CAPSULE NETWORK WITH SHORTCUT ROUTING-BASED LEARNING METHOD AND APPARATUS

In the overall capsule network shortcut routing, image arrangement is transmitted to the entire capsule network. The image arrangement goes through a general convolutional layer and is transformed into (b, n, d, w, h) through a step called a primary capsule to obtain the result. Here, n is the numbe...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: LEE JU HWAN, DANG THANH VU, YU GWANGHYUN, KIM JINYEONG
Format: Patent
Sprache:eng ; kor
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 LEE JU HWAN
DANG THANH VU
YU GWANGHYUN
KIM JINYEONG
description In the overall capsule network shortcut routing, image arrangement is transmitted to the entire capsule network. The image arrangement goes through a general convolutional layer and is transformed into (b, n, d, w, h) through a step called a primary capsule to obtain the result. Here, n is the number of capsule channels, d is the size of a capsule, d is the horizontal and vertical spatial dimension, and b is the arrangement size. Then, the next local capsule layer is obtained using a local capsule layer. A method of obtaining a first global capsule to use a shortcut routing algorithm uses the result of a forward propagation step of the last local capsule block. In order to continuously use the shortcut routing algorithm, a channel wise result in each local capsule block is used as an input of a global capsule block to calculate routing coefficients, and a global capsule is calculated by calculating with the local capsule layer. Accordingly, a capsule network can be built more effectively. 전체적인 캡슐 네트워크 숏컷 라우팅에 있어서, 이미지 배치가 전체 캡슐 네트워크에 전송된다. 이미지 배치는 일반적인 컨벌루션 레이어를 거치고, primary capsule이라고 부르는 단계를 거쳐 (b,n,d,w,h)로 변형한 결과물을 얻게 된다. 여기서 n은 캡슐 채널 수이고, d는 캡슐의 사이즈를 말하며, d는 가로, 세로를 나타내는 공간 차원을 의미하며, b는 배치 사이즈를 말한다. 그 다음에 로컬 캡슐 블락을 이용하여 다음 로컬 캡슐 레이어를 얻는다. 숏컷 라우팅 알고리즘을 사용하기 위한 최초의 글로벌 캡슐을 얻는 방법은 마지막 로컬 캡슐 블락의 순전파 단계의 결과를 사용한다. 그리고 지속적인 숏컷 라우팅 알고리즘을 사용하기 위해서 각각의 로컬 캡슐 블락에서 channel wise 결과를 글로벌 캡슐 블락의 입력으로 라우팅 계수를 산출하여 로컬 캡슐 레이어와 연산하여 글로벌 캡슐을 구하는 과정을 거친다.
format Patent
fullrecord <record><control><sourceid>epo_EVB</sourceid><recordid>TN_cdi_epo_espacenet_KR20220094868A</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>KR20220094868A</sourcerecordid><originalsourceid>FETCH-epo_espacenet_KR20220094868A3</originalsourceid><addsrcrecordid>eNrjZPB0dgwIDvVxVfBzDQn3D_JWCPcM8VAI9vAPCnEODVEI8g8N8fRz13VyDHZ1UfBxdQzyA3IVfF1DPPxdFBz9gDggwDHIMSQ0mIeBNS0xpziVF0pzMyi7uYY4e-imFuTHpxYXJCan5qWWxHsHGRkYGRkYWJpYmFk4GhOnCgAS4S5W</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>patent</recordtype></control><display><type>patent</type><title>CAPSULE NETWORK WITH SHORTCUT ROUTING-BASED LEARNING METHOD AND APPARATUS</title><source>esp@cenet</source><creator>LEE JU HWAN ; DANG THANH VU ; YU GWANGHYUN ; KIM JINYEONG</creator><creatorcontrib>LEE JU HWAN ; DANG THANH VU ; YU GWANGHYUN ; KIM JINYEONG</creatorcontrib><description>In the overall capsule network shortcut routing, image arrangement is transmitted to the entire capsule network. The image arrangement goes through a general convolutional layer and is transformed into (b, n, d, w, h) through a step called a primary capsule to obtain the result. Here, n is the number of capsule channels, d is the size of a capsule, d is the horizontal and vertical spatial dimension, and b is the arrangement size. Then, the next local capsule layer is obtained using a local capsule layer. A method of obtaining a first global capsule to use a shortcut routing algorithm uses the result of a forward propagation step of the last local capsule block. In order to continuously use the shortcut routing algorithm, a channel wise result in each local capsule block is used as an input of a global capsule block to calculate routing coefficients, and a global capsule is calculated by calculating with the local capsule layer. Accordingly, a capsule network can be built more effectively. 