{"id":475822,"date":"2023-08-09T07:23:51","date_gmt":"2023-08-09T07:23:51","guid":{"rendered":""},"modified":"2023-09-05T11:11:17","modified_gmt":"2023-09-05T11:11:17","slug":"adversarial-machine-learning","status":"publish","type":"wiki","link":"https:\/\/oneproxy.pro\/tr\/wiki\/adversarial-machine-learning\/","title":{"rendered":"\u00c7eli\u015fkili makine \u00f6\u011frenimi"},"content":{"rendered":"

\u00c7eki\u015fmeli makine \u00f6\u011frenimi, yapay zeka ve siber g\u00fcvenli\u011fin kesi\u015fiminde yer alan geli\u015fen bir aland\u0131r. Tasar\u0131m\u0131ndaki g\u00fcvenlik a\u00e7\u0131klar\u0131ndan yararlanarak modelin performans\u0131n\u0131 aldatmaya veya tehlikeye atmaya \u00e7al\u0131\u015fan, makine \u00f6\u011frenimi modellerine y\u00f6nelik d\u00fc\u015fmanca sald\u0131r\u0131lar\u0131 anlamaya ve bunlara kar\u015f\u0131 koymaya odaklan\u0131r. Rekabet\u00e7i makine \u00f6\u011freniminin amac\u0131, bu t\u00fcr sald\u0131r\u0131lara kar\u015f\u0131 savunma sa\u011flayabilecek sa\u011flam ve dayan\u0131kl\u0131 makine \u00f6\u011frenimi sistemleri olu\u015fturmakt\u0131r.<\/p>\n

Adversarial Machine Learning'in k\u00f6keninin tarihi ve bundan ilk s\u00f6z<\/h2>\n

Rekabet\u00e7i makine \u00f6\u011frenimi kavram\u0131n\u0131n k\u00f6keni, ara\u015ft\u0131rmac\u0131lar\u0131n makine \u00f6\u011frenimi algoritmalar\u0131n\u0131n ince girdi manip\u00fclasyonlar\u0131na kar\u015f\u0131 savunmas\u0131zl\u0131\u011f\u0131n\u0131 fark etmeye ba\u015flad\u0131klar\u0131 2000'li y\u0131llar\u0131n ba\u015flar\u0131na kadar uzanabilir. D\u00fc\u015fmanca sald\u0131r\u0131lardan ilk s\u00f6z Szegedy ve arkada\u015flar\u0131n\u0131n \u00e7al\u0131\u015fmalar\u0131na atfedilebilir. 2013 y\u0131l\u0131nda, insan g\u00f6z\u00fcyle alg\u0131lanamayan bir sinir a\u011f\u0131n\u0131 yanl\u0131\u015f y\u00f6nlendirebilecek rahats\u0131z edici girdiler gibi kar\u015f\u0131t \u00f6rneklerin varl\u0131\u011f\u0131n\u0131 g\u00f6sterdiler.<\/p>\n

Adversarial Machine Learning hakk\u0131nda detayl\u0131 bilgi<\/h2>\n

\u00c7eki\u015fmeli makine \u00f6\u011frenimi, \u00e7e\u015fitli d\u00fc\u015fman sald\u0131r\u0131lar\u0131n\u0131 anlamay\u0131 ve bunlara kar\u015f\u0131 savunma mekanizmalar\u0131 tasarlamay\u0131 ama\u00e7layan karma\u015f\u0131k ve \u00e7ok y\u00f6nl\u00fc bir aland\u0131r. Bu alandaki temel zorluk, makine \u00f6\u011frenimi modellerinin rakip girdiler kar\u015f\u0131s\u0131nda do\u011frulu\u011funu ve g\u00fcvenilirli\u011fini korumas\u0131n\u0131 sa\u011flamakt\u0131r.<\/p>\n

Adversarial Machine Learning'in i\u00e7 yap\u0131s\u0131: Nas\u0131l \u00e7al\u0131\u015f\u0131r?<\/h2>\n

\u00c7eki\u015fmeli makine \u00f6\u011frenimi \u00f6z\u00fcnde iki temel bile\u015feni i\u00e7erir: rakip ve savunucu. Rakip, rakip \u00f6rnekler \u00fcretirken, savunan taraf da bu sald\u0131r\u0131lara dayanabilecek sa\u011flam modeller tasarlamaya \u00e7al\u0131\u015f\u0131r. Rekabet\u00e7i makine \u00f6\u011frenimi s\u00fcreci \u015fu \u015fekilde \u00f6zetlenebilir:<\/p>\n

