Saved in:
Bibliographic Details
Main Authors: Yorsh, Uladzislau, Holeňa, Martin, Bojar, Ondřej, Herel, David
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2404.00798
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929298899730432
author Yorsh, Uladzislau
Holeňa, Martin
Bojar, Ondřej
Herel, David
author_facet Yorsh, Uladzislau
Holeňa, Martin
Bojar, Ondřej
Herel, David
contents Transformers have revolutionized deep learning in numerous fields, including natural language processing, computer vision, and audio processing. Their strength lies in their attention mechanism, which allows for the discovering of complex input relationships. However, this mechanism's quadratic time and memory complexity pose challenges for larger inputs. Researchers are now investigating models like Linear Unified Nested Attention (Luna) or Memory Augmented Transformer, which leverage external learnable memory to either reduce the attention computation complexity down to linear, or to propagate information between chunks in chunk-wise processing. Our findings challenge the conventional thinking on these models, revealing that interfacing with the memory directly through an attention operation is suboptimal, and that the performance may be considerably improved by filtering the input signal before communicating with memory.
format Preprint
id arxiv_https___arxiv_org_abs_2404_00798
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle On Difficulties of Attention Factorization through Shared Memory
Yorsh, Uladzislau
Holeňa, Martin
Bojar, Ondřej
Herel, David
Machine Learning
Transformers have revolutionized deep learning in numerous fields, including natural language processing, computer vision, and audio processing. Their strength lies in their attention mechanism, which allows for the discovering of complex input relationships. However, this mechanism's quadratic time and memory complexity pose challenges for larger inputs. Researchers are now investigating models like Linear Unified Nested Attention (Luna) or Memory Augmented Transformer, which leverage external learnable memory to either reduce the attention computation complexity down to linear, or to propagate information between chunks in chunk-wise processing. Our findings challenge the conventional thinking on these models, revealing that interfacing with the memory directly through an attention operation is suboptimal, and that the performance may be considerably improved by filtering the input signal before communicating with memory.
title On Difficulties of Attention Factorization through Shared Memory
topic Machine Learning
url https://arxiv.org/abs/2404.00798