전체적인 캡슐 네트워크 숏컷 라우팅에 있어서, 이미지 배치가 전체 캡슐 네트워크에 전송된다. 이미지 배치는 일반적인 컨벌루션 레이어를 거치고, primary capsule이라고 부르는 단계를 거쳐 (b,n,d,w,h)로 변형한 결과물을 얻게 된다. 여기서 n은 캡슐 채널 수이고, d는 캡슐의 사이즈를 말하며, d는 가로, 세로를 나타내는 공간 차원을 의미하며, b는 배치 사이즈를 말한다. 그 다음에 로컬 캡슐 블락을 이용하여 다음 로컬 캡슐 레이어를 얻는다. 숏컷 라우팅 알고리즘을 사용하기 위한 최초의 글로벌 캡슐을 얻는 방법은 마지막 로컬 캡슐 블락의 순전파 단계의 결과를 사용한다. 그리고 지속적인 숏컷 라우팅 알고리즘을 사용하기 위해서 각각의 로컬 캡슐 블락에서 channel wise 결과를 글로벌 캡슐 블락의 입력으로 라우팅 계수를 산출하여 로컬 캡슐 레이어와 연산하여 글로벌 캡슐을 구하는 과정을 거친다.</description><language>eng ; kor</language><subject>CALCULATING ; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS ; COMPUTING ; COUNTING ; PHYSICS</subject><creationdate>2022</creationdate><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><linktohtml>$$Uhttps://worldwide.espacenet.com/publicationDetails/biblio?FT=D&amp;date=20220706&amp;DB=EPODOC&amp;CC=KR&amp;NR=20220094868A$$EHTML$$P50$$Gepo$$Hfree_for_read</linktohtml><link.rule.ids>230,308,780,885,25563,76318</link.rule.ids><linktorsrc>$$Uhttps://worldwide.espacenet.com/publicationDetails/biblio?FT=D&amp;date=20220706&amp;DB=EPODOC&amp;CC=KR&amp;NR=20220094868A$$EView_record_in_European_Patent_Office$$FView_record_in_$$GEuropean_Patent_Office$$Hfree_for_read</linktorsrc></links><search><creatorcontrib>LEE JU HWAN</creatorcontrib><creatorcontrib>DANG THANH VU</creatorcontrib><creatorcontrib>YU GWANGHYUN</creatorcontrib><creatorcontrib>KIM JINYEONG</creatorcontrib><title>CAPSULE NETWORK WITH SHORTCUT ROUTING-BASED LEARNING METHOD AND APPARATUS</title><description>In the overall capsule network shortcut routing, image arrangement is transmitted to the entire capsule network. The image arrangement goes through a general convolutional layer and is transformed into (b, n, d, w, h) through a step called a primary capsule to obtain the result. Here, n is the number of capsule channels, d is the size of a capsule, d is the horizontal and vertical spatial dimension, and b is the arrangement size. Then, the next local capsule layer is obtained using a local capsule layer. A method of obtaining a first global capsule to use a shortcut routing algorithm uses the result of a forward propagation step of the last local capsule block. In order to continuously use the shortcut routing algorithm, a channel wise result in each local capsule block is used as an input of a global capsule block to calculate routing coefficients, and a global capsule is calculated by calculating with the local capsule layer. Accordingly, a capsule network can be built more effectively. 전체적인 캡슐 네트워크 숏컷 라우팅에 있어서, 이미지 배치가 전체 캡슐 네트워크에 전송된다. 이미지 배치는 일반적인 컨벌루션 레이어를 거치고, primary capsule이라고 부르는 단계를 거쳐 (b,n,d,w,h)로 변형한 결과물을 얻게 된다. 여기서 n은 캡슐 채널 수이고, d는 캡슐의 사이즈를 말하며, d는 가로, 세로를 나타내는 공간 차원을 의미하며, b는 배치 사이즈를 말한다. 그 다음에 로컬 캡슐 블락을 이용하여 다음 로컬 캡슐 레이어를 얻는다. 숏컷 라우팅 알고리즘을 사용하기 위한 최초의 글로벌 캡슐을 얻는 방법은 마지막 로컬 캡슐 블락의 순전파 단계의 결과를 사용한다. 그리고 지속적인 숏컷 라우팅 알고리즘을 사용하기 위해서 각각의 로컬 캡슐 블락에서 channel wise 결과를 글로벌 캡슐 블락의 입력으로 라우팅 계수를 산출하여 로컬 캡슐 레이어와 연산하여 글로벌 캡슐을 구하는 과정을 거친다.