    \n
  1. \n

    \u00c7eli\u015fkili \u00d6rneklerin Olu\u015fturulmas\u0131<\/strong>: D\u00fc\u015fman, hedef makine \u00f6\u011frenimi modelinde yanl\u0131\u015f s\u0131n\u0131fland\u0131rmaya veya di\u011fer istenmeyen davran\u0131\u015flara neden olmak amac\u0131yla giri\u015f verilerine tedirginlikler uygular. Kar\u015f\u0131t \u00f6rnekler olu\u015fturmak i\u00e7in H\u0131zl\u0131 Gradyan \u0130\u015faret Y\u00f6ntemi (FGSM) ve Tahmini Gradyan \u0130ni\u015fi (PGD) gibi \u00e7e\u015fitli teknikler kullan\u0131l\u0131r.<\/p>\n<\/li>\n

  2. \n

    \u00c7eli\u015fkili \u00d6rneklerle E\u011fitim<\/strong>: Savunmac\u0131lar, sa\u011flam bir model olu\u015fturmak i\u00e7in e\u011fitim s\u00fcreci s\u0131ras\u0131nda rakip \u00f6rnekleri dahil eder. Rekabet\u00e7i e\u011fitim olarak bilinen bu s\u00fcre\u00e7, modelin rahats\u0131z edici girdilerle ba\u015fa \u00e7\u0131kmay\u0131 \u00f6\u011frenmesine yard\u0131mc\u0131 olur ve genel sa\u011flaml\u0131\u011f\u0131n\u0131 art\u0131r\u0131r.<\/p>\n<\/li>\n

  3. \n

    De\u011ferlendirme ve Test<\/strong>: Savunmac\u0131, modelin performans\u0131n\u0131, farkl\u0131 sald\u0131r\u0131 t\u00fcrlerine kar\u015f\u0131 dayan\u0131kl\u0131l\u0131\u011f\u0131n\u0131 \u00f6l\u00e7mek i\u00e7in rakip test setleri kullanarak de\u011ferlendirir. Bu ad\u0131m, ara\u015ft\u0131rmac\u0131lar\u0131n modelin g\u00fcvenlik a\u00e7\u0131klar\u0131n\u0131 analiz etmesine ve savunmas\u0131n\u0131 geli\u015ftirmesine olanak tan\u0131r.<\/p>\n<\/li>\n<\/ol>\n

    Adversarial Machine Learning'in temel \u00f6zelliklerinin analizi<\/h2>\n

    Rekabet\u00e7i makine \u00f6\u011freniminin temel \u00f6zellikleri \u015fu \u015fekilde \u00f6zetlenebilir:<\/p>\n

      \n
    1. \n

      \u00c7eli\u015fkili \u00d6rneklerin Varl\u0131\u011f\u0131<\/strong>: \u00c7eki\u015fmeli makine \u00f6\u011frenimi, en son teknolojiye sahip modellerin bile dikkatle haz\u0131rlanm\u0131\u015f rakip \u00f6rneklere kar\u015f\u0131 savunmas\u0131z oldu\u011funu g\u00f6stermi\u015ftir.<\/p>\n<\/li>\n

    2. \n

      Aktar\u0131labilirlik<\/strong>: Bir model i\u00e7in olu\u015fturulan kar\u015f\u0131t \u00f6rnekler, farkl\u0131 mimarilere sahip olsa bile s\u0131kl\u0131kla di\u011fer modellere aktar\u0131l\u0131r ve bu da onu ciddi bir g\u00fcvenlik sorunu haline getirir.<\/p>\n<\/li>\n

    3. \n

      Sa\u011flaml\u0131k ve Do\u011fruluk Dengesi<\/strong>: Modeller rakip sald\u0131r\u0131lara kar\u015f\u0131 daha dayan\u0131kl\u0131 hale getirildik\u00e7e, temiz verilerdeki do\u011fruluklar\u0131 zarar g\u00f6rebilir ve bu da sa\u011flaml\u0131k ile genelleme aras\u0131nda bir denge kurulmas\u0131na yol a\u00e7abilir.<\/p>\n<\/li>\n

