Welcome to Jiyuniverse

Part 1. Introduction to Deep Neural Networks 본문

ML, DL/Lg Aimers

Part 1. Introduction to Deep Neural Networks

JJiiyun 2025. 1. 27. 00:49

 

KAIST 주재걸 교수님의 딥러닝 강의를 듣고 필기한 내용입니다. 추후 정리 예정

 

  • One kind of neural networks
  • Input node, Weights → Output node

Single Layer Perceptron

  • 활성함수 : Hard thresholding function
    • 내적 + bias 값이 0 이상이면 1, 음수이면 0
  • AND, OR, XOR Gate로 표현 가능

Decision Boundary in Perception

하나의 직선만으로 양수, 음수 구분 가능 (AND, OR)

 

But, It is not possible to solve a XOR problem using a single layer perceptron

→ Single layer perceptron can only solve linear problem. XOR problem is non-linear

 

XOR problem을 해결할 때에는 Multi-Layer Perceptron 사용 (Hidden layer)

multi-Layer perceptron

 

Forward Propagation

  • 수학적으로 표현함
  • 활성함수 : Sigmoid function
  • 내적한 값을 활성함수 g(x)에 통과시키면 output node 값이 됨.

Linear Layer / Fully-connected Layer

  • The layer where perceptrons are fully connected between two layers

Loss function

  • MSE (Mean-Squared Error)
  • Softmax Loss (or Cross-Entropy Loss)
  • Logistic Regression (Binary Class, Softmax Classfier 사용) = BCE?