Related Experiment Video
Updated: Aug 12, 2025

An Operant Intra-/Extra-dimensional Set-shift Task for Mice
Published on: January 22, 2016
On separating long- and short-term memories in hyperdimensional computing
Jeffrey L Teeters1, Denis Kleyko1,2, Pentti Kanerva1
1Redwood Center for Theoretical Neuroscience, University of California, Berkeley, Berkeley, CA, United States.
Associative memory offers a more efficient method for storing numerous key-value pairs compared to superposition vectors, requiring less memory and computation for reliable data retrieval. This approach enhances memory capacity without increasing vector width.
Area of Science:
- Computational Neuroscience
- Machine Learning
- Information Theory
Background:
- High-dimensional, fixed-width vectors enable information distribution via superposition, encoding key-value pairs through component-wise multiplication and addition.
- Superposition vectors act as approximate memories, with exact retrieval from a codebook (item memory) requiring vectors of the same width.
- Increasing memory capacity necessitates wider vectors, posing scalability challenges for superposition-based storage.
Purpose of the Study:
- To investigate the efficiency of associative memory versus superposition vectors for storing a large number of key-value pairs.
- To demonstrate that associative memory offers advantages in memory and computation for reliable storage.
- To explore the potential of associative memory to augment superposition vector-based systems.
Main Methods:
- Comparing the memory and computational requirements of associative memory and superposition vector approaches for storing over 1,000 key-value pairs.
- Experimentally validating the advantages of associative memory using the storage of large finite-state automata.
- Analyzing the reliability of storage and retrieval for both memory paradigms.
Main Results:
- Associative memory requires less memory and computation than superposition vectors for storing a large number of key-value pairs with comparable reliability.
- The number of storage locations in associative memory can be increased without increasing vector width, overcoming a key limitation of superposition vectors.
- Experimental results with finite-state automata confirm the practical advantages of associative memory.
Conclusions:
- Associative memory presents a more scalable and efficient solution for large-scale key-value pair storage compared to superposition vectors.
- Associative memory can augment superposition vector systems, analogous to the relationship between human long-term and working memory.
- The findings suggest potential applications in modeling brain-like memory and state-dependent behaviors.
Related Concept Videos
Storage
System of Memory
Long-Term Memory
Long-term memory can be categorized into two primary types: explicit and implicit memory. Explicit memory, also known as declarative memory, involves the conscious recollection of information that we deliberately try to remember, recall, and articulate. This type of memory encompasses specific facts, events, and...
Chunking and Rehearsal in Sensory Memory
Working Memory
Higher Mental Functions of Brain: Learning and Memory

