Adaptive Computation with Elastic Input Sequence

Humans have the ability to adapt the type of information they use, the procedure they employ, and the amount of time they spend when solving problems. However, most standard neural networks have a fixed function type and computation budget regardless of the sample's nature or difficulty. Adapti...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Xue, Fuzhao, Likhosherstov, Valerii, Arnab, Anurag, Houlsby, Neil, Dehghani, Mostafa, You, Yang
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page
container_issue
container_start_page
container_title
container_volume
creator Xue, Fuzhao
Likhosherstov, Valerii
Arnab, Anurag
Houlsby, Neil
Dehghani, Mostafa
You, Yang
description Humans have the ability to adapt the type of information they use, the procedure they employ, and the amount of time they spend when solving problems. However, most standard neural networks have a fixed function type and computation budget regardless of the sample's nature or difficulty. Adaptivity is a powerful paradigm as it not only imbues practitioners with flexibility pertaining to the downstream usage of these models but can also serve as a powerful inductive bias for solving certain challenging classes of problems. In this work, we introduce a new approach called AdaTape, which allows for dynamic computation in neural networks through adaptive tape tokens. AdaTape utilizes an elastic input sequence by equipping an architecture with a dynamic read-and-write tape. Specifically, we adaptively generate input sequences using tape tokens obtained from a tape bank which can be either trainable or derived from input data. We examine the challenges and requirements to obtain dynamic sequence content and length, and propose the Adaptive Tape Reading (ATR) algorithm to achieve both goals. Through extensive experiments on image recognition tasks, we show that AdaTape can achieve better performance while maintaining the computational cost. To facilitate further research, we have released code at https://github.com/google-research/scenic.
doi_str_mv 10.48550/arxiv.2301.13195
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2301_13195</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2301_13195</sourcerecordid><originalsourceid>FETCH-LOGICAL-a675-e10efc395003bae9d69d92c981a1874bb37675a073d2324df97931b1a41bd0bd3</originalsourceid><addsrcrecordid>eNotzs1uwjAQBGBfekC0D8AJv0DCbjbG8RFFFJCQeoB7tI4dYQlCCIbSt-evp5FGo9EnxAghzQulYML9LVzTjABTJDRqIGDmuIvh6mV5PHSXyDEcW_kb4k7O93yOoZar9tHLjT9dfFv7T_HR8P7sv_5zKLbf8225TNY_i1U5Wyc81SrxCL6pySgAsuyNmxpnstoUyFjo3FrSjxmDJpdRlrvGaENokXO0DqyjoRi_b1_kquvDgfu_6kmvXnS6AyfkPZ4</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>Adaptive Computation with Elastic Input Sequence</title><source>arXiv.org</source><creator>Xue, Fuzhao ; Likhosherstov, Valerii ; Arnab, Anurag ; Houlsby, Neil ; Dehghani, Mostafa ; You, Yang</creator><creatorcontrib>Xue, Fuzhao ; Likhosherstov, Valerii ; Arnab, Anurag ; Houlsby, Neil ; Dehghani, Mostafa ; You, Yang</creatorcontrib><description>Humans have the ability to adapt the type of information they use, the procedure they employ, and the amount of time they spend when solving problems. However, most standard neural networks have a fixed function type and computation budget regardless of the sample's nature or difficulty. Adaptivity is a powerful paradigm as it not only imbues practitioners with flexibility pertaining to the downstream usage of these models but can also serve as a powerful inductive bias for solving certain challenging classes of problems. In this work, we introduce a new approach called AdaTape, which allows for dynamic computation in neural networks through adaptive tape tokens. AdaTape utilizes an elastic input sequence by equipping an architecture with a dynamic read-and-write tape. Specifically, we adaptively generate input sequences using tape tokens obtained from a tape bank which can be either trainable or derived from input data. We examine the challenges and requirements to obtain dynamic sequence content and length, and propose the Adaptive Tape Reading (ATR) algorithm to achieve both goals. Through extensive experiments on image recognition tasks, we show that AdaTape can achieve better performance while maintaining the computational cost. To facilitate further research, we have released code at https://github.com/google-research/scenic.</description><identifier>DOI: 10.48550/arxiv.2301.13195</identifier><language>eng</language><subject>Computer Science - Artificial Intelligence ; Computer Science - Computer Vision and Pattern Recognition ; Computer Science - Learning</subject><creationdate>2023-01</creationdate><rights>http://arxiv.org/licenses/nonexclusive-distrib/1.0</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,781,886</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/2301.13195$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2301.13195$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Xue, Fuzhao</creatorcontrib><creatorcontrib>Likhosherstov, Valerii</creatorcontrib><creatorcontrib>Arnab, Anurag</creatorcontrib><creatorcontrib>Houlsby, Neil</creatorcontrib><creatorcontrib>Dehghani, Mostafa</creatorcontrib><creatorcontrib>You, Yang</creatorcontrib><title>Adaptive Computation with Elastic Input Sequence</title><description>Humans have the ability to adapt the type of information they use, the procedure they employ, and the amount of time they spend when solving problems. However, most