Scholarlink

This chapter introduces the foundational concepts of artificial neural networksArtificialartificial neural network by focusing on the Perceptron, its limitations, and the development of Multi-LayerLayer Perceptrons (MLPs)Perceptronmulti-layer Perceptron (MLP) or feedforward neural networksFeedforward neural network. Beginning with early neuronNeuron models inspired by biologicalBiological systems, it presents the Perceptron as the first trainable neural networkNeural network and explores its capacity and limitations, such as the inability to solve nonlinearly separable problems. These challenges motivate the transition to multi-layerLayer architecturesArchitecture, which form the basis of deepLearningdeep learning learningDeepdeep learning. The chapter explains the structure of MLPsPerceptronmulti-layer Perceptron (MLP), including neuronsNeuron, layersLayer, activation functionsActivation function, and common loss functionsFunction. The chapter also offers an overview of related models—including ADALINEADALINE, logistic regressionLogisticlogistic regression, radial basis functionFunction networks, andSelf-organizing map self-organizing map—that provide additional historical and conceptual context but are not central to the main developmental trajectory of deep Learningdeep learning learningDeepdeep learning. The universal approximationApproximation theorem, in neural networksNeural network, is also discussed in this chapter.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

From Perceptron to Feedforward Neural Networks: Foundations and Early Models

  • 章节

  • 发表日期：

    2026年

查看全文
  • Benyamin Ghojogh,
  • Ali Ghodsi

摘要

This chapter introduces the foundational concepts of artificial neural networksArtificialartificial neural network by focusing on the Perceptron, its limitations, and the development of Multi-LayerLayer Perceptrons (MLPs)Perceptronmulti-layer Perceptron (MLP) or feedforward neural networksFeedforward neural network. Beginning with early neuronNeuron models inspired by biologicalBiological systems, it presents the Perceptron as the first trainable neural networkNeural network and explores its capacity and limitations, such as the inability to solve nonlinearly separable problems. These challenges motivate the transition to multi-layerLayer architecturesArchitecture, which form the basis of deepLearningdeep learning learningDeepdeep learning. The chapter explains the structure of MLPsPerceptronmulti-layer Perceptron (MLP), including neuronsNeuron, layersLayer, activation functionsActivation function, and common loss functionsFunction. The chapter also offers an overview of related models—including ADALINEADALINE, logistic regressionLogisticlogistic regression, radial basis functionFunction networks, andSelf-organizing map self-organizing map—that provide additional historical and conceptual context but are not central to the main developmental trajectory of deep Learningdeep learning learningDeepdeep learning. The universal approximationApproximation theorem, in neural networksNeural network, is also discussed in this chapter.