Word Window Classification and Neural Network 201983289 1

  • Slides: 16
Download presentation
Word Window Classification and Neural Network 201983289 이유진 1

Word Window Classification and Neural Network 201983289 이유진 1

A Neural Network A single neuron A computational unit with n (3) inputs and

A Neural Network A single neuron A computational unit with n (3) inputs and 1 output and parameters W, b Bias Activation Output Inputs function 1. Single neuron as a linear classifier 2. Activation functions 3. Neural Network architectures 2

 • A Neural Network z = Wx +b 3

• A Neural Network z = Wx +b 3

A Neural Network 1. Single neuron as a linear classifier • 4

A Neural Network 1. Single neuron as a linear classifier • 4

A Neural Network 2. Activation functions • 활성화 함수 사용의 이유는 Data를 비선형으로 바꾸기

A Neural Network 2. Activation functions • 활성화 함수 사용의 이유는 Data를 비선형으로 바꾸기 위해 서이다. ① Sigmoid ②Tanh ③Re. LU ④Leaky Re. LU ⑤Maxout 5

④ Leaky Re. LU 9

④ Leaky Re. LU 9

3. Neural Network architectures multilayer neural network 11

3. Neural Network architectures multilayer neural network 11

3. Neural Network architectures 12

3. Neural Network architectures 12

Feed-forward Computation 13

Feed-forward Computation 13

The max-margin loss • 14

The max-margin loss • 14

Backprop • s는 U transpose a, a는 f(z), z는 Wx+b 16

Backprop • s는 U transpose a, a는 f(z), z는 Wx+b 16