Deep Neural Network Linguistic Expressions for Explainable Predictive Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current predictive models are unintelligible to human users, making them difficult to understand, modify, and optimize, which requires additional computing resources and hinders efficient production.
Innovation Solution
A deep neural network (DNN) is trained on labeled sentences to produce linguistic expressions that can be modified by human experts, forming explainable predictive models capable of labeling unlabeled sentences, using techniques such as semantic role labeling and neural logic programming.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If current predictive models are used, then prediction accuracy is achieved, but model intelligibility deteriorates making them difficult to understand and modify
Solution Approach 1:
The patent introduces linguistic expressions as an intermediary layer between the deep neural network and human users. These linguistic expressions serve as a mediator that translates the internal representations of the DNN into human-understandable forms, allowing users to comprehend and modify model behavior without needing to understand the complex neural network architecture directly.
Solution Approach 2:
The patent segments the predictive model into distinct components: the deep neural network for accurate prediction and the linguistic expressions for human interpretation. This segmentation allows each component to fulfill its specific function - the DNN handles complex pattern recognition while the linguistic expressions provide interpretable representations that can be easily understood and modified by humans.
2Reliability
If current predictive models are used, then prediction capability is maintained, but additional computing resources are required due to lack of optimization
Solution Approach 1:
The patent enables the predictive model to serve itself by generating linguistic expressions that can be used to optimize and refine the model's own behavior. The linguistic expressions provide a mechanism for the system to automatically understand and adjust its own functioning, reducing the need for external computational resources for model optimization and interaction.
3Measurement precision
If current predictive models are used, then classification accuracy is achieved, but model modification efficiency deteriorates due to difficulty in understanding
Solution Approach 1:
The linguistic expressions act as an intermediary that bridges the gap between high-accuracy classification and ease of modification. By providing human-understandable representations of the classification logic, the linguistic expressions enable users to efficiently identify and modify specific aspects of the model's decision-making process without needing to understand or modify the entire complex neural network architecture.
Data Source
AI summary
A computer-implemented method according to one embodiment includes receiving, at a deep neural network (DNN), a plurality of sentences each having an associated label; training the DNN, utilizing the plurality of sentences and associated labels; and producing a linguistic expression (LE) utilizing the trained DNN.