Episodic Memory Graph Retrieval With Shared Event Nodes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing episodic memory systems for automated systems face challenges in reducing storage requirements and access time to retrieve matching episodes efficiently.
Innovation Solution
The system employs a collective event database using a hash function for efficient storage and retrieval, shares nodes in the event-sequence graph for common events, and discards similar new episodes to save storage, thereby reducing storage needs and access time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If traditional tree structure is used to store working memory elements, then episodes can be stored, but storage space is wasted and access time is increased
Solution Approach 1:
The patent merges multiple episode storage into a shared graph structure where common events are represented by shared nodes. Instead of storing each episode independently in separate tree structures, the system combines episodes into a single graph where nodes represent events and edges represent temporal relationships. This merging eliminates redundant storage of common events across multiple episodes while maintaining efficient access paths for each episode.
Solution Approach 2:
The patent uses node sharing as a form of copying where a single node representing a common event can be referenced by multiple episodes simultaneously. Rather than creating duplicate copies of event data for each episode, the system creates references from multiple episodes to the same underlying node, reducing storage requirements while allowing each episode to maintain its complete event sequence through the graph structure.
2Productivity
If every episode maintains its own element list, then complete episode information is preserved, but retrieval efficiency decreases when many similar episodes exist
Solution Approach 1:
The patent merges the element lists of multiple episodes into a shared graph structure where common events are represented by shared nodes. This consolidation reduces storage redundancy by eliminating duplicate event representations while maintaining the ability to retrieve complete episode information through the graph's path structure. The merging enables efficient retrieval by allowing the system to traverse shared paths for common events rather than processing redundant data.
Solution Approach 2:
The patent segments episodes into discrete events that can be independently represented as nodes in the graph. By segmenting the continuous episode data into atomic event units, the system can efficiently share common segments (events) across multiple episodes while maintaining the unique sequence and context of each episode. This segmentation enables selective sharing of common portions while preserving episode-specific information.
3Quantity of substance
If tree-search is used to find matching episodes, then episodes can be retrieved, but access time increases with larger numbers of episodes
Solution Approach 1:
The patent performs preliminary organization of episodes into a graph structure during the storage phase, creating an optimized layout that facilitates efficient retrieval. By pre-organizing episodes into a shared graph with common events grouped into shared nodes and edges representing temporal relationships, the system eliminates the need for exhaustive tree searches during retrieval. This preliminary structuring enables faster access by allowing direct traversal along pre-established paths rather than searching through unorganized episode lists.
Solution Approach 2:
The patent transitions from a one-dimensional linear search through episode lists to a multi-dimensional graph structure where episodes are interconnected through shared events. This dimensional change allows the system to navigate through the episode space along multiple paths simultaneously, exploiting the graph's topology to find matching episodes more efficiently. The graph structure adds dimensional relationships between episodes through shared nodes, enabling parallel traversal paths and reducing access time compared to sequential tree-search methods.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Described is a system for an episodic memory used by an automated platform. The system acquires data from an episodic memory that, comprises an event database, an event-sequence graph, and an episode list. Using the event-sequence graph, the system identifies a closest node to a current environment for the automated platform. Based on the closest node and using a hash function or key based on the hash function, the system retrieves from the event database an episode that corresponds to the closest node, the episode including a sequence of even ts, Behavior of the automated platform in the current environment is guided based on the data from the episodic memory.