Saved in:
Bibliographic Details
Main Authors: Muller, Lyle, Churchland, Patricia S., Sejnowski, Terrence J.
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2401.14267
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909289286729728
author Muller, Lyle
Churchland, Patricia S.
Sejnowski, Terrence J.
author_facet Muller, Lyle
Churchland, Patricia S.
Sejnowski, Terrence J.
contents The capabilities of transformer networks such as ChatGPT and other Large Language Models (LLMs) have captured the world's attention. The crucial computational mechanism underlying their performance relies on transforming a complete input sequence - for example, all the words in a sentence - into a long "encoding vector" that allows transformers to learn long-range temporal dependencies in naturalistic sequences. Specifically, "self-attention" applied to this encoding vector enhances temporal context in transformers by computing associations between pairs of words in the input sequence. We suggest that waves of neural activity traveling across single cortical areas or multiple regions at the whole-brain scale could implement a similar encoding principle. By encapsulating recent input history into a single spatial pattern at each moment in time, cortical waves may enable temporal context to be extracted from sequences of sensory inputs, the same computational principle used in transformers.
format Preprint
id arxiv_https___arxiv_org_abs_2401_14267
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Transformers and Cortical Waves: Encoders for Pulling In Context Across Time
Muller, Lyle
Churchland, Patricia S.
Sejnowski, Terrence J.
Computation and Language
Artificial Intelligence
The capabilities of transformer networks such as ChatGPT and other Large Language Models (LLMs) have captured the world's attention. The crucial computational mechanism underlying their performance relies on transforming a complete input sequence - for example, all the words in a sentence - into a long "encoding vector" that allows transformers to learn long-range temporal dependencies in naturalistic sequences. Specifically, "self-attention" applied to this encoding vector enhances temporal context in transformers by computing associations between pairs of words in the input sequence. We suggest that waves of neural activity traveling across single cortical areas or multiple regions at the whole-brain scale could implement a similar encoding principle. By encapsulating recent input history into a single spatial pattern at each moment in time, cortical waves may enable temporal context to be extracted from sequences of sensory inputs, the same computational principle used in transformers.
title Transformers and Cortical Waves: Encoders for Pulling In Context Across Time
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2401.14267