standard neural networks have a fixed function type and computation budget regardless of the sample's nature or difficulty. Adaptivity is a powerful paradigm as it not only imbues practitioners with flexibility pertaining to the downstream usage of these models but can also serve as a powerful inductive bias for solving certain challenging classes of problems. In this work, we introduce a new approach called AdaTape, which allows for dynamic computation in neural networks through adaptive tape tokens. AdaTape utilizes an elastic input sequence by equipping an architecture with a dynamic read-and-write tape. Specifically, we adaptively generate input sequences using tape tokens obtained from a tape bank which can be either trainable or derived from input data. We examine the challenges and requirements to obtain dynamic sequence content and length, and propose the Adaptive Tape Reading (ATR) algorithm to achieve both goals. Through extensive experiments on image recognition tasks, we show that AdaTape can achieve better performance while maintaining the computational cost. To facilitate further research, we have released code at https://github.com/google-research/scenic.</description><subject>Computer Science - Artificial Intelligence</subject><subject>Computer Science - Computer Vision and Pattern Recognition</subject><subject>Computer Science - Learning</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2023</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNotzs1uwjAQBGBfekC0D8AJv0DCbjbG8RFFFJCQeoB7tI4dYQlCCIbSt-evp5FGo9EnxAghzQulYML9LVzTjABTJDRqIGDmuIvh6mV5PHSXyDEcW_kb4k7O93yOoZar9tHLjT9dfFv7T_HR8P7sv_5zKLbf8225TNY_i1U5Wyc81SrxCL6pySgAsuyNmxpnstoUyFjo3FrSjxmDJpdRlrvGaENokXO0DqyjoRi_b1_kquvDgfu_6kmvXnS6AyfkPZ4</recordid><startdate>20230130</startdate><enddate>20230130</enddate><creator>Xue, Fuzhao</creator><creator>Likhosherstov, Valerii</creator><creator>Arnab, Anurag</creator><creator>Houlsby, Neil</creator><creator>Dehghani, Mostafa</creator><creator>You, Yang</creator><scope>AKY</scope><scope>GOX</scope></search><sort><creationdate>20230130</creationdate><title>Adaptive Computation with Elastic Input Sequence</title><author>Xue, Fuzhao ; Likhosherstov, Valerii ; Arnab, Anurag ; Houlsby, Neil ; Dehghani, Mostafa ; You, Yang</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a675-e10efc395003bae9d69d92c981a1874bb37675a073d2324df97931b1a41bd0bd3</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2023</creationdate><topic>Computer Science - Artificial Intelligence</topic><topic>Computer Science - Computer Vision and Pattern Recognition</topic><topic>Computer Science - Learning</topic><toplevel>online_resources</toplevel><creatorcontrib>Xue, Fuzhao</creatorcontrib><creatorcontrib>Likhosherstov, Valerii</creatorcontrib><creatorcontrib>Arnab, Anurag</creatorcontrib><creatorcontrib>Houlsby, Neil</creatorcontrib><creatorcontrib>Dehghani, Mostafa</creatorcontrib><creatorcontrib>You, Yang</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Xue, Fuzhao</au><au>Likhosherstov, Valerii</au><au>Arnab, Anurag</au><au>Houlsby, Neil</au><au>Dehghani, Mostafa</au><au>You, Yang</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Adaptive Computation with Elastic Input Sequence</atitle><date>2023-01-30</date><risdate>2023</risdate><abstract>Humans have the ability to adapt the type of information they use, the procedure they employ, and the amount of time they spend when solving problems. However, most standard neural networks have a fixed function type and computation budget regardless of the sample's nature or difficulty. Adaptivity is a powerful paradigm as it not only imbues practitioners with flexibility pertaining to the downstream usage of these models but can also serve as a powerful inductive bias for solving certain challenging classes of problems. In this work, we introduce a new approach called AdaTape, which allows for dynamic computation in neural networks through adaptive tape tokens. AdaTape utilizes an elastic input sequence by equipping an architecture with a dynamic read-and-write tape. Specifically, we adaptively generate input sequences using tape tokens obtained from a tape bank which can be either trainable or derived from input data. We examine the challenges and requirements to obtain dynamic sequence content and length, and propose the Adaptive Tape Reading (ATR) algorithm to achieve both goals. Through extensive experiments on image recognition tasks, we show that AdaTape can achieve better performance while maintaining the computational cost. To facilitate further research, we have released code at https://github.com/google-research/scenic.</abstract><doi>10.48550/arxiv.2301.13195</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.2301.13195
ispartof
issn
language eng
recordid cdi_arxiv_primary_2301_13195
source arXiv.org
subjects Computer Science - Artificial Intelligence
Computer Science - Computer Vision and Pattern Recognition
Computer Science - Learning
title Adaptive Computation with Elastic Input Sequence
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2024-12-16T14%3A38%3A40IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-arxiv_GOX&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=Adaptive%20Computation%20with%20Elastic%20Input%20Sequence&rft.au=Xue,%20Fuzhao&rft.date=2023-01-30&rft_id=info:doi/10.48550/arxiv.2301.13195&rft_dat=%3Carxiv_GOX%3E2301_13195%3C/arxiv_GOX%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_id=info:pmid/&rfr_iscdi=true