Embedding Object Memory for Semantic Retrieval at Lower Memory Cost
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computing systems face inefficiencies in storing and retrieving large quantities of content information due to reliance on keyword searches that do not consider abstract meanings, leading to inaccuracies and high memory requirements.
Innovation Solution
Implementing semantic embedding models to generate embeddings that are stored in an embedding object memory, allowing for semantic proximity matching and efficient retrieval of content objects using geometric distance functions, with mechanisms for fine-grained indexing and hierarchical data structures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If keyword searches are used to iterate through content objects, then the search process is simple to implement, but the search accuracy is relatively inaccurate because it does not consider abstract meaning
Solution Approach 1:
The patent introduces semantic embeddings as an intermediary representation between content objects and search queries. Instead of directly matching keywords with content, the system transforms both queries and content into semantic embedding vectors, enabling accurate semantic similarity matching while maintaining a relatively simple overall system architecture
Solution Approach 2:
The patent changes the search parameter from keyword matching to semantic embedding similarity. By transforming content objects into vector representations that capture abstract meaning, the system achieves higher search accuracy through parameter transformation rather than fundamental architectural complexity
2Quantity of substance
If large quantities of content information are stored for searching, then more comprehensive search results are available, but the memory requirements become relatively large
Solution Approach 1:
The patent extracts only the essential semantic information from content objects by generating embeddings, while discarding redundant raw data. This extraction approach allows the system to store compact vector representations instead of full content objects, significantly reducing memory requirements while maintaining search comprehensiveness
Solution Approach 2:
The patent creates a compressed copy of content information in the form of semantic embeddings. These embedding vectors serve as efficient representations that capture the essential meaning of content objects, allowing extensive content storage with minimal memory footprint compared to storing original content
3Productivity
If content information is stored and retrieved using traditional methods, then the system is easy to implement, but the computational efficiency is relatively low
Solution Approach 1:
The patent performs preliminary action by pre-computing and storing semantic embeddings for all content objects before retrieval operations. This allows the system to quickly compare query embeddings against pre-processed content embeddings using efficient similarity metrics, significantly improving retrieval speed and computational efficiency
Data Source
AI summary
Methods, systems, and media for storing entries in and/or retrieving information from an embedding object memory are provided. In some examples, a content item is received that has content data. The content data associated with the content item may be provided to one or more semantic embedding models that generate semantic embeddings. From one or more of the semantic embedding models, one or semantic embeddings may be received. The one or more semantic embeddings may then be inserted into the embedding object memory. The semantic embeddings may be associated with respective indications corresponding to a reference to source data associated with the semantic embeddings. Further, the insertion may trigger a spatial storage operation to store a vector representation of the one or more semantic embeddings. A plurality of collections of stored embeddings may be received from the embedding object memory, based on a provided input, to determine an action.