[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"tag-catastrophic-forgetting":3},{"tag":4,"articles":11},{"id":5,"name":6,"slug":7,"article_count":8,"description_zh":9,"description_en":10},"19976ff4-6433-42d3-afcc-da60454d663f","catastrophic forgetting","catastrophic-forgetting",3,"災難性遺忘描述模型在持續學習時，學新任務就快速失去舊知識的現象。它影響安全強化學習、長序列 4D 重建、線上適應與多階段訓練，關鍵在於如何保住既有能力又能吸收新資料。","Catastrophic forgetting is the tendency of a model to lose earlier skills or representations when it learns new tasks. It matters in continual learning, safe RL, long-sequence 4D reconstruction, and online adaptation, where retaining past behavior is as important as fitting new data.",[12,21,28,35],{"id":13,"slug":14,"title":15,"summary":16,"category":17,"image_url":18,"cover_image":18,"language":19,"created_at":20},"a4cb421e-464e-4933-9e1c-6371d3cd1503","prevent-catastrophic-forgetting-llm-fine-tuning-zh","如何防止 LLM 微調災難性遺忘","用 Anchored Weight Decay 在 LLM 微調時降低舊任務漂移，保住原有能力並檢查模型是否回復。","research","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780730281470-8i97.png","zh","2026-06-06T07:17:28.426709+00:00",{"id":22,"slug":23,"title":24,"summary":25,"category":17,"image_url":26,"cover_image":26,"language":19,"created_at":27},"923bb0c4-95f3-49a0-8e01-5cdd6bcd2e32","fixing-llm-forgetting-es-fine-tuning-zh","ES 微調忘記問題有解了","這篇論文指出，LLM 用 evolution strategies 微調時的「忘記」多半是可回復的漂移，靠 anchored weight decay 就能壓住。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780604276240-arx4.png","2026-06-04T20:17:25.720929+00:00",{"id":29,"slug":30,"title":31,"summary":32,"category":17,"image_url":33,"cover_image":33,"language":19,"created_at":34},"947e3be0-2b4b-4719-90d1-ddd1ac80f18a","safe-continual-rl-changing-environments-zh","安全持續學習還沒解題","這篇 arXiv 研究把安全 RL 和持續 RL 放在一起看，指出環境一變，現有方法常常顧不了安全，也守不住舊行為。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1776838196623-anqk.png","2026-04-22T06:09:32.609993+00:00",{"id":36,"slug":37,"title":38,"summary":39,"category":17,"image_url":40,"cover_image":40,"language":19,"created_at":41},"7e3fc38d-5744-4f1d-8941-643ed78be513","fast-spatial-memory-elastic-test-time-training-zh","長序列4D重建的彈性記憶法","FSM 用彈性 test-time training 穩住長序列 4D 重建的記憶更新，降低遺忘與記憶瓶頸，讓多 chunk 推論更可行。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775714633904-j3go.png","2026-04-09T06:03:34.127299+00:00"]