Distributed Associative Memory for Linear Scaling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Associative memories face challenges in scaling effectively to address real-world problems due to geometric scaling issues, which limit their application in complex systems, and existing technologies struggle with spurious memories and inefficient data handling.
Innovation Solution
A distributed associative memory system that includes a network of networks with a processing system configured to observe and imagine associations using multiple streaming queues, translating semantic-space queries into physical-space queries, and utilizing a query map to estimate taxonomic meanings and expand query terms, while providing count data and reverse look-up capabilities to associate results with query terms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional associative memories are used to handle pattern matching and identification, then pattern recognition capability is achieved, but geometric scaling limitations prevent effective handling of real-world complex problems
Solution Approach 1:
The system segments the associative memory into multiple independent networks, each handling specific patterns or features. This segmentation allows the overall system to scale linearly by adding modular network units rather than geometrically increasing complexity, while maintaining comprehensive pattern recognition capabilities across distributed memory structures.
Solution Approach 2:
The patent introduces a distributed network dimension, transitioning from a single associative memory unit to a multi-network architecture. This dimensional expansion enables the system to handle real-world problem complexity by distributing computations across multiple networks, achieving linear scalability through additive network layers rather than geometric growth within a single unit.
2Adaptability or versatility
If the number of inputs in associative memory is increased to solve real-world problems, then problem-solving capability improves, but geometric scaling causes unreasonable system growth
Solution Approach 1:
Instead of concentrating all inputs in a single associative memory unit, the system segments inputs across multiple distributed networks. Each network receives and processes a subset of inputs independently, allowing the system to handle large numbers of real-world inputs through linear addition of network units rather than geometric explosion within one unit.
Solution Approach 2:
The patent distributes inputs across a multi-network dimensional space, where each network layer handles a portion of the input data. This dimensional distribution transforms the scaling challenge from geometric growth in a single unit to linear growth across multiple units, enabling effective handling of real-world input volumes.
3Reliability
If traditional associative memory structures are used, then basic memory functions are provided, but spurious memories and inefficient data handling occur
Solution Approach 1:
The distributed network architecture segments memory storage and retrieval operations across multiple independent networks. This segmentation isolates spurious memory formation to specific network subsets, preventing contamination across the entire system and enabling more reliable data handling through distributed verification and error isolation.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Associative memory systems, methods and computer program products are provided. An associative memory system includes a distributed associative memory base including a network of networks of associative memory networks. A respective associative memory network includes associations among a respective observer memories and a plurality of observed memories that are observed by the respective observer memory. Ones of the associative memory networks are physically and/or logically independent from other ones of the associative memory networks. A processing system is configured to observe associations into and imagine associations from, the distributed associative memory base using multiple streaming queues that correspond to respective ones of multiple rows in the associative memory networks. The processing system is further configured to determine a cognitive distance between a term and a class of terms, the cognitive distance being returned responsive to a query of the distributed associative memory base.