{"topic":{"id":4584,"title":"粤港澳大湾区算法比赛 1000 万奖金咱组个队吧","slug":"topic","categoryId":6,"createdAt":"2023-07-28 12:27:46.926102","bumpedAt":"2023-07-30 03:07:30.367871","views":496,"postsCount":4,"highestPostNumber":4,"likeCount":1,"closed":false,"pinned":false,"pinnedGlobally":false,"pinnedAt":null,"pinnedUntil":null,"isBanner":false,"excerpt":"这个比赛奖金丰厚，求组队人员，最好是组里有显卡的同学，https://iacc.pazhoulab-huangpu.com/contestdetail?id=64af505d4a0ed647faca627e&award=1,000,000","deletedAt":null,"isAnonymous":false,"category":{"id":6,"name":"寻🐱/寻物/表白/交友","slug":"people","color":"F1592A"},"author":{"id":964,"username":"aikedaer","name":"aikedaer","avatarUrl":"/uploads/image/fa/fa/fafa2a0e6422ac9c1d7b17984a92dc9112d8352aa3f942ab99b35e356c47bf22.avif"},"tags":["competition"]},"posts":[{"id":8168,"postNumber":1,"raw":"这个比赛奖金丰厚，求组队人员，最好是组里有显卡的同学，https://iacc.pazhoulab-huangpu.com/contestdetail?id=64af505d4a0ed647faca627e&award=1,000,000\n![image|502x319](/uploads/image/e7/78/e778b2916c0dd64a57a95ad84343a02928f171635e2ef2857999026ff1c3714f.avif)","cooked":"

这个比赛奖金丰厚，求组队人员，最好是组里有显卡的同学，2024第三届大赛赛题一览-琶洲实验室 (黄埔）
\n $\"image\"$

\n","createdAt":"2023-07-28 12:27:47.203899","updatedAt":"2023-09-09 08:57:00.350598","replyToPostNumber":null,"replyCount":0,"likeCount":0,"wordCount":28,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":3,"publicVersion":3,"deletedAt":null,"isAnonymous":false,"user":{"id":964,"username":"aikedaer","name":"aikedaer","avatarUrl":"/uploads/image/fa/fa/fafa2a0e6422ac9c1d7b17984a92dc9112d8352aa3f942ab99b35e356c47bf22.avif","admin":false,"moderator":false,"signatureRaw":null,"signatureCooked":null}},{"id":8174,"postNumber":2,"raw":"> 背景\n\n 当前深度学习的应用通常只关注在单一任务上的表现。为实现通用人工智能的目标，AI 模型需要具有在序列任务上不断学习新知识的能力。在该场景下，学习任务按照指定顺序依次到达，并且出于隐私保护和数据安全等目的，历史任务的训练数据不能被保存。\n 深度学习模型虽然可以在当前任务上学习到新知识，但存在“灾难性遗忘”现象，即容易遗忘在历史任务上学到的知识，从而不能保持在历史任务上的性能。\n 灾难性遗忘现象使得深度学习模型在序列任务上的整体表现欠佳。与之相比，人类在学习新知识的同时，往往能很好地保留旧知识。\n\n> 意义\n\n 研究持续学习算法，解决灾难性遗忘现象是实现通用人工智能的关键途径。\n 针对动态学习环境下的灾难性遗忘现象，开发高效且有新意的持续学习算法，提升深度学习模型在序列任务上的整体表现。\n\n\n这个问题属于 Multi-task / Meta Learning / life-long learning\n\n[cs330_lifelonglearning_2022.pdf|attachment](/uploads/pdf/8f/5b/8f5be50912ece3185cab21efd551b025cc49b6da09dda5e30fb2d6abdd8d6939.pdf) (5.0 MB)","cooked":"

\n
背景
\n

当前深度学习的应用通常只关注在单一任务上的表现。为实现通用人工智能的目标，AI 模型需要具有在序列任务上不断学习新知识的能力。在该场景下，学习任务按照指定顺序依次到达，并且出于隐私保护和数据安全等目的，历史任务的训练数据不能被保存。\n深度学习模型虽然可以在当前任务上学习到新知识，但存在“灾难性遗忘”现象，即容易遗忘在历史任务上学到的知识，从而不能保持在历史任务上的性能。\n灾难性遗忘现象使得深度学习模型在序列任务上的整体表现欠佳。与之相比，人类在学习新知识的同时，往往能很好地保留旧知识。\n

\n
意义
\n

研究持续学习算法，解决灾难性遗忘现象是实现通用人工智能的关键途径。\n针对动态学习环境下的灾难性遗忘现象，开发高效且有新意的持续学习算法，提升深度学习模型在序列任务上的整体表现。\n

这个问题属于 Multi-task / Meta Learning / life-long learning

cs330_lifelonglearning_2022.pdf|attachment (5.0 MB)

\n","createdAt":"2023-07-28 12:45:10.857244","updatedAt":"2024-03-17 13:05:02.258389","replyToPostNumber":null,"replyCount":1,"likeCount":0,"wordCount":349,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":3,"publicVersion":3,"deletedAt":null,"isAnonymous":false,"user":{"id":1047,"username":"anonymous_coward_old","name":"Anonymous Coward Old","avatarUrl":"/uploads/image/43/32/4332638c11248b8ff07033c11de0759a55205facfbec85e14458245a91991faf.avif","admin":false,"moderator":false,"signatureRaw":null,"signatureCooked":null}},{"id":8211,"postNumber":3,"raw":"是的，还涉及增量学习，开放世界目标检测等","cooked":"

是的，还涉及增量学习，开放世界目标检测等

\n","createdAt":"2023-07-29 02:05:59.782316","updatedAt":"2023-07-29 02:05:59.782316","replyToPostNumber":2,"replyCount":1,"likeCount":1,"wordCount":20,"liked":false,"bookmarked":false,"polls":[],"reactions":[{"reaction":"❤️","count":1,"users":[{"id":259,"username":"ReturnDream","name":"保密"}],"hasReacted":false}],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":964,"username":"aikedaer","name":"aikedaer","avatarUrl":"/uploads/image/fa/fa/fafa2a0e6422ac9c1d7b17984a92dc9112d8352aa3f942ab99b35e356c47bf22.avif","admin":false,"moderator":false,"signatureRaw":null,"signatureCooked":null}},{"id":8252,"postNumber":4,"raw":"# 深度神经网络中的持续学习 [ICML2021 Tutorial]\nhttps://zhuanlan.zhihu.com/p/438766442","cooked":"

深度神经网络中的持续学习 [ICML2021 Tutorial]

https://zhuanlan.zhihu.com/p/438766442

","createdAt":"2023-07-30 03:07:30.367871","updatedAt":"2023-07-30 03:07:30.367871","replyToPostNumber":3,"replyCount":0,"likeCount":0,"wordCount":16,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":964,"username":"aikedaer","name":"aikedaer","avatarUrl":"/uploads/image/fa/fa/fafa2a0e6422ac9c1d7b17984a92dc9112d8352aa3f942ab99b35e356c47bf22.avif","admin":false,"moderator":false,"signatureRaw":null,"signatureCooked":null}}],"hasMore":false,"lastReadPostNumber":3,"hasUrlPostNumber":true}