NLU Lookup Source Framework for Efficient Data Protection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Modern natural language understanding (NLU) systems face challenges in efficiently leveraging structured information from various data sources to enhance domain-specific operations, while also needing to minimize computational resources and protect sensitive data.
Innovation Solution
A NLU framework that includes a lookup source framework, enabling the definition of a lookup source system with multiple lookup sources. Each lookup source has a compact source data representation, optimized during compilation, and supports efficient matching of user utterances during inference, with capabilities for exact and fuzzy matching, and data protection techniques like encryption and obfuscation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If source data is stored in traditional formats for NLU processing, then data accessibility is maintained, but computational resource usage increases and data protection becomes difficult
Solution Approach 1:
The system performs preliminary action by compiling source data into optimized representations during a compilation phase before actual NLU inference occurs. This pre-processing transforms raw source data into a format that reduces computational load during runtime while maintaining data accessibility for the NLU framework.
Solution Approach 2:
The invention introduces an intermediary representation layer between the original source data and the NLU processing components. This intermediate form serves as a mediator that both reduces computational requirements and enables data protection mechanisms to operate on the optimized representation rather than the raw data during inference.
2Productivity
If source data is transformed during compile-time operation, then computational efficiency during inference is improved, but system complexity increases
Solution Approach 1:
The system segments the data processing workflow into distinct phases: a compilation phase where source data is transformed into optimized representations, and an inference phase where the optimized data is consumed by NLU components. This segmentation allows the complex transformation logic to be isolated in the compilation phase, simplifying the inference phase and improving overall productivity.
Solution Approach 2:
By performing the complex data transformation as a preliminary action during compilation rather than during inference, the system prepares optimized data representations in advance. This eliminates the need for complex real-time transformations during inference, thereby improving inference speed while containing system complexity in the compilation phase.
3Measurement precision
If exact matching is used for source data, then matching precision is high, but language flexibility is reduced
Solution Approach 1:
The system applies different matching qualities to different parts of the data representation. The optimized source data representation maintains precise structural information for exact matching where needed, while allowing for flexible pattern matching in other dimensions. This local differentiation of matching quality enables both high precision and language flexibility simultaneously.
Solution Approach 2:
The invention creates a composite matching approach by combining multiple matching strategies within the optimized data representation. The representation structure allows both exact matching operations and fuzzy pattern matching operations to work together, providing both high precision for exact matches and flexibility for linguistic variations.
4Ease of operation
If PII is stored in memory for processing, then data accessibility is maintained, but security risk increases
Solution Approach 1:
The system performs preliminary action by protecting sensitive data during the compilation phase before it enters the inference phase. The optimized representation is created in a way that minimizes the time PII remains in unprotected memory form, reducing the window of vulnerability while maintaining accessibility during the necessary processing operations.
Solution Approach 2:
The system rushes through the processing of PII by using optimized representations that enable faster processing. The compilation phase creates a streamlined data structure that allows the NLU system to process and complete operations on PII more quickly, reducing the total time the data is exposed in memory and thereby reducing security risk.
Data Source
AI summary
A natural language understanding (NLU) framework includes a lookup source framework, which enables a lookup source system to be defined having one or more lookup sources. Each lookup source of the lookup source system includes a respective source data representation that is compiled from respective source data. For example, a source data representation may include source data arranged in a finite state transducer (IFST) structure as a set of finite-state automata (FSA) states, wherein each state is associated with a token that represents underlying source data. Different producers can be applied during compilation of a source data representation to derive additional states within the source data representation from the source data. Certain states of the source data representation that contain sensitive data can be selectively protected through encryption and/or obfuscation, while other portions of the source data representation that are not sensitive may remain in clear-text form.