TTNT_Chapter9.ppt

Artificial Intelligence Trí Tuệ Nhân Tạo Chapter 9 – Học Máy Lê Quân Hà Outline • Machine Learning introduction  How Does Machine Learning Work?  Types of Machine Learning • Learning by Decision trees  Procedure: Buildtree for Building good ones  Entropy  Attribute-based representations  Decision tree learning algorithm • Sub-symbolic learning  Perceptrons  Linear Regression  Neural networks  NN Supervised Learning  Learning Example  Backpropagation observations  Kohonen learning  Problems with NNs Machine Learning Introduction • Why is machine learning important?  Early AI systems were brittle, learning can improve such a system’s capabilities  AI systems require some form of knowledge acquisition, learning can reduce this effort  KBS research clearly shows that producing a KBS is extremely time consuming – dozens of man-years per system is the norm  in some cases, there is too much knowledge for humans to enter (e.g., common sense reasoning, natural language processing)  Some problems are not well understood but can be learned (e.g., speech recognition, visual recognition)  AI systems are often placed into real-world problem solving situations  the flexibility to learn how to solve new problem instances can be invaluable  A system can improve its problem solving accuracy (and possibly efficiency) by learning how to do something better How Does Machine Learning Work? • Learning in general breaks down into one of two forms  Learning something new  no prior knowledge of the domain/concept so no previous representation of that knowledge  in ML, this requires adding new information to the knowledge base  Learning something new about something you already knew  add to the knowledge base or refine the knowledge base  modification of the previous representation  new classes, new features, new connections between them  Learning how to do something better, either more efficiently or with more accuracy  previous problem solving instance (case, chain of logic) can be “chunked” into a new rule (also called memoizing)  previous knowledge can be modified – typically this is a parameter adjustment like a weight or probability in a network that indicates that this was more or less important than previously thought Types of Machine Learning • There are many ways to implement ML  Supervised vs. Unsupervised vs. Reinforcement  is there a “teacher” that rewards/punishes right/wrong answers?  Symbolic vs. Subsymbolic vs. Evolutionary  at what level is the representation?  subsymbolic is the fancy name for neural networks  evolutionary learning is actually a subtype of symbolic learning  Knowledge acquisition vs. Learning through problem solving vs. Explanation-based learning vs. Analogy • We can also focus on what is being learned  Learning functions  Learning rules  Parameter adjustment  Learning classifications  these are not mutually exclusive, for instance learning classification is often done by parameter adjustment Learning by Decision trees: issues • Constructing a decision tree is very difficult! • Difficulty: how can we extract a simplified decision tree?  This implies (among other things) establishing a preference order (bias) among alternative decision trees.  Finding the smallest one proves to be VERY hard. Improving over the trivial one is okay. Procedure: Buildtree If all of the training examples are in the same class, then quit, else 1. Choose an attribute to split the examples. 2. Create a new child node for each value of the attribute. 3. Redistribute the examples among the children according to the attribute values. 4. Apply buildtree to each child node. Is this a good decision tree? Maybe? How do we decide? Entropy Measures the (im) purity in collection S of examples Entropy(S) = - [ p + log 2 (p + ) + p - log 2 (p - ) ] • p + is the proportion of positive examples. • p - is the proportion of negative examples. N.B. This is not a fully general definition of entropy. Learning decision trees Problem: decide whether to wait for a table at a restaurant, based on the following attributes: 1. Alternate: is there an alternative restaurant nearby? 2. Bar: is there a comfortable bar area to wait in? 3. Fri/Sat: is today Friday or Saturday? 4. Hungry: are we hungry? 5. Patrons: number of people in the restaurant (None, Some, Full) 6. Price: price range ($, $$, $$$) 7. Raining: is it raining outside? 8. Reservation: have we made a reservation? 9. Type: kind of restaurant (French, Italian, Thai, Burger) 10. WaitEstimate: estimated waiting time (0-10, 10-30, 30-60, >60) Attribute-based representations • Examples described by attribute values (Boolean, discrete, continuous) • E.g., situations where I will/won't wait for a table: • Classification of examples is positive (T) or negative (F)