models and their associated data structures into the database system in advance. This preliminary action allows the models to be readily available in memory when queries are executed, eliminating the need to load large datasets and models from external storage during real-time analysis, thus reducing processing time while maintaining accuracy.
Data Source
AI summary
Techniques for massively-parallel real-time database-integrated machine learning (ML) inference are described. An ML model is deployed as one or more model serving units behind an endpoint. The ML model can be associated with a virtual table or function, and a query that is received that references the virtual table or function can be processed by issuing inference requests to the endpoint by the query execution engine(s).