2024. 1. 13. 09:04ㆍ코딩 도구/LG Aimers
LG Aimers: AI전문가과정 4차
Module 4. 『지도학습(분류/회귀)』
ㅇ 교수 : 이화여자대학교 강제원 교수
ㅇ 학습목표
Machine Learning의 한 부류인 지도학습(Supervised Learning)에 대한 기본 개념과 regression/classification의 목적 및 차이점에 대해 이해하고, 다양한 모델 및 방법 (linear and nonlinear regression, classification, ensemble methods, kernel methods 등)을 통해 언제 어떤 모델을 사용해야 하는지, 왜 사용하는지, 모델 성능을 향상시키는 방법을 학습하게 됩니다.
Advanced Classification Model
ㄴ Linear and non-linear model…
• Support vector machine (SVM)
• Neural network (NN)
-Support vector machine (SVM)
• Choose the linear separator (hyperplane) with the largest margin on either side
• Maximum margin hyperplane with support vectors
• Robust to outliers
Support vector :
an instance with the minimum margin, which will be the most sensible data points to affect the performance
-Margin
• Twice the distance from the hyperplane to the nearest instance on either side
-Optimization
• Optimal weight 𝑤 and bias 𝑏
• Classifies points correctly as well as achieves the largest possible margin
• Hard margin SVM – assumes linear separability
• Soft margin SVM – extends to non-separable cases
• Nonlinear transform & kernel trick
-Optimization
constraints: linearly separable; hard-margin linear SVM
objective function: linearly separable; hard-margin linear SVM
-Problem of SVM
• What if the data samples are not linearly separable?
-Support Vector Machine
not linearly separable; Kernel Trick
• Data is not linearly separable in the input space
• Data is linearly separable in the feature space obtained by a kernel
-Radial-basis function (RBF) kernel
• Radial-basis function kernel
-Artificial neural network (ANN)
non-linear classification model
-Artificial neural network (ANN)
activation functions
• Sigmoid neurons give a real-valued output that is a smooth and bounded function of their total input.
• Non-linearity due to the activation functions
-Artificial neural network (ANN)
deep neural network
-Artificial neural network (ANN)
multilayer perceptron
• Multilayer Perceptron (MLP)
• Proposed by Prof. Marvin Minsky at MIT (1969)
• Can solve XOR Problem
-Artificial neural network (ANN)
ANN for non-linear problem
Observation
• There exists cases when the accuracy is low even if the # layers is high. Why?
Answer
• The result of one ANN is the result of sigmoid function (between 0 and 1).
• The numerous multiplication of this result converges to near zero.
Gradient Vanishing Problem
-Breakthrough in Back Propagation
• Backpropagation (BP) barely changes lower-layer parameters (vanishing gradient)
Breakthrough
• Pre-training+ fine tuning
• Convolutional neural networks (CNN) for reducing redundant parameters.
• Rectified linear unit (constant gradient propagation)
• Dropout
Quiz
What answers are correct? Select all that apply.
A. Support vector machine can be applied to non-linear classification
Correct.
Kernel transform converts an input into a new feature space so that data is linearly separable in the feature space obtained by a kernel
B.
Neural network is a non-linear classifier given with an activation function, when a neuron produces an output. The performance is always improved with an increasing number of layers.
False.
The performance degrades when gradients are vanishing through activation functions
Summary
SVM
• Uses fewer support vectors in training data sets -> computationally efficient.
• For non linear data : use Kernel transformation
NN
• Provides a non-linear classification framework
• Needs elaborated training schemes to improve performance
'코딩 도구 > LG Aimers' 카테고리의 다른 글
LG Aimers 4기 인과성과 기본개념 (1) | 2024.01.15 |
---|---|
LG Aimers 4기 그리고 Ensemble Learning (0) | 2024.01.14 |
LG Aimers 4기 그리고 Linear Classification (0) | 2024.01.12 |
LG Aimers 4기 그리고 Gradient Descent (2) | 2024.01.11 |
LG Aimers 4기 그리고 Linear Regression (2) | 2024.01.10 |