Neural Network Controller for Database Query Generation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current neural networks lack the capability to efficiently process natural language questions to generate executable database queries, requiring significant human supervision and being limited in their ability to perform complex reasoning across various domains.

Innovation Solution

A neural network system comprising a controller neural network and an operation subsystem that processes inputs to generate operation and data score distributions, allowing for the performance of arithmetic, logic, aggregation, comparison, print, reset, and string-matching operations on database tables, rows, and columns, enabling the generation of executable queries with minimal human intervention.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Extent of automation

If current neural networks are used to process natural language questions, then they can generate outputs, but they require significant human supervision and cannot efficiently generate executable database queries

Engineering Contradiction:
Improveautomation of query generationVSAvoidaccuracy of query generation
Core Design Contradiction:
Extent of automationVSReliability

Solution Approach 1:

The system segments the query generation process into multiple time steps, where at each step the controller neural network generates operation score distributions and data score distributions that guide specific operations. This step-by-step approach allows the system to build complex queries reliably through sequential decision-making rather than attempting to generate complete queries in a single step.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system uses dynamic score distributions that are updated at each time step based on previous operations and current state. The operation scores and data scores are recalculated dynamically, allowing the system to adapt its query generation strategy in real-time and improve reliability through iterative refinement rather than static single-step generation.

Inventive Principle:
Principle #15Dynamics

2Adaptability or versatility

If neural networks perform complex reasoning across various domains, then they can handle diverse tasks, but they lack the capability to efficiently process natural language questions to generate executable database queries

Engineering Contradiction:
Improvedomain applicabilityVSAvoidquery generation efficiency
Core Design Contradiction:
Adaptability or versatilityVSProductivity

Solution Approach 1:

The controller neural network is designed with a universal architecture that can handle multiple operation types (arithmetic, logic, aggregation, comparison, print, reset, string-matching) and work across different data domains. The same core network generates operation score distributions and data score distributions regardless of the specific domain or operation type, enabling versatile application while maintaining efficient query generation through a single unified system.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Extent of automation

If the system generates operation score distribution and data score distribution at each time step, then it can perform complex operations on database tables, rows and columns, but the system complexity increases

Engineering Contradiction:
Improveautonomous program inductionVSAvoidneural network architecture
Core Design Contradiction:
Extent of automationVSDevice complexity

Solution Approach 1:

The system implements feedback loops where the output of each time step (operation scores and data scores) becomes input for the next time step. The controller neural network uses the previous state and current input to generate updated score distributions, creating a recurrent architecture that provides feedback mechanisms. This allows autonomous program induction through iterative refinement while managing complexity through structured feedback rather than uncontrolled system growth.

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP3362950B1Neural network for querying a database
Publication Date: 2021.07.14 GOOGLE LLC
  • EP3362950B1 patent drawingFigure 1
  • EP3362950B1 patent drawingFigure 2
  • EP3362950B1 patent drawingFigure 3

AI summary

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for performing operations using data from a data source. In one aspect, a method includes a neural network system including a controller neural network configured to: receive a controller input for a time step and process the controller input and a representation of a system input to generate: an operation score distribution that assigns a respective operation score to an operation and a data score distribution that assigns a respective data score in the data source. The neural network system can also include an operation subsystem configured to: perform operations to generate operation outputs, wherein at least one of the operations is performed on data in the data source, and combine the operation outputs in accordance with the operation score distribution and the data score distribution to generate a time step output for the time step.