    4. \n

      Sald\u0131r\u0131 Geli\u015fmi\u015fli\u011fi<\/strong>: \u00c7eki\u015fmeli sald\u0131r\u0131lar, optimizasyona dayal\u0131 y\u00f6ntemleri, kara kutu sald\u0131r\u0131lar\u0131n\u0131 ve fiziksel d\u00fcnya senaryolar\u0131ndaki sald\u0131r\u0131lar\u0131 i\u00e7erecek \u015fekilde daha karma\u015f\u0131k olacak \u015fekilde geli\u015fti.<\/p>\n<\/li>\n<\/ol>\n

      \u00c7eli\u015fkili Makine \u00d6\u011frenimi T\u00fcrleri<\/h2>\n

      \u00c7eki\u015fmeli makine \u00f6\u011frenimi \u00e7e\u015fitli sald\u0131r\u0131 ve savunma tekniklerini kapsar. Rekabet\u00e7i makine \u00f6\u011freniminin baz\u0131 t\u00fcrleri \u015funlard\u0131r:<\/p>\n

      D\u00fc\u015fmanca Sald\u0131r\u0131lar:<\/h3>\n
        \n
      1. \n

        Beyaz Kutu Sald\u0131r\u0131lar\u0131<\/strong>: Sald\u0131rgan\u0131n modelin mimarisine ve parametrelerine tam eri\u015fimi vard\u0131r.<\/p>\n<\/li>\n

      2. \n

        Kara Kutu Sald\u0131r\u0131lar\u0131<\/strong>: Sald\u0131rgan\u0131n hedef modele eri\u015fimi s\u0131n\u0131rl\u0131d\u0131r veya hi\u00e7 yoktur ve kar\u015f\u0131t \u00f6rnekler olu\u015fturmak i\u00e7in yedek modeller kullanabilir.<\/p>\n<\/li>\n

      3. \n

        Transfer Sald\u0131r\u0131lar\u0131<\/strong>: Bir model i\u00e7in olu\u015fturulan kar\u015f\u0131t \u00f6rnekler ba\u015fka bir modele sald\u0131rmak i\u00e7in kullan\u0131l\u0131r.<\/p>\n<\/li>\n

      4. \n

        Fiziksel D\u00fcnya Sald\u0131r\u0131lar\u0131<\/strong>: Otonom ara\u00e7lar\u0131 yan\u0131ltmaya y\u00f6nelik g\u00f6r\u00fcnt\u00fc bozulmalar\u0131 gibi ger\u00e7ek d\u00fcnya senaryolar\u0131nda etkili olacak \u015fekilde tasarlanm\u0131\u015f kar\u015f\u0131t \u00f6rnekler.<\/p>\n<\/li>\n<\/ol>\n

        D\u00fc\u015fman Savunmalar\u0131:<\/h3>\n
          \n
        1. \n

          Rekabet\u00e7i E\u011fitim<\/strong>: Sa\u011flaml\u0131\u011f\u0131 art\u0131rmak i\u00e7in model e\u011fitimi s\u0131ras\u0131nda kar\u015f\u0131t \u00f6rneklerin dahil edilmesi.<\/p>\n<\/li>\n

        2. \n

          Savunma Dam\u0131tma<\/strong>: \u00c7\u0131kt\u0131 da\u011f\u0131l\u0131mlar\u0131n\u0131 s\u0131k\u0131\u015ft\u0131rarak d\u00fc\u015fman sald\u0131r\u0131lar\u0131na direnecek e\u011fitim modelleri.<\/p>\n<\/li>\n

        3. \n

          Sertifikal\u0131 Savunmalar<\/strong>: S\u0131n\u0131rl\u0131 bozulmalara kar\u015f\u0131 sa\u011flaml\u0131\u011f\u0131 garanti etmek i\u00e7in do\u011frulanm\u0131\u015f s\u0131n\u0131rlar\u0131n kullan\u0131lmas\u0131.<\/p>\n<\/li>\n

        4. \n

          Giri\u015f \u00d6n \u0130\u015fleme<\/strong>: Potansiyel olumsuz tedirginlikleri ortadan kald\u0131rmak i\u00e7in giri\u015f verilerinin de\u011fi\u015ftirilmesi.<\/p>\n<\/li>\n<\/ol>\n

