第一章 绪论
深度学习:神经网络类的算法均可称为深度学习
涉及概念:
数据集dataset,示例instance,样本sample,属性attribute,特征feature,属性值attribute value,属性空间attribute space,样本空间samplespace,特征向量feature vector,维数dimensionality
训练数据training data,训练样本training sample,训练集training set,假设hypothesis,真相ground-truth,学习器learner
标记label,标记空间label space,分类classification,回归regression,二分类binaryclassification,正类positive class,反类negative class,多分类multi-classclassification,聚类clustering,簇cluster,监督学习supervised learning,无监督学习unsupervised learning,泛化generalization,独立同分布independent and identically distributed
归纳induction,演绎deduction,版本空间version space
归纳偏好inductive bias,特征选择feature selection,奥卡姆剃刀Occam’srazor,NFL定理no free lunch theorem
发展历程:
1950s-1970s,推理期,逻辑理论家logictheorists,通用问题求解general problem solving;连接主义connectionism(感知机perceptron),符号主义symbolism
1970s,知识期,专家系统;
1980s-1990s,学习期,符号主义学习(决策树decisiontree,归纳逻辑程序设计inductive logic programming),基于神经网络的连接主义学习(BP算法)
1990s,统计学习statistical learning(支持向量机supportvector machine,核方法kernel methods)
2000s,深度学习
2010s,大数据时代关键技术(机器学习,云计算,众包crowdsourcing),数据挖掘data mining