[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$fpZ9ViRDtuC47r5iDGagdNrM8OYOaAjkJsZOaUfPZqPs":3},{"answer":4,"createTime":5,"id":6,"options":7,"origin":12,"question":18,"related":19,"source":27,"type":68},[],"2023-05-07 18:05:41",4700770,[8,9,10,11],"BP网络是一种前馈网络,其隐单元必须分层,又称为多层前馈网络","反向传播的目的是为了修改各层神经元的权值,使误差信号最小","正向传播是输入信息由输入层传至隐层,最终在输出层输出","BP神经网络学习算法最核心的三部分是权值调整、输出层连接权调整、隐层连接权调整",{"courseId":13,"courseImg":14,"courseName":15,"workId":16,"workName":17},"1000009025","https:\u002F\u002Ftihai-oss-cloud.itihey.com\u002Fimg\u002Ff54758487d8301812024ebde85c2bf0d.jpg","人工智能导论","2843209","第四章单元测试","以下有关BP网络说法错误的是( )",[20,29,34,39,44,49,59,69,72,81],{"answer":21,"createTime":5,"id":22,"options":23,"question":26,"source":27,"type":28},[],4700747,[24,25],"对","错","人工神经网络从一个方面上模拟大脑: 组成人工神经网络的神经元之间的连接强度,即突触权值w ,可用于储存获取的知识.( )","v2",3,{"answer":30,"createTime":5,"id":31,"options":32,"question":33,"source":27,"type":28},[],4700751,[24,25],"感知机中,通过学习来调整权值,以使网络对任何的输入都能得到期望的输出.( )",{"answer":35,"createTime":5,"id":36,"options":37,"question":38,"source":27,"type":28},[],4700754,[24,25],"多层前馈网络的每一层都是单层的网络,却无法用单层感知器的学习方法.( )",{"answer":40,"createTime":5,"id":41,"options":42,"question":43,"source":27,"type":28},[],4700755,[24,25],"卷积神经网络是所谓深度神经网络的最重要的模型,深度就是隐层非常多的意思,深度越深,性能越好.( )",{"answer":45,"createTime":5,"id":46,"options":47,"question":48,"source":27,"type":28},[],4700756,[24,25],"不同的深度神经网络产生的深度学习技术性能是有差异的,用途也不一样.( )",{"answer":50,"createTime":5,"id":51,"options":52,"question":57,"source":27,"type":58},[],4700762,[53,54,55,56],"1949年心理学家赫布在其论著《行为自组织》,提出赫布规则","1957年罗森布拉特定义了一个神经网络结构,称为感知器","1982年物理学家霍普菲尔德提出了全联接网络,离散的神经网络模型.这是一种全新的具有完整理论基础的神经网络模型","1986年美国的一个平行计算研究小组提出了前项反馈神经网络的反向传播学习(BP)学习算法","以下关于人工智能发展史的说法正确的是( )",1,{"answer":60,"createTime":5,"id":61,"options":62,"question":67,"source":27,"type":68},[],4700763,[63,64,65,66],"感知器模型由输入层和输出层两层构成,不失一般性","感知机的本质是一种适合用于将某些数据分为两种类型的线性分类模型","感知机能够求解异或问题","感知器学习的基本思想:神经单元之间连接权的变化正比于输出单元期望输出与实际的输出之差","以下关于感知机模型说法错误的是( )",0,{"answer":70,"createTime":5,"id":6,"options":71,"question":18,"source":27,"type":68},[],[8,9,10,11],{"answer":73,"createTime":5,"id":74,"options":75,"question":80,"source":27,"type":68},[],4700773,[76,77,78,79],"人工神经网络主要由输入层、隐藏层和输出层组成.深度学习一般要求有多个隐藏层","卷积神经网络具有天然的网络权重值共享和网络局部稀疏性连接的特性,这种连接恰好也符合生物神经元的稀疏性响应特性","卷积神经网络有两个基本概念:权值共享和池化.权值共享使权值参数的个数减小;池化可以使特征图减小,简化计算","损失函数层的作用是用来估算模型的预测值与实际值的差距的函数,它是一个非负的实数值函数,它的值越小,反应该网络的数据拟合性能越好,也就是其结果越逼近原始输入数据","以下关于卷积神经网络说法错误的是( )",{"answer":82,"createTime":5,"id":83,"options":84,"question":89,"source":27,"type":68},[],4700778,[85,86,87,88],"RNN:语音识别、问答系统、语言建模和文本生成等诸多领域","GAN:图像合成、语义图像编辑、风格迁移、图像超分辨率技术和分类","VAE:从生成假人脸到合成音乐等","LSTM:捕捉到序列中长距离历史信息,但不能缓解长期依赖问题,应使用RNN实现有效缓解长期依赖","以下有关深度神经网络应用说法错误的是( )"]