电子科技大学：《统计学习理论及应用 Statistical Learning Theory and Applications》课程教学资源（课件讲稿，英文版）Lecture 04 Perceptron

Statistical Learning Theory and Applications Lecture 4 Perceptron Instructor:Quan Wen SCSE@UESTC Fal,2021

Statistical Learning Theory and Applications Lecture 4 Perceptron Instructor: Quan Wen SCSE@UESTC Fall, 2021

Outline (Level 1) 1Preparatory knowledge 2 Perceptron concept 3 Learning strategies of perceptron 4 Perceptron learning algorithm 5 Convergence of perceptron learning algorithm 6 Dual form of perceptron learning algorithm 1/66

Outline (Level 1) 1 Preparatory knowledge 2 Perceptron concept 3 Learning strategies of perceptron 4 Perceptron learning algorithm 5 Convergence of perceptron learning algorithm 6 Dual form of perceptron learning algorithm 1 / 66

Topics: ●Perceptron concept Perceptron learning strategy o Perceptron learning algorithm o Dual form of perceptron algorithm Key points and difficulties: oKey points:Perceptron learning strategy o Difficulties:Dual form of perceptron algorithm 2/66

Topics: Perceptron concept Perceptron learning strategy Perceptron learning algorithm Dual form of perceptron algorithm Key points and difficulties: Key points: Perceptron learning strategy Difficulties: Dual form of perceptron algorithm 2 / 66

Outline (Level 1) Preparatory knowledge Perceptron concept Learning strategies of perceptron Perceptron learning algorithm Convergence of perceptron learning algorithm Dual form of perceptron learning algorithm 3/66

Outline (Level 1) 1 Preparatory knowledge 2 Perceptron concept 3 Learning strategies of perceptron 4 Perceptron learning algorithm 5 Convergence of perceptron learning algorithm 6 Dual form of perceptron learning algorithm 3 / 66

1.Preparatory knowledge Structure of a Typical Neuron Dendrites- Axon Terminals Nucleus Node of Ranvier Schwann's Cells Axon Myelin Sheath Cell Body 4/66

1. Preparatory knowledge 4 / 66

The Simple Perceptron I 5 W =1 dendrite:input; axon:output 5166

The Simple Perceptron I o = fact( P 5 n=1 in · wn) dendrite: input; axon: output 5 / 66

Threshold Logic Unit (TLU) inputs N weights activation output W2 X ∑ o n a= WiXi y= {0 ,ifa≥0 otherwise 6/66

Threshold Logic Unit (TLU) a = P n i=1 wixi y = n 1 , if a ≥ θ 0 , otherwise 6 / 66

Outline (Level 1) Preparatory knowledge ②Perceptron concept Learning strategies of perceptron Perceptron learning algorithm Convergence of perceptron learning algorithm Dual form of perceptron learning algorithm 7166

Outline (Level 1) 1 Preparatory knowledge 2 Perceptron concept 3 Learning strategies of perceptron 4 Perceptron learning algorithm 5 Convergence of perceptron learning algorithm 6 Dual form of perceptron learning algorithm 7 / 66

2.Perceptron concept Perceptron is a linear binary classification model. 1 Input is the feature vector of the training sample 2 output is the category of the sample,taking +1 and-1 3 a hyper-plane in input space(feature space),dividing samples into positive and negative types 4 a discriminative model Linear regression model:Output is continuous Perceptron:Output is discrete 8/66

2. Perceptron concept ▶ Perceptron is a linear binary classification model. 1 Input is the feature vector of the training sample 2 output is the category of the sample, taking +1 and −1 3 a hyper-plane in input space (feature space), dividing samples into positive and negative types 4 a discriminative model ▶ Linear regression model: Output is continuous ▶ Perceptron: Output is discrete 8 / 66

Outline (Level 1-2) Perceptron concept Definition of Perceptron o Geometric interpretation of perceptron 9166

Outline (Level 1-2) 2 Perceptron concept Definition of Perceptron Geometric interpretation of perceptron 9 / 66