Dense Associative Memory with Epanechnikov Energy

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Hoover, Benjamin, Shi, Zhaoyang, Balasubramanian, Krishnakumar, Krotov, Dmitry, Ram, Parikshit
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866908803599958016
author Hoover, Benjamin
Shi, Zhaoyang
Balasubramanian, Krishnakumar
Krotov, Dmitry
Ram, Parikshit
author_facet Hoover, Benjamin
Shi, Zhaoyang
Balasubramanian, Krishnakumar
Krotov, Dmitry
Ram, Parikshit
contents We propose a novel energy function for Dense Associative Memory (DenseAM) networks, the log-sum-ReLU (LSR), inspired by optimal kernel density estimation. Unlike the common log-sum-exponential (LSE) function, LSR is based on the Epanechnikov kernel and enables exact memory retrieval with exponential capacity without requiring exponential separation functions. Moreover, it introduces abundant additional \emph{emergent} local minima while preserving perfect pattern recovery -- a characteristic previously unseen in DenseAM literature. Empirical results show that LSR energy has significantly more local minima (memories) that have comparable log-likelihood to LSE-based models. Analysis of LSR's emergent memories on image datasets reveals a degree of creativity and novelty, hinting at this method's potential for both large-scale memory storage and generative tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2506_10801
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Dense Associative Memory with Epanechnikov Energy
Hoover, Benjamin
Shi, Zhaoyang
Balasubramanian, Krishnakumar
Krotov, Dmitry
Ram, Parikshit
Machine Learning
We propose a novel energy function for Dense Associative Memory (DenseAM) networks, the log-sum-ReLU (LSR), inspired by optimal kernel density estimation. Unlike the common log-sum-exponential (LSE) function, LSR is based on the Epanechnikov kernel and enables exact memory retrieval with exponential capacity without requiring exponential separation functions. Moreover, it introduces abundant additional \emph{emergent} local minima while preserving perfect pattern recovery -- a characteristic previously unseen in DenseAM literature. Empirical results show that LSR energy has significantly more local minima (memories) that have comparable log-likelihood to LSE-based models. Analysis of LSR's emergent memories on image datasets reveals a degree of creativity and novelty, hinting at this method's potential for both large-scale memory storage and generative tasks.
title Dense Associative Memory with Epanechnikov Energy
topic Machine Learning
url https://arxiv.org/abs/2506.10801