{"id":1726,"date":"2026-05-12T08:58:45","date_gmt":"2026-05-12T08:58:45","guid":{"rendered":"https:\/\/hria.utcluj.ro\/?p=1726"},"modified":"2026-05-18T20:04:27","modified_gmt":"2026-05-18T20:04:27","slug":"arhitecturi-agentice-si-modele-de-deep-learning-in-cadrul-romanian-ai-hospital","status":"publish","type":"post","link":"https:\/\/hria.utcluj.ro\/en\/arhitecturi-agentice-si-modele-de-deep-learning-in-cadrul-romanian-ai-hospital\/","title":{"rendered":"Romanian AI Hospital – arhitecturi agentice \u0219i modele de deep learning"},"content":{"rendered":"

Agentic AI: Video Segmentation for Polyp Detection (Gastro Clinic)<\/strong><\/h2>\n\n\n\n

Implementarea vizeaz\u0103 optimizarea detec\u021biei polipilor prin segmentare video \u00een timp real, utiliz\u00e2nd un agent colonoscopic bazat pe arhitecturi SOTA (State-of-the-Art)<\/sup>.<\/p>\n\n\n\n

Arhitectur\u0103 \u0219i Metodologie<\/strong><\/h3>\n\n\n\n
    \n
  • Backbone:<\/strong> Se utilizeaz\u0103 ConvNeXt<\/strong>, o arhitectur\u0103 bazat\u0103 pe CNN (ResNet) dar \u00eembun\u0103t\u0103\u021bit\u0103 cu principii specifice Transformerelor.<\/li>\n\n\n\n
  • Feature Extraction:<\/strong> Modelul este structurat \u00een 4 etape; ultimele dou\u0103 furnizeaz\u0103 tr\u0103s\u0103turi de nivel sc\u0103zut (H\/8 x W\/8) \u0219i de nivel \u00eenalt (H\/16 x W\/16).<\/li>\n\n\n\n
  • Inovarea KAN:<\/strong> Arhitectura a fost \u00eembun\u0103t\u0103\u021bit\u0103 prin ad\u0103ugarea a 3 straturi Kolmogorov-Arnold (KAN)<\/strong>.<\/li>\n\n\n\n
  • Configura\u021bie straturi KAN:<\/strong> Un strat KAN este definit ca o matrice de func\u021bii de activare \u0219i utilizeaz\u0103 convolu\u021bii de tip depth-wise<\/em>.<\/li>\n\n\n\n
  • Antrenare:<\/strong> Greut\u0103\u021bile pentru backbone-ul ConvNeXt au fost ob\u021binute prin pre-antrenare pe setul de date ImageNet-22K<\/strong>.<\/li>\n<\/ul>\n\n\n
    \n
    \"\"<\/figure>\n<\/div>\n\n\n

    Performan\u021b\u0103 \u0219i Validare<\/strong><\/h3>\n\n\n\n
      \n
    • Dataset:<\/strong> Evaluarea s-a realizat pe SUN-SEG<\/strong>, cel mai mare set de date de secven\u021be colonoscopice complet adnotate, utiliz\u00e2nd 19.544 de imagini ce con\u021bin polipi \u0219i m\u0103\u0219tile aferente.<\/li>\n\n\n\n
    • Metrici de referin\u021b\u0103 (Dice SOTA):<\/strong> Modelele bazate pe Cross-Attention au ca referin\u021b\u0103 PNS+ (73.7%), SALI (82.2%) \u0219i YOLO-SAM2 (90.2%).<\/li>\n\n\n\n
    • Rezultate Model Propus:<\/strong> Dep\u0103\u0219e\u0219te PNS+ pe fiecare metric\u0103 pentru ambele grade de dificultate ale setului de date.<\/li>\n\n\n\n
    • Eficien\u021b\u0103 computa\u021bional\u0103:<\/strong> Modelul men\u021bine o vitez\u0103 de procesare de peste 10 FPS, ating\u00e2nd o medie de 12.91 FPS<\/strong>.<\/li>\n<\/ul>\n\n\n\n
      \n\n\n\n

      Automatic Invoice Management & Resident Support System<\/strong><\/h2>\n\n\n\n

      Sistemul integreaz\u0103 procesarea documentelor administrative \u0219i suportul academic prin metode neuro-simbolice \u0219i ontologii<\/sup><\/sup>.<\/p>\n\n\n\n

      Gestiunea Automat\u0103 a Facturilor<\/strong><\/h3>\n\n\n\n
        \n
      • Ingestie de date:<\/strong> Facturile sunt desc\u0103rcate automat din SPV (Spa\u021biul Privat Virtual) \u0219i procesate \u00eentr-un knowledge graph aliniat cu ontologia sistemului.<\/li>\n\n\n\n
      • Capabilit\u0103\u021bi Ontologice:<\/strong> Integrarea ontologiilor pentru produse \u0219i facturi permite sugestii de actualizare bazate pe con\u021binutul facturat.<\/li>\n\n\n\n
      • Automatizare GRN:<\/strong> Generarea automat\u0103 a Notelor de Recep\u021bie \u0219i Constatare Diferen\u021be (Goods Receipt Note) pe baza istoricului \u0219i a caracteristicilor produselor din ontologie.<\/li>\n<\/ul>\n\n\n\n

