[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"glossary-articles-diffusion-model-zh-TW":3},[4,20,30,40],{"id":5,"slug":6,"title":7,"summary":8,"source":9,"category":10,"image_url":11,"cover_image":11,"language":12,"views":13,"created_at":14,"published_at":15,"topic_cluster_id":16,"topic_clusters":17},"b630264c-6adf-4808-8c75-2b887020e0d9","diffusiongemma-transparency-measured-zh","DiffusionGemma 的透明度問題被量化了","這篇研究把擴散式語言模型的透明度拆成兩部分，並證明 DiffusionGemma 可透過 token bottleneck 變得明顯更可解釋。","arxiv.org","research","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781848974850-kk3o.png","zh",3,"2026-06-19T06:02:30.127489+00:00","2026-06-19T06:02:30.113+00:00","0c35a120-52fc-41fc-afa3-d404eb934158",{"slug":18,"title":19},"longmemeval-v2-agent-","LongMemEval-V2：測 agent 長期記憶",{"id":21,"slug":22,"title":23,"summary":24,"source":9,"category":10,"image_url":25,"cover_image":25,"language":12,"views":26,"created_at":27,"published_at":28,"topic_cluster_id":16,"topic_clusters":29},"b1ac2a1c-7e6e-48b8-a776-5d9d8d126787","klip-localized-shift-detection-inverse-problems-zh","KLIP 用擴散先驗抓局部異常","KLIP 把擴散先驗和後驗的 KL 散度拿來看逆問題中的 OOD 變化，還能把異常定位到局部區域。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780294694039-s7go.png",6,"2026-06-01T06:17:40.554648+00:00","2026-06-01T06:17:40.531+00:00",{"slug":18,"title":19},{"id":31,"slug":32,"title":33,"summary":34,"source":9,"category":10,"image_url":35,"cover_image":35,"language":12,"views":36,"created_at":37,"published_at":38,"topic_cluster_id":16,"topic_clusters":39},"8a7df89c-afa4-44f5-992b-a32618239019","carv-cuts-diffusion-teacher-gradient-variance-zh","CARV 讓 diffusion 老師梯度更穩","CARV 用分層蒙地卡羅、重要性取樣與重用昂貴前處理，降低 diffusion-teacher 管線的梯度方差與計算浪費。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779343563009-evcu.png",5,"2026-05-21T06:05:29.409637+00:00","2026-05-21T06:05:29.187+00:00",{"slug":18,"title":19},{"id":41,"slug":42,"title":43,"summary":44,"source":9,"category":10,"image_url":45,"cover_image":45,"language":12,"views":46,"created_at":47,"published_at":48,"topic_cluster_id":49,"topic_clusters":49},"be28a180-07a1-433c-bb6a-6f015d7291c2","actcam-joint-camera-motion-control-zh","ActCam 讓鏡頭和動作一起控","ActCam 用零樣本方式，同時控制影片生成中的角色動作與攝影機路徑，重點是不必重新訓練新模型。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778220675487-z4sz.png",11,"2026-05-08T06:10:33.672754+00:00","2026-05-08T06:10:33.494+00:00",null]