#2. ML 기본 개념

- 기존의 프로그램의 경우 Rule 기반으로 개발됨

- 1959년 Arthur Samuel이 제안: 

Field of study that gives computers the ability to learn without being explicitly programmed

학습을 위한 과정은 Data storage > Abstraction > Generalization > Evaluation 단계를 거침.

ML은 목적에 따라 크게 4가지 카테고리로 구분됨.

출처: http://en.proft.me/media/science/ml_types2.png

▣ Supervised learning

• Predictive Model
• Labeled Data를 사용
• 주요 알고리즘
• Classification: Logistic Regression, Decision Tree, KNN, Random Forest, SVM, Naive Bayes, etc
• Numeric prediction: Linear Regression, KNN, Gradient Boosting & AdaBoost, etc
• ex) Image labeling, Email span filter, Predicting exam score
• 유형은 크게 Regression과 Classification임.
• Regression: 범위 내 값 예측
• Binary Classification: Pass / Non-pass
• Multi-label Classification: 등급 판정

출처: http://en.proft.me/2015/12/24/types-machine-learning-algorithms/

▣ Unsupervised learning

• Descriptive Model
• Unlabeled Data
• 주요 알고리즘
• Clustering: K-Means
• Pattern Discovery
• ex) Google news grouping, Word clustering

▣ Semi-supervised learning

• labeled / unlabeled 데이터를 사용

▣ Reinforcement learning. 

• Using this algorithm, the machine is trained to make specific decisions. 
• It works this way: the machine is exposed to an environment where it trains itself continually using trial and error. This machine learns from past experience and tries to capture the best possible knowledge to make accurate business decisions. 
• Example of Reinforcement Learning: Markov Decision Process.