Paying Attention to Facts: Quantifying the Knowledge Capacity of Attention Layers

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autore principale: Wong, Liang Ze
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866916603339210752
author Wong, Liang Ze
author_facet Wong, Liang Ze
contents In this paper, we investigate the ability of single-layer attention-only transformers (i.e. attention layers) to memorize facts contained in databases from a linear-algebraic perspective. We associate with each database a 3-tensor, propose the rank of this tensor as a measure of the size of the database, and provide bounds on the rank in terms of properties of the database. We also define a 3-tensor corresponding to an attention layer, and empirically demonstrate the relationship between its rank and database rank on a dataset of toy models and random databases. By highlighting the roles played by the value-output and query-key weights, and the effects of argmax and softmax on rank, our results shed light on the `additive motif' of factual recall in transformers, while also suggesting a way of increasing layer capacity without increasing the number of parameters.
format Preprint
id arxiv_https___arxiv_org_abs_2502_05076
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Paying Attention to Facts: Quantifying the Knowledge Capacity of Attention Layers
Wong, Liang Ze
Machine Learning
Computation and Language
In this paper, we investigate the ability of single-layer attention-only transformers (i.e. attention layers) to memorize facts contained in databases from a linear-algebraic perspective. We associate with each database a 3-tensor, propose the rank of this tensor as a measure of the size of the database, and provide bounds on the rank in terms of properties of the database. We also define a 3-tensor corresponding to an attention layer, and empirically demonstrate the relationship between its rank and database rank on a dataset of toy models and random databases. By highlighting the roles played by the value-output and query-key weights, and the effects of argmax and softmax on rank, our results shed light on the `additive motif' of factual recall in transformers, while also suggesting a way of increasing layer capacity without increasing the number of parameters.
title Paying Attention to Facts: Quantifying the Knowledge Capacity of Attention Layers
topic Machine Learning
Computation and Language
url https://arxiv.org/abs/2502.05076