{"id":13556,"date":"2024-08-14T08:42:47","date_gmt":"2024-08-14T05:42:47","guid":{"rendered":"https:\/\/umram.bilkent.edu.tr\/index.php\/2024\/08\/14\/vispool-vektor-gorunurluk-cizgesi-sinir-aglariyla-donusturucu-kodlayicilarin-guclendirilmesi\/"},"modified":"2026-11-04T10:57:19","modified_gmt":"2026-11-04T07:57:19","slug":"vispool-vektor-gorunurluk-cizgesi-sinir-aglariyla-donusturucu-kodlayicilarin-guclendirilmesi","status":"publish","type":"post","link":"https:\/\/umram.bilkent.edu.tr\/index.php\/tr\/2024\/08\/14\/vispool-vektor-gorunurluk-cizgesi-sinir-aglariyla-donusturucu-kodlayicilarin-guclendirilmesi\/","title":{"rendered":"VISPool: Vekt\u00f6r g\u00f6r\u00fcn\u00fcrl\u00fck \u00e7izgesi sinir a\u011flar\u0131yla d\u00f6n\u00fc\u015ft\u00fcr\u00fcc\u00fc kodlay\u0131c\u0131lar\u0131n g\u00fc\u00e7lendirilmesi"},"content":{"rendered":"<p>\u201cVISPool: Enhancing Transformer Encoders with Vector Visibility Graph Neural Networks\u201d ba\u015fl\u0131kl\u0131 \u00e7al\u0131\u015fmam\u0131z Bangkok&#8217;ta d\u00fczenlenen ACL 2024 konferans\u0131nda sunulmu\u015ftur.<\/p>\n<p>\u00c7al\u0131\u015fmada, do\u011fal dil i\u015flemenin s\u0131n\u0131rlar\u0131n\u0131 geni\u015fletmek i\u00e7in d\u00f6n\u00fc\u015ft\u00fcr\u00fcc\u00fcler ile \u00e7izge sinir a\u011flar\u0131n\u0131n (GNN) birle\u015fimi incelenmektedir. D\u00f6n\u00fc\u015ft\u00fcr\u00fcc\u00fcler do\u011fal dil i\u015flemeyi k\u00f6kl\u00fc bi\u00e7imde de\u011fi\u015ftirmi\u015ftir. GNN&#8217;lerin bu yap\u0131lara kat\u0131lmas\u0131 umut verici bir yol sunsa da mevcut y\u00f6ntemler \u00e7\u0131kar\u0131msal (transd\u00fcktif) yakla\u015f\u0131m\u0131n s\u0131n\u0131rlar\u0131, \u00f6l\u00e7eklenebilirlik ve karma\u015f\u0131k metin i\u015fleme gereksinimi gibi sorunlarla kar\u015f\u0131la\u015fmaktad\u0131r.<\/p>\n<p>Bu sorunlar\u0131 gidermek i\u00e7in, d\u00f6n\u00fc\u015ft\u00fcr\u00fcc\u00fc \u00e7\u0131kt\u0131lar\u0131ndan \u00fcretilen vekt\u00f6r g\u00f6r\u00fcn\u00fcrl\u00fck \u00e7izgelerini (VVG) kullanan, metin belgeleri i\u00e7in yeni bir dinamik \u00e7izge olu\u015fturma y\u00f6ntemi \u00f6nerilmektedir. \u00d6nerilen g\u00f6r\u00fcn\u00fcrl\u00fck havuzlay\u0131c\u0131s\u0131 (VISPool), VVG evri\u015fim a\u011flar\u0131n\u0131 d\u00f6n\u00fc\u015ft\u00fcr\u00fcc\u00fc i\u015flem zincirlerine do\u011frudan katmaktad\u0131r. VISPool, daha az e\u011fitilebilir parametreyle GLUE veri setlerinde kar\u015f\u0131la\u015ft\u0131r\u0131lan y\u00f6ntemlerden daha ba\u015far\u0131l\u0131 olmaktad\u0131r.<\/p>\n\n<p>Makale: <a href=\"https:\/\/aclanthology.org\/2024.findings-acl.149\" target=\"_blank\" rel=\"noopener\">https:\/\/aclanthology.org\/2024.findings-acl.149<\/a><\/p>\n<p>Kod: <a href=\"https:\/\/github.com\/koc-lab\/vispool\" target=\"_blank\" rel=\"noopener\">https:\/\/github.com\/koc-lab\/vispool<\/a><\/p>\n\n<p><strong>\u00d6zet:<\/strong><\/p>\n<p>D\u00f6n\u00fc\u015ft\u00fcr\u00fcc\u00fclerin ortaya \u00e7\u0131k\u0131\u015f\u0131, \u00e7e\u015fitli g\u00f6revlerde g\u00f6r\u00fcld\u00fc\u011f\u00fc gibi do\u011fal dil i\u015flemeyi k\u00f6kl\u00fc bi\u00e7imde de\u011fi\u015ftirmi\u015ftir. \u00c7izge sinir a\u011flar\u0131 (GNN) do\u011fal dil i\u015flemede umut vaat etse de d\u00f6n\u00fc\u015ft\u00fcr\u00fcc\u00fclerin yerini tek ba\u015f\u0131na alamaz. Bu nedenle son \u00e7al\u0131\u015fmalar d\u00f6n\u00fc\u015ft\u00fcr\u00fcc\u00fcler ile GNN&#8217;lerin birle\u015ftirilmesini ara\u015ft\u0131rmaktad\u0131r. Mevcut GNN tabanl\u0131 yakla\u015f\u0131mlar a\u015f\u0131r\u0131 metin i\u015fleme gerektiren dura\u011fan \u00e7izge olu\u015fturma y\u00f6ntemlerine dayanmakta ve \u00e7o\u011fu belge ve s\u00f6zc\u00fck say\u0131s\u0131 artt\u0131k\u00e7a \u00f6l\u00e7eklenememektedir. Bu s\u0131n\u0131rlamalar, d\u00f6n\u00fc\u015ft\u00fcr\u00fcc\u00fc \u00e7\u0131kt\u0131s\u0131ndan \u00fcretilen vekt\u00f6r g\u00f6r\u00fcn\u00fcrl\u00fck \u00e7izgelerine (VVG) dayanan, metin belgeleri i\u00e7in yeni bir dinamik \u00e7izge olu\u015fturma y\u00f6ntemiyle giderilmektedir. Ard\u0131ndan VVG evri\u015fim a\u011flar\u0131n\u0131 d\u00f6n\u00fc\u015ft\u00fcr\u00fcc\u00fc i\u015flem zincirlerine do\u011frudan katan, \u00f6l\u00e7eklenebilir bir model mimarisi olan g\u00f6r\u00fcn\u00fcrl\u00fck havuzlay\u0131c\u0131s\u0131 (VISPool) tan\u0131t\u0131lmaktad\u0131r. \u00d6nerilen model, Genel Dil Anlama De\u011ferlendirmesi (GLUE) veri setlerinde de\u011ferlendirilmi\u015ftir. VISPool, daha az e\u011fitilebilir parametreyle kar\u015f\u0131la\u015ft\u0131r\u0131lan y\u00f6ntemlerden daha ba\u015far\u0131l\u0131 olmakta ve g\u00f6r\u00fcn\u00fcrl\u00fc\u011fe dayal\u0131 \u00e7izge olu\u015fturman\u0131n d\u00f6n\u00fc\u015ft\u00fcr\u00fcc\u00fcleri GNN&#8217;lerle g\u00fc\u00e7lendirmede uygulanabilir oldu\u011funu g\u00f6stermektedir.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u201cVISPool: Enhancing Transformer Encoders with Vector Visibility Graph Neural Networks\u201d ba\u015fl\u0131kl\u0131 \u00e7al\u0131\u015fmam\u0131z Bangkok&#8217;ta d\u00fczenlenen ACL 2024 konferans\u0131nda sunulmu\u015ftur. \u00c7al\u0131\u015fmada, do\u011fal dil i\u015flemenin s\u0131n\u0131rlar\u0131n\u0131 geni\u015fletmek i\u00e7in d\u00f6n\u00fc\u015ft\u00fcr\u00fcc\u00fcler ile \u00e7izge sinir a\u011flar\u0131n\u0131n (GNN) birle\u015fimi incelenmektedir. D\u00f6n\u00fc\u015ft\u00fcr\u00fcc\u00fcler do\u011fal dil i\u015flemeyi k\u00f6kl\u00fc bi\u00e7imde de\u011fi\u015ftirmi\u015ftir. GNN&#8217;lerin bu yap\u0131lara kat\u0131lmas\u0131 umut verici bir yol sunsa da mevcut y\u00f6ntemler \u00e7\u0131kar\u0131msal (transd\u00fcktif) yakla\u015f\u0131m\u0131n\u2026<\/p>\n","protected":false},"author":9,"featured_media":12401,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[445],"tags":[],"class_list":["post-13556","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-duyurular"],"_links":{"self":[{"href":"https:\/\/umram.bilkent.edu.tr\/index.php\/wp-json\/wp\/v2\/posts\/13556","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/umram.bilkent.edu.tr\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/umram.bilkent.edu.tr\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/umram.bilkent.edu.tr\/index.php\/wp-json\/wp\/v2\/users\/9"}],"replies":[{"embeddable":true,"href":"https:\/\/umram.bilkent.edu.tr\/index.php\/wp-json\/wp\/v2\/comments?post=13556"}],"version-history":[{"count":1,"href":"https:\/\/umram.bilkent.edu.tr\/index.php\/wp-json\/wp\/v2\/posts\/13556\/revisions"}],"predecessor-version":[{"id":13557,"href":"https:\/\/umram.bilkent.edu.tr\/index.php\/wp-json\/wp\/v2\/posts\/13556\/revisions\/13557"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/umram.bilkent.edu.tr\/index.php\/wp-json\/wp\/v2\/media\/12401"}],"wp:attachment":[{"href":"https:\/\/umram.bilkent.edu.tr\/index.php\/wp-json\/wp\/v2\/media?parent=13556"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/umram.bilkent.edu.tr\/index.php\/wp-json\/wp\/v2\/categories?post=13556"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/umram.bilkent.edu.tr\/index.php\/wp-json\/wp\/v2\/tags?post=13556"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}