Selection-Inference Neural Networks With Traceable Reasoning

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Neural network-based approaches for making control decisions or diagnosing faults in mechanical systems are often opaque, making it difficult to understand the reasoning behind decisions, which is a concern for safety-critical applications.

Innovation Solution

A system that generates responses to queries by alternating between selection and inference steps, providing a traceable reasoning process in natural language that justifies the final answer, ensuring transparency and interpretability.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Extent of automation

If a neural network-based approach is used to make control decisions, then the decision-making capability is improved, but the interpretability of the reasoning process deteriorates

Engineering Contradiction:
Improvedecision-making capabilityVSAvoidinterpretability of reasoning
Core Design Contradiction:
Extent of automationVSLoss of information

Solution Approach 1:

The patent introduces an intermediary component that generates natural language explanations as a mediator between the neural network's internal reasoning process and the external observer. This intermediary translates the network's decisions into human-interpretable language without altering the underlying decision-making capability, thus resolving the contradiction between automation performance and interpretability.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Productivity

If a neural network is used as a black box for control decisions, then the productivity is improved, but the reliability for safety-critical applications deteriorates

Engineering Contradiction:
Improvedecision-making speedVSAvoidtrust in decision-making
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent implements a feedback mechanism where natural language explanations are generated and provided back to users alongside the neural network's decisions. This feedback loop allows users to understand and verify the reasoning process, thereby building trust and reliability for safety-critical applications while maintaining the high productivity benefits of neural network-based decision-making.

Inventive Principle:
Principle #23Feedback

Data Source

PatentUS20230365146A1Selection-inference neural network systems
Publication Date: 2023.11.16 GDM HOLDING LLC
  • US20230365146A1 patent drawing
  • US20230365146A1 patent drawing
  • US20230365146A1 patent drawing

AI summary

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating a response to a query input using a selection-inference neural network.