Prime-Number Token Encoding for Parallel GPU Knowledge Graphs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Multilayer networks face challenges in efficiently processing large volumes of data due to increased system costs and computational demands, particularly when handling knowledge graphs with numerous tokens, leading to organizations being unable to utilize their full potential.
Innovation Solution
A method involving a parallel data processing circuit that assigns unique contiguous prime numbers to tokens in a knowledge graph, encoding both tokens and their relationships, allowing for parallel execution of set operations without separately encoding edges.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional encoding methods are used for tokens in knowledge graphs, then the system can process the data, but the computational overhead increases and processing efficiency decreases
Solution Approach 1:
The patent changes the encoding parameter from traditional sequential or hierarchical encoding to prime number-based encoding. Each token is assigned a unique prime number, and set relationships are represented by mathematical operations on these prime numbers. This parameter change enables parallel processing of set operations while reducing the computational complexity of encoding and querying operations.
Solution Approach 2:
The patent replaces the mechanical system of traditional graph traversal and set operations with a mathematical system based on prime number properties. Instead of mechanically traversing graph edges to determine set relationships, the system uses mathematical operations (multiplication, division, modulo) on prime number encodings to efficiently determine relationships, enabling parallel computation.
2Reliability
If the number of tokens in the knowledge graph increases to provide more relevant information, then the quality of multilayer network output improves, but the system cost and hardware resource requirements increase
Solution Approach 1:
The patent changes the parameter representation from traditional graph structures to prime number encodings, which compress the representation of set relationships. This parameter change allows the system to handle larger numbers of tokens and more complex relationships without proportionally increasing hardware resource requirements, as the prime number encoding efficiently represents multiple relationships simultaneously through mathematical properties.
3Loss of information
If separate encoding is performed for tokens and edges in knowledge graphs, then the relationships are clearly represented, but the encoding complexity and processing time increase
Solution Approach 1:
The patent merges the encoding of tokens and their set relationships into a unified prime number-based system. Instead of separately encoding tokens and edges, the system assigns prime numbers to tokens and uses mathematical operations on these encodings to represent relationships. This merging eliminates the need for separate encoding processes while preserving complete relationship information through the mathematical properties of prime numbers.
Data Source
AI summary
An apparatus and method for efficiently encoding tokens of a knowledge graph. In various implementations, a computing system includes a includes a processing circuit and a memory. The memory stores the instructions of an application that relies on a data model such as a large language model (LLM) to process natural language processing (NLP) tasks that analyze and extract meaning and relationships from text provided by a user's input. The processing circuit receives a full tokens list of a knowledge graph and receives set relationships for the full tokens list. When executing the application, the processing circuit assigns unique contiguous prime numbers to the tokens of the full tokens list and generates encoded values for the tokens based at least on the assigned prime numbers. The processing circuit provides the encoded full tokens list to the data model.