        Sistemul de Suport pentru Reziden\u021bi<\/strong><\/h3>\n\n\n\n
          \n
        • Abordare Neuro-Simbolic\u0103:<\/strong> Clasificarea fi\u0219ierelor \u0219i adnotarea acestora folose\u0219te ontologia ca baz\u0103, \u00een timp ce clasificarea propriu-zis\u0103 utilizeaz\u0103 o abordare neural\u0103.<\/li>\n\n\n\n
        • Interogare:<\/strong> Func\u021bia de c\u0103utare este implementat\u0103 prin interog\u0103ri SPARQL<\/strong> peste ontologia sistemului.<\/li>\n\n\n\n
        • Status Dezvoltare:<\/strong> Se analizeaz\u0103 complexitatea ontologiei \u0219i selectarea motorului de ra\u021bionament adecvat (GraphDB, Stardog sau solu\u021bii mai expresive).<\/li>\n<\/ul>\n\n\n\n
          \n\n\n\n

          Feedback-Driven Agentic LLM for Autonomous Research<\/strong><\/h2>\n\n\n\n

          Arhitectura multi-agent este conceput\u0103 pentru fluxuri de lucru academice structurate \u0219i cercetare autonom\u0103<\/sup><\/sup><\/sup><\/sup>.<\/p>\n\n\n\n

          Fluxul de Lucru (Pipeline)<\/strong><\/h3>\n\n\n\n

          Sistemul utilizeaz\u0103 contracte JSON<\/strong> pentru comunicarea \u00eentre agen\u021bi \u0219i include bucle de revizie de tip editorial<\/sup><\/sup><\/sup>:<\/p>\n\n\n\n

            \n
          1. Researcher:<\/strong> Prime\u0219te task-ul de cercetare \u0219i colecteaz\u0103 datele.<\/li>\n\n\n\n
          2. Writer:<\/strong> Elaboreaz\u0103 con\u021binutul pe baza cercet\u0103rii.<\/li>\n\n\n\n
          3. Fact Checker:<\/strong> Realizeaz\u0103 validarea afirma\u021biilor \u0219i asigur\u0103 trasabilitatea dovezilor (claim-level evidence traceability).<\/li>\n\n\n\n
          4. Editor:<\/strong> Refineaz\u0103 output-ul final; poate solicita reluarea procesului (redo<\/em>) dac\u0103 criteriile de calitate nu sunt \u00eendeplinite.<\/li>\n<\/ol>\n\n\n
            \n
            \"\"<\/figure>\n<\/div>\n\n\n

            Obiective Tehnice<\/strong><\/h3>\n\n\n\n
              \n
            • Ra\u021bionament Reproductibil:<\/strong> Documentarea procesului de ra\u021bionament pentru a asigura reproductibilitatea.<\/li>\n\n\n\n
            • Output Final:<\/strong> Generarea unui rezumat de o pagin\u0103, incluz\u00e2nd lista de referin\u021be \u0219i cit\u0103rile corespunz\u0103toare pentru subiectul analizat.<\/li>\n<\/ul>\n\n\n\n

              <\/p>","protected":false},"excerpt":{"rendered":"

              Agentic AI: Video Segmentation for Polyp Detection (Gastro Clinic) Implementarea vizeaz\u0103 optimizarea detec\u021biei polipilor prin segmentare video \u00een timp real, utiliz\u00e2nd un agent colonoscopic bazat pe arhitecturi SOTA (State-of-the-Art). Arhitectur\u0103 \u0219i Metodologie Performan\u021b\u0103 \u0219i Validare Automatic Invoice Management & Resident Support System Sistemul integreaz\u0103 procesarea documentelor administrative \u0219i suportul academic prin metode neuro-simbolice \u0219i ontologii.… <\/p>","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"neve_meta_sidebar":"","neve_meta_container":"default","neve_meta_enable_content_width":"on","neve_meta_content_width":100,"neve_meta_title_alignment":"","neve_meta_author_avatar":"off","neve_post_elements_order":"[\"content\"]","neve_meta_disable_header":"off","neve_meta_disable_footer":"","neve_meta_disable_title":"","_themeisle_gutenberg_block_has_review":false,"footnotes":""},"categories":[7],"tags":[],"class_list":["post-1726","post","type-post","status-publish","format-standard","hentry","category-rezultate"],"_links":{"self":[{"href":"https:\/\/hria.utcluj.ro\/en\/wp-json\/wp\/v2\/posts\/1726","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/hria.utcluj.ro\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/hria.utcluj.ro\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/hria.utcluj.ro\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/hria.utcluj.ro\/en\/wp-json\/wp\/v2\/comments?post=1726"}],"version-history":[{"count":4,"href":"https:\/\/hria.utcluj.ro\/en\/wp-json\/wp\/v2\/posts\/1726\/revisions"}],"predecessor-version":[{"id":1844,"href":"https:\/\/hria.utcluj.ro\/en\/wp-json\/wp\/v2\/posts\/1726\/revisions\/1844"}],"wp:attachment":[{"href":"https:\/\/hria.utcluj.ro\/en\/wp-json\/wp\/v2\/media?parent=1726"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/hria.utcluj.ro\/en\/wp-json\/wp\/v2\/categories?post=1726"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/hria.utcluj.ro\/en\/wp-json\/wp\/v2\/tags?post=1726"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}