Category of Machine Learning

지도학습

• Classification
• Regreesion

비지도학습

• Clustering

강화학습

• Markov Decision Process

Supervised Learning

• Output y is categorical variable : Classification
• Output y is continuous : Regression

Regression

• Ouput(y) , Input(x)
• x = n*p matrix 형태
  

Classification

• Output(y), Input(x)
• Each image is an input and corresponding class is output
  

Model Evaluation and Performance

• We must train the function for appropriately demonstrating the relationship between input and output variables
• The word 'appropriately' means the difference between results from real data and model's prediction is small
  
  

-> which model is better than the other? => A

Machine Learning and Optimization

• Machine learning models optimize the function by minimizing the loss
• The process of optimization for the machine learning functions represents learning
  

Generalization Error and Hyperparameter

• Training error
  - prediction error to training data(Loss)

• Validation error
  - prediction error to validation data(Generalizaion error)

• As the complexity of the model increases,
  the training error decreases but the validation error increases from a critical point

• Machine learning model's goal is predicting unseen data accurately
  

Model Validation with Data

• Validation with training data only (Bad)
• Training, Testing data (After parameter tuning)
• Training, Validation, Testing data (Best)
• Cross-Validation(with small number of data)
  

reference : K-MOOC 실습으로 배우는 머신러닝