Non-standard memory models with indexed retrieval
Abstract
The standard memory models for neural networks are variants of the Hopfield network, where feature representations are stored as vectors in a matrix. Retrieval happens based on similarity between an input vector and the set of stored vectors in a content-addressable manner such that the network evolves towards the closest stored attractor. In other words, the useful property of addressing items in memory directly by index is lost in Hopfield-style neural network models ("associative memory"). In this extended abstract, we present a new model which is extremely simple, derived from biological observation, yet introduces a significant conceptual advance and technical benefits. The goal is to establish adaptivity based on the neuron-centric hypothesis: Plasticity is organized by the neuron which regulates its own synapses. Accordingly we implemented a localist, neuron-centric one-shot learning method and applied it to a simple pattern classification problem (MNIST). We were looking for the existence of high information neurons, to act as indices into the representations. The idea was that we would be able to restore full patterns by indexed retrieval, instead of associative vector retrieval from attractors.
Related
- Theta sequences support real-time spatial memory retrieval
- Retrieval tasks differentially influence memory for object features
- The Semantic Shift in Autobiographical Memory and Aging: Do Retrieval Goals Contribute to Age-Related Memory Differences?
- A long neural circuit spanning multiple brain regions is involved in context-induced retrieval of fear memory
- Neuromodulation-inspired gated associative memory networks: Extended memory retrieval and emergent multistability
BibTeX
@misc{scheler2026nonstandard,
title = {{Non-standard memory models with indexed retrieval}},
author = {Gabriele Scheler and Martin L. Schumann and Johann Schumann},
howpublished = {arXiv},
year = {2026},
month = aug,
doi = {10.48550/arxiv.2608.27479},
eprint = {2608.27479},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2608.27479},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - UNPB
TI - Non-standard memory models with indexed retrieval
AU - Gabriele Scheler
AU - Martin L. Schumann
AU - Johann Schumann
PY - 2026
DA - 2026/08/21
PB - arXiv
DO - 10.48550/arxiv.2608.27479
UR - https://arxiv.org/abs/2608.27479
AB - The standard memory models for neural networks are variants of the Hopfield network, where feature representations are stored as vectors in a matrix. Retrieval happens based on similarity between an input vector and the set of stored vectors in a content-addressable manner such that the network evolves towards the closest stored attractor. In other words, the useful property of addressing items in memory directly by index is lost in Hopfield-style neural network models ("associative memory"). In this extended abstract, we present a new model which is extremely simple, derived from biological observation, yet introduces a significant conceptual advance and technical benefits. The goal is to establish adaptivity based on the neuron-centric hypothesis: Plasticity is organized by the neuron which regulates its own synapses. Accordingly we implemented a localist, neuron-centric one-shot learning method and applied it to a simple pattern classification problem (MNIST). We were looking for the existence of high information neurons, to act as indices into the representations. The idea was that we would be able to restore full patterns by indexed retrieval, instead of associative vector retrieval from attractors.
ER -