          Adversarial Machine Learning'i kullanma yollar\u0131, sorunlar ve kullan\u0131mla ilgili \u00e7\u00f6z\u00fcmleri<\/h2>\n

          \u00c7eli\u015fkili makine \u00f6\u011frenimi, bilgisayarl\u0131 g\u00f6rme, do\u011fal dil i\u015fleme ve siber g\u00fcvenlik dahil olmak \u00fczere \u00e7e\u015fitli alanlarda uygulama alan\u0131 bulur. Ancak rakip makine \u00f6\u011freniminin kullan\u0131m\u0131 ayn\u0131 zamanda zorluklar\u0131 da beraberinde getirir:<\/p>\n

            \n
          1. \n

            Rakiplere Kar\u015f\u0131 Sa\u011flaml\u0131k<\/strong>: Modeller, mevcut savunmalar\u0131 a\u015fabilecek yeni ve uyarlanabilir sald\u0131r\u0131lara kar\u015f\u0131 h\u00e2l\u00e2 savunmas\u0131z kalabilir.<\/p>\n<\/li>\n

          2. \n

            Hesaplamal\u0131 Ek Y\u00fck<\/strong>: \u00c7eli\u015fkili e\u011fitim ve savunma mekanizmalar\u0131, model e\u011fitimi ve \u00e7\u0131kar\u0131m i\u00e7in hesaplama gereksinimlerini art\u0131rabilir.<\/p>\n<\/li>\n

          3. \n

            Veri kalitesi<\/strong>: Kar\u015f\u0131t \u00f6rnekler, tespit edilmesi zor olabilecek ve potansiyel veri kalitesi sorunlar\u0131na yol a\u00e7abilecek k\u00fc\u00e7\u00fck bozulmalara dayan\u0131r.<\/p>\n<\/li>\n<\/ol>\n

            Bu zorluklar\u0131n \u00fcstesinden gelmek i\u00e7in devam eden ara\u015ft\u0131rmalar, daha verimli savunma mekanizmalar\u0131 geli\u015ftirmeye, transfer \u00f6\u011freniminden yararlanmaya ve rakip makine \u00f6\u011freniminin teorik temellerini ke\u015ffetmeye odaklan\u0131yor.<\/p>\n

            Ana \u00f6zellikler ve benzer terimlerle kar\u015f\u0131la\u015ft\u0131rmalar<\/h2>\n\n\n\n\n\n\n\n\n
            Terim<\/th>\nTan\u0131m<\/th>\n<\/tr>\n<\/thead>\n
            \u00c7eli\u015fkili Makine \u00d6\u011frenimi<\/td>\nMakine \u00f6\u011frenimi modellerine y\u00f6nelik sald\u0131r\u0131lar\u0131 anlamaya ve bunlara kar\u015f\u0131 savunmaya odaklan\u0131r.<\/td>\n<\/tr>\n
            Siber g\u00fcvenlik<\/td>\nBilgisayar sistemlerini sald\u0131r\u0131lardan ve tehditlerden korumaya y\u00f6nelik teknolojileri ve uygulamalar\u0131 kapsar.<\/td>\n<\/tr>\n
            Makine \u00f6\u011frenme<\/td>\nBilgisayarlar\u0131n verilerden \u00f6\u011frenmesini sa\u011flayan algoritmalar\u0131 ve istatistiksel modelleri i\u00e7erir.<\/td>\n<\/tr>\n
            Yapay Zeka (AI)<\/td>\n\u0130nsan benzeri g\u00f6revleri yerine getirebilen ve ak\u0131l y\u00fcr\u00fctebilen ak\u0131ll\u0131 makineler yaratman\u0131n daha geni\u015f alan\u0131.<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

            \u00c7eli\u015fkili Makine \u00d6\u011frenimi ile ilgili gelece\u011fin perspektifleri ve teknolojileri<\/h2>\n

            Rekabet\u00e7i makine \u00f6\u011freniminin gelece\u011fi, hem sald\u0131r\u0131 hem de savunma tekniklerinde umut verici geli\u015fmeler bar\u0131nd\u0131r\u0131yor. Baz\u0131 perspektifler \u015funlar\u0131 i\u00e7erir:<\/p>\n

