Fundamental of Machine Learning
1 like90 views
The document presents an overview of machine learning, detailing its definitions, applications, and various paradigms such as supervised, unsupervised, and reinforcement learning. It also explains linear regression for predicting continuous values and logistic regression for classification tasks, highlighting their functionalities and limitations. Additionally, it addresses common issues in machine learning like overfitting and underfitting, along with potential solutions such as feature reduction and regularization.
Fundamental of Machine Learning
- 1. Some Basics of Machine Learning Sidik Soleman Prosa.ai Email: sidik.soleman@prosa.ai
- 2. 1. Machine Learning 2. Linear Regression 3. Logistic Regression Agenda 2
- 4. Machine Learning in Everyday Life Search Engine Machine Translation Spelling Checker Spam Detection Chatbot 4
- 5. Definition of Artificial Intelligence (AI) 1. Acting Humanly ○ The turing test approach ○ Computer should to be able to do ■ Natural language processing ■ Knowledge representation ■ Automated reasoning ■ Machine learning (ML) ● To adapt to new circumstances and to detect and extrapolate patterns. ■ Computer vision ■ Robotics 5 Artificial Intelligence - A Modern Approach by Peter Norvig, Stuart J. Russell
- 6. Definition of Artificial Intelligent (AI) 2. Thinking Humanly ○ The cognitive approach ○ Understanding human brain by means ■ Introspection—trying to catch our own thoughts as they go by; ■ Psychological experiments—observing a person in action; and ■ Brain imaging—observing the brain in action 6 Artificial Intelligence - A Modern Approach by Peter Norvig, Stuart J. Russell
- 7. Definition of Artificial Intelligent (AI) 3. Thinking Rationally ○ The “law of thought” approach ○ Logic → correct inference 4. Acting Rationally ○ The rational agent approach ○ Act rationally is to reason logically to the conclusion and then to act on that. ○ But, correct inference is not all of rationality; in some situations, there is no provably correct thing to do, but something must still be done 7 Artificial Intelligence - A Modern Approach by Peter Norvig, Stuart J. Russell
- 8. Machine Learning Paradigms 1. Supervised Learning ○ Learning based on experience/examples, e.g. classification task 8 Training Data Machine (Model) Input Cat
- 9. Machine Learning Paradigms (Cont’d) 2. Unsupervised Learning ○ Finding hidden pattern from data, e.g. clustering task 9 Data Machine (Model) Output
- 10. Machine Learning Paradigms (Cont’d) 3. Reinforcement Learning ○ Learning based on the interaction with environment to achieve a goal, e.g. to win chess game ○ Decision will be rewarded or punished 10 Machine (Model) Environmen t (action) (state, reward)
- 11. Machine Learning Algorithms 11 Supervised Learning Unsupervised Learning Reinforcement Learning - Linear Regression - Logistic Regression - Support Vector Machine - Decision Tree - Random Forest - Naive Bayes - Deep Learning - K-Means Clustering - K-Nearest Neighbour - Apriori Algorithm - Principal Component Analysis - Latent Dirichlet Allocation - Deep Learning - Temporal Difference Learning - Monte Carlo - Q-learning - Policy Gradients - Deep Reinforcement Learning
- 12. 12 2.Linear Regression Based on machine learning course in Coursera by Andrew Ng https://www.coursera.org/learn/machine-learning
- 13. Linear Regression ● Predicting the price of house 13 What is the price of a 200m2 house? Finding the best linear function f(x) g(x) h(x)
- 14. Linear Regression (Cont’d) ● Predicting the price of house 14 What is the price of a 200m2 house? The best linear function ● ● Finding the best parameters so that is close to y General Setup ● m= # of training examples ● x = input variable/features ● y = output variable/target ● (x, y) = one training example ● (xi, yi) = i-th training example h(x)
- 15. Linear Regression (Cont’d) ● Predicting the price of house 15 What is the price of a 200m2 house? Finding the best parameter h(x) cost function
- 16. Gradient Descent: Cost Function Derivation ● Predicting the price of house 16 What is the price of a 200m2 house? Finding the best parameter h(x) Repeat until convergence { } Simultaneous Update
- 17. Derivation Intuition 17 Positive Slope Negative Slope Derivation guide the direction of should be adjusted. α is the learning rate to control how big the movement of
- 18. Gradient Descent Algorithm ● Predicting the price of house 18 What is the price of a 200m2 house? Finding the best parameter h(x) Repeat until convergence { } Gradient Descent Algorithm
- 19. The Answer of the Prediction ● Predicting the price of house 19 What is the price of a 200m2 house? h(x)
- 20. Multivariable Linear Regression 20 Previous linear regression function ● Multivariable linear regression ● m= # of training examples ● x(i) = input variables/features of i-th example ● y(i) = output variable of i-th example ● x(i) j = value of feature j in i-th example repeat { } Cost function Gradient descent
- 21. 21 3.Logistic Regression Based on machine learning course in Coursera by Andrew Ng https://www.coursera.org/learn/machine-learning
- 22. Classification Task 22 ● Linear regression issues ○ Prediction score can be more or less than 1 ○ Not suitable for classification task ● Solution: logistic regression Parameter and x as matrix 1 0.5 Score is also probability
- 23. Regression Model 23 Training Set n parameter Cost repeat { } Gradient descent
- 24. Multi-class Classification ● Binary vs multi-class classification ● Handling multi-class classification ○ One vs Rest /One vs All ■ 4 Class Classification → 4 Classification Model ■ Label with maximum score is the answer 24
- 25. Machine Learning Problems ● Overfit ○ Hypothesis fit well with the training data → high performance ○ But fail to make generalization → poor performance on test data ○ High variance ○ Too many features ● Underfit ○ High bias → low performance ○ Fail to make generalization → poor performance on test data 25
- 26. Some solution for the problem ● Reduce number of features ○ Select features to keep ○ Model selection ● Regularization ○ Keep all features, but reduce magnitude or values of parameters (theta) ○ Works well when having a lot of features 26
- 27. Regularization 27 Regularization Note: Lambda high → underfit and high bias Lambda too small → overfit
- 28. Take Home Message ● Machine learning is to adapt to new circumstances and to detect and extrapolate patterns. ● Linear regression for predicting real continuous values ● Logistic regression for predicting class in classification 28