Logical Inference Training Data via Argument Stacking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for generating training data for artificial intelligence models that simulate human-like logical inference are limited in generating proofs with multiple steps and stacking of arguments, making them inadequate for complex decision-making scenarios.
Innovation Solution
A computer system that generates training data by representing arguments as logical expressions, searching for connectable proof trees, and converting them into text form to create a model capable of performing logical inference through repeated non-biased arguments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If existing methods (NPL 1 and NPL 2) are used to generate training data, then simple proofs or single arguments can be generated, but complex proofs with multiple steps and argument stacking cannot be generated
Solution Approach 1:
The patent segments complex proofs into individual argument units, each representing a single inference step from premises to conclusion. These segmented arguments can then be systematically combined and stacked to form complex multi-step proofs, enabling the generation of training data with arbitrary complexity levels while maintaining manageable individual components
Solution Approach 2:
The patent implements argument stacking where arguments are nested within proof structures, and multiple arguments are combined to form higher-level proofs. This nested structure allows simple arguments to be embedded within complex proofs, creating a hierarchical organization that scales from basic to advanced logical reasoning tasks
2Reliability
If more complex proofs with multiple argument steps are generated, then the model's logical inference capability is improved, but the generation process becomes more difficult
Solution Approach 1:
The patent performs preliminary action by pre-defining argument templates with standardized logical structures (premises, inference rules, conclusions). These pre-prepared argument units can be systematically combined without requiring complex real-time generation logic, making the creation of multi-step proofs straightforward while ensuring logical correctness
Solution Approach 2:
The patent implements a self-service mechanism where the system automatically searches for compatible arguments to stack based on premise-conclusion matching. The argument stacking process is automated through algorithmic combination of pre-defined argument units, eliminating the need for manual construction of complex proofs while maintaining high logical inference accuracy
Data Source
AI summary
Training data used for training a model that performs a logical inference is generated. A computer system that generates the training data used for training the model configured to perform the logical inference holds argument data representing an argument that leads to a conclusion proposition from a plurality of premise propositions. The proposition is expressed as a logical expression. The computer system searches for the argument data whose conclusion is the premise proposition of the argument data or the argument data whose premise is the conclusion proposition of the argument data and performs combination to generate proof data representing a proof that leads to a conclusion proposition by repeating the argument a plurality of times, converts the proof data into a text expressed as a language expression, and generates the training data using the text.