</description><subject>CALCULATING</subject><subject>COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS</subject><subject>COMPUTING</subject><subject>COUNTING</subject><subject>PHYSICS</subject><fulltext>true</fulltext><rsrctype>patent</rsrctype><creationdate>2022</creationdate><recordtype>patent</recordtype><sourceid>EVB</sourceid><recordid>eNrjZPB0dgwIDvVxVfBzDQn3D_JWCPcM8VAI9vAPCnEODVEI8g8N8fRz13VyDHZ1UfBxdQzyA3IVfF1DPPxdFBz9gDggwDHIMSQ0mIeBNS0xpziVF0pzMyi7uYY4e-imFuTHpxYXJCan5qWWxHsHGRkYGRkYWJpYmFk4GhOnCgAS4S5W</recordid><startdate>20220706</startdate><enddate>20220706</enddate><creator>LEE JU HWAN</creator><creator>DANG THANH VU</creator><creator>YU GWANGHYUN</creator><creator>KIM JINYEONG</creator><scope>EVB</scope></search><sort><creationdate>20220706</creationdate><title>CAPSULE NETWORK WITH SHORTCUT ROUTING-BASED LEARNING METHOD AND APPARATUS</title><author>LEE JU HWAN ; DANG THANH VU ; YU GWANGHYUN ; KIM JINYEONG</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-epo_espacenet_KR20220094868A3</frbrgroupid><rsrctype>patents</rsrctype><prefilter>patents</prefilter><language>eng ; kor</language><creationdate>2022</creationdate><topic>CALCULATING</topic><topic>COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS</topic><topic>COMPUTING</topic><topic>COUNTING</topic><topic>PHYSICS</topic><toplevel>online_resources</toplevel><creatorcontrib>LEE JU HWAN</creatorcontrib><creatorcontrib>DANG THANH VU</creatorcontrib><creatorcontrib>YU GWANGHYUN</creatorcontrib><creatorcontrib>KIM JINYEONG</creatorcontrib><collection>esp@cenet</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>LEE JU HWAN</au><au>DANG THANH VU</au><au>YU GWANGHYUN</au><au>KIM JINYEONG</au><format>patent</format><genre>patent</genre><ristype>GEN</ristype><title>CAPSULE NETWORK WITH SHORTCUT ROUTING-BASED LEARNING METHOD AND APPARATUS</title><date>2022-07-06</date><risdate>2022</risdate><abstract>In the overall capsule network shortcut routing, image arrangement is transmitted to the entire capsule network. The image arrangement goes through a general convolutional layer and is transformed into (b, n, d, w, h) through a step called a primary capsule to obtain the result. Here, n is the number of capsule channels, d is the size of a capsule, d is the horizontal and vertical spatial dimension, and b is the arrangement size. Then, the next local capsule layer is obtained using a local capsule layer. A method of obtaining a first global capsule to use a shortcut routing algorithm uses the result of a forward propagation step of the last local capsule block. In order to continuously use the shortcut routing algorithm, a channel wise result in each local capsule block is used as an input of a global capsule block to calculate routing coefficients, and a global capsule is calculated by calculating with the local capsule layer. Accordingly, a capsule network can be built more effectively. 전체적인 캡슐 네트워크 숏컷 라우팅에 있어서, 이미지 배치가 전체 캡슐 네트워크에 전송된다. 이미지 배치는 일반적인 컨벌루션 레이어를 거치고, primary capsule이라고 부르는 단계를 거쳐 (b,n,d,w,h)로 변형한 결과물을 얻게 된다. 여기서 n은 캡슐 채널 수이고, d는 캡슐의 사이즈를 말하며, d는 가로, 세로를 나타내는 공간 차원을 의미하며, b는 배치 사이즈를 말한다. 그 다음에 로컬 캡슐 블락을 이용하여 다음 로컬 캡슐 레이어를 얻는다. 숏컷 라우팅 알고리즘을 사용하기 위한 최초의 글로벌 캡슐을 얻는 방법은 마지막 로컬 캡슐 블락의 순전파 단계의 결과를 사용한다. 그리고 지속적인 숏컷 라우팅 알고리즘을 사용하기 위해서 각각의 로컬 캡슐 블락에서 channel wise 결과를 글로벌 캡슐 블락의 입력으로 라우팅 계수를 산출하여 로컬 캡슐 레이어와 연산하여 글로벌 캡슐을 구하는 과정을 거친다.</abstract><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier
ispartof
issn
language eng ; kor
recordid cdi_epo_espacenet_KR20220094868A
source esp@cenet
subjects CALCULATING
COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
COMPUTING
COUNTING
PHYSICS
title CAPSULE NETWORK WITH SHORTCUT ROUTING-BASED LEARNING METHOD AND APPARATUS
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-10T10%3A06%3A35IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-epo_EVB&rft_val_fmt=info:ofi/fmt:kev:mtx:patent&rft.genre=patent&rft.au=LEE%20JU%20HWAN&rft.date=2022-07-06&rft_id=info:doi/&rft_dat=%3Cepo_EVB%3EKR20220094868A%3C/epo_EVB%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