              \n
            1. \n

              \u00dcretken Rekabet\u00e7i A\u011flar (GAN'lar)<\/strong>: G\u00fcvenlik a\u00e7\u0131klar\u0131n\u0131 anlamak ve savunmalar\u0131 geli\u015ftirmek amac\u0131yla rakip \u00f6rnekler olu\u015fturmak i\u00e7in GAN'lar\u0131n kullan\u0131lmas\u0131.<\/p>\n<\/li>\n

            2. \n

              A\u00e7\u0131klanabilir Yapay Zeka<\/strong>: Rakiplere y\u00f6nelik g\u00fcvenlik a\u00e7\u0131klar\u0131n\u0131 daha iyi anlamak i\u00e7in yorumlanabilir modeller geli\u015ftirmek.<\/p>\n<\/li>\n

            3. \n

              Hizmet Olarak Rekabet\u00e7i Sa\u011flaml\u0131k (ARAaS)<\/strong>: \u0130\u015fletmelerin yapay zeka modellerini g\u00fcvence alt\u0131na almalar\u0131 i\u00e7in bulut tabanl\u0131 sa\u011flaml\u0131k \u00e7\u00f6z\u00fcmleri sa\u011flamak.<\/p>\n<\/li>\n<\/ol>\n

              Proxy sunucular\u0131 nas\u0131l kullan\u0131labilir veya Adversarial Machine Learning ile nas\u0131l ili\u015fkilendirilebilir?<\/h2>\n

              Proxy sunucular\u0131 internet kullan\u0131c\u0131lar\u0131n\u0131n g\u00fcvenli\u011fini ve gizlili\u011fini artt\u0131rmada \u00e7ok \u00f6nemli bir rol oynamaktad\u0131r. Kullan\u0131c\u0131n\u0131n IP adresini gizleyerek istekleri ve yan\u0131tlar\u0131 ileterek kullan\u0131c\u0131lar ve internet aras\u0131nda arac\u0131 g\u00f6revi g\u00f6r\u00fcrler. Proxy sunucular\u0131, \u00e7eki\u015fmeli makine \u00f6\u011frenimi ile a\u015fa\u011f\u0131daki yollarla ili\u015fkilendirilebilir:<\/p>\n

                \n
              1. \n

                ML Altyap\u0131s\u0131n\u0131 Korumak<\/strong>: Proxy sunucular\u0131, makine \u00f6\u011frenimi altyap\u0131s\u0131n\u0131 do\u011frudan sald\u0131r\u0131lara ve yetkisiz eri\u015fim giri\u015fimlerine kar\u015f\u0131 koruyabilir.<\/p>\n<\/li>\n

              2. \n

                D\u00fc\u015fmanca Sald\u0131r\u0131lara Kar\u015f\u0131 Savunma<\/strong>: Proxy sunucular\u0131, k\u00f6t\u00fc ama\u00e7l\u0131 istekleri makine \u00f6\u011frenimi modeline ula\u015fmadan \u00f6nce filtreleyerek, potansiyel rakip faaliyetlere kar\u015f\u0131 gelen trafi\u011fi analiz edebilir.<\/p>\n<\/li>\n

              3. \n

                Gizlilik korumas\u0131<\/strong>: Proxy sunucular\u0131, verilerin ve kullan\u0131c\u0131 bilgilerinin anonimle\u015ftirilmesine yard\u0131mc\u0131 olarak olas\u0131 veri zehirlenmesi sald\u0131r\u0131lar\u0131 riskini azaltabilir.<\/p>\n<\/li>\n<\/ol>\n

                \u0130lgili Ba\u011flant\u0131lar<\/h2>\n

                Adversarial Machine Learning hakk\u0131nda daha fazla bilgi i\u00e7in a\u015fa\u011f\u0131daki kaynaklar\u0131 ke\u015ffedebilirsiniz:<\/p>\n

