[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"tags-groups":3,"chains-groupcat":4,"experts-list":5,"experts-coverage":35},[],[],{"list":6,"total":33,"page":33,"pageSize":34,"totalPages":33},[7],{"id":8,"name":9,"orgName":10,"school":10,"gender":11,"title":12,"degree":11,"honor":11,"position":11,"researchDirection":13,"intro":14,"province":11,"completeness":15,"avatarAuthorized":16,"sourceUrl":17,"nameValid":18,"tags":19,"tagNames":32},"85bc933a-5d51-45c3-a7db-ee9a531eb213","张闯","南京信息工程大学",null,"副教授","；l 深度学习模型在长期学习时会出现灾难性遗忘现象。比如，学习过任务A的神经网络在学习任务B时，它会忘记学过的任务A，只记得B。如何在动态的环境中像人一样具备长期学习的能力成为了学术界关注的热点。这固然是深度学习研究的一个新的方向，但我们应该探究给SNN增加额外的时间维度是否有助于实现持续性学习型任务。另一个类似的任务就是，利用少量数据进行学习，这也是SNN能超过深度学习的领域。SNN中的无监督学习可以与提供少量数据的监督学习相结合，只使用一小部分标记的训练数据得到高效的训练结果。；·；张海超，“融合注意力的轻量级行为识别网络研究 ”，已录用，电子测量与仪器学报。；·；乔丹， 张闯 ，朱晨雨。基于多尺度融合卷积神经网络的图像去雾算法，液晶与显示， vol.36, no.10, pp:1420-1429, 2021.10 。；葛涛，张闯，张海超，乔丹。基于注意力机制的图像去雾方法，已投稿，计算机工程与应用。","【产业方向】人工智能",72,false,"https:\u002F\u002Ffaculty.nuist.edu.cn\u002Fzhangchuang\u002Fzh_CN\u002Findex.htm",true,[20,23,26,29],{"code":21,"name":22},"CH-33","人工智能",{"code":24,"name":25},"GB-39","计算机、通信和其他电子设备制造业",{"code":27,"name":28},"GB-73","研究和试验发展",{"code":30,"name":31},"SEI-01","新一代信息技术产业",[22,25,28,31],1,30,{"schools":36},297]