                  \n
                1. OpenAI Blogu \u2013 Tart\u0131\u015fmal\u0131 \u00d6rnekler<\/a><\/li>\n
                2. Google AI Blogu \u2013 Kar\u015f\u0131t \u00d6rnekleri A\u00e7\u0131klamak ve Kullanmak<\/a><\/li>\n
                3. MIT Teknoloji \u0130ncelemesi \u2013 Yapay Zeka Dedektifleri<\/a><\/li>\n<\/ol>","protected":false},"featured_media":0,"menu_order":0,"template":"","meta":{"_acf_changed":false,"content-type":"","inline_featured_image":false,"footnotes":""},"class_list":["post-475822","wiki","type-wiki","status-publish","hentry"],"acf":{"faq_title":"Frequently Asked Questions about Adversarial Machine Learning: Enhancing Proxy Server Security<\/mark>","faq_items":[{"question":"What is Adversarial Machine Learning?","answer":"

                  Adversarial Machine Learning is a field that focuses on understanding and countering adversarial attacks on machine learning models. It aims to build robust and resilient AI systems that can defend against attempts to deceive or compromise their performance.<\/p>"},{"question":"How did Adversarial Machine Learning originate?","answer":"

                  The concept of Adversarial Machine Learning emerged in the early 2000s when researchers noticed vulnerabilities in machine learning algorithms. The first mention of adversarial attacks can be traced back to the work of Szegedy et al. in 2013, where they demonstrated the existence of adversarial examples.<\/p>"},{"question":"How does Adversarial Machine Learning work?","answer":"

                  Adversarial Machine Learning involves two key components: the adversary and the defender. The adversary crafts adversarial examples, while the defender designs robust models to withstand these attacks. Adversarial examples are perturbed inputs that aim to mislead the target machine learning model.<\/p>"},{"question":"What are the key features of Adversarial Machine Learning?","answer":"

                  The key features of Adversarial Machine Learning include the existence of adversarial examples, their transferability between models, and the trade-off between robustness and accuracy. Additionally, adversaries use sophisticated attacks, such as white-box, black-box, transfer, and physical-world attacks.<\/p>"},{"question":"What types of Adversarial Machine Learning attacks exist?","answer":"

                  Adversarial attacks come in various forms:<\/p>

                  • White-box Attacks: The attacker has complete access to the model's architecture and parameters.<\/li>
                  • Black-box Attacks: The attacker has limited access to the target model and may use substitute models.<\/li>
                  • Transfer Attacks: Adversarial examples generated for one model are used to attack another model.<\/li>
                  • Physical-world Attacks: Adversarial examples designed to work in real-world scenarios, such as fooling autonomous vehicles.<\/li><\/ul>"},{"question":"How can Adversarial Machine Learning be used?","answer":"

                    Adversarial Machine Learning finds applications in computer vision, natural language processing, and cybersecurity. It helps enhance the security of AI models and protects against potential threats posed by adversarial attacks.<\/p>"},{"question":"What are the challenges in using Adversarial Machine Learning?","answer":"

                    Some challenges include ensuring robustness against novel attacks, dealing with computational overhead, and maintaining data quality when handling adversarial examples.<\/p>"},{"question":"How does Adversarial Machine Learning compare to other terms?","answer":"

                    Adversarial Machine Learning is related to cybersecurity, machine learning, and artificial intelligence (AI), but it specifically focuses on defending machine learning models against adversarial attacks.<\/p>"},{"question":"What does the future hold for Adversarial Machine Learning?","answer":"

                    The future of Adversarial Machine Learning includes advancements in attack and defense techniques, leveraging GANs, developing interpretable models, and providing robustness as a service.<\/p>"},{"question":"How are proxy servers associated with Adversarial Machine Learning?","answer":"

                    Proxy servers play a vital role in enhancing security by protecting ML infrastructure, defending against adversarial attacks, and safeguarding user privacy and data. They act as intermediaries, filtering out potential malicious traffic before it reaches the machine learning model.<\/p>"}]},"_links":{"self":[{"href":"https:\/\/oneproxy.pro\/tr\/wp-json\/wp\/v2\/wiki\/475822","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/oneproxy.pro\/tr\/wp-json\/wp\/v2\/wiki"}],"about":[{"href":"https:\/\/oneproxy.pro\/tr\/wp-json\/wp\/v2\/types\/wiki"}],"version-history":[{"count":0,"href":"https:\/\/oneproxy.pro\/tr\/wp-json\/wp\/v2\/wiki\/475822\/revisions"}],"wp:attachment":[{"href":"https:\/\/oneproxy.pro\/tr\/wp-json\/wp\/v2\/media?parent=475822"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}