1. 1 What is learning? “Learning denotes changes in a system that... enable a system to do the same task more efficiently the next time.” –Herbert Simon “Learning is any process by which a system improves performance from experience.” –Herbert Simon “Learning is constructing or modifying representations of what is being experienced.” –Ryszard Michalski “Learning is making useful changes in our minds.” – Marvin Minsky

2. 2 Machine Learning - Example One of my favorite AI/Machine Learning sites:  http://www.20q.net/ http://www.20q.net/

3. 3 Why learn? Build software agents that can adapt to their users or to other software agents or to changing environments  Personalized news or mail filter  Personalized tutoring  Mars robot Develop systems that are too difficult/expensive to construct manually because they require specific detailed skills or knowledge tuned to a specific task  Large, complex AI systems cannot be completely derived by hand and require dynamic updating to incorporate new information. Discover new things or structure that were previously unknown to humans  Examples: data mining, scientific discovery

4. 4

5. 5 Applications Assign object/event to one of a given finite set of categories.  Medical diagnosis  Credit card applications or transactions  Fraud detection in e-commerce  Spam filtering in email  Recommended books, movies, music  Financial investments  Spoken words  Handwritten letters

6. 6 Major paradigms of machine learning Rote learning – “Learning by memorization.”  Employed by first machine learning systems, in 1950s Samuel’s Checkers program Supervised learning – Use specific examples to reach general conclusions or extract general rules Classification (Concept learning) Regression Unsupervised learning (Clustering) – Unsupervised identification of natural groups in data Reinforcement learning– Feedback (positive or negative reward) given at the end of a sequence of steps Analogy – Determine correspondence between two different representations Discovery – Unsupervised, specific goal not given

7. 7 Rote Learning is Limited Memorize I/O pairs and perform exact matching with new inputs If a computer has not seen the precise case before, it cannot apply its experience We want computers to “generalize” from prior experience  Generalization is the most important factor in learning

8. 8 The inductive learning problem Extrapolate from a given set of examples to make accurate predictions about future examples Supervised versus unsupervised learning  Learn an unknown function f(X) = Y, where X is an input example and Y is the desired output.  Supervised learning implies we are given a training set of (X, Y) pairs by a “teacher”  Unsupervised learning means we are only given the Xs.  Semi-supervised learning: mostly unlabelled data

9. 9 Types of supervised learning a)Classification: We are given the label of the training objects: {(x1,x2,y=T/O)} We are interested in classifying future objects: (x1’,x2’) with the correct label. I.e. Find y’ for given (x1’,x2’). b)Concept Learning: We are given positive and negative samples for the concept we want to learn (e.g.Tangerine): {(x1,x2,y=+/-)} We are interested in classifying future objects as member of the class (or positive example for the concept) or not. I.e. Answer +/- for given (x1’,x2’). x1=size x2=color Tangerines Oranges Tangerines Not Tangerines

10. 10 Types of Supervised Learning Regression  Target function is continuous rather than class membership  y=f(x)

11. 11 Example Positive ExamplesNegative Examples How does this symbol classify? Concept Solid Red Circle in a (regular?) Polygon What about? Figures on left side of page Figures drawn before 5pm 2/2/89

12. 12 Inductive learning framework: Feature Space Raw input data from sensors are typically preprocessed to obtain a feature vector, X, that adequately describes all of the relevant features for classifying examples Each x is a list of (attribute, value) pairs x = [Color=Orange Shape=Round Weight=200g ] Each attribute can be discrete or continuous Each example can be interpreted as a point in an n-dimensional feature space, where n is the number of attributes Model space M defines the possible hypotheses  M: X → C, M = {m 1, … m n } (possibly infinite) Training data can be used to direct the search for a good (consistent, complete, simple) hypothesis in the model space

13. 13 Feature Space Size Color Weight ? Big 2500 Gray A “concept” is then a (possibly disjoint) volume in this space.

14. 14 Learning: Key Steps data and assumptions – what data is available for the learning task? – what can we assume about the problem? representation – how should we represent the examples to be classified method and estimation – what are the possible hypotheses? – what learning algorithm to use to infer the most likely hypothesis? – how do we adjust our predictions based on the feedback? evaluation – how well are we doing? …

15. 15

16. 16 Evaluation of Learning Systems Experimental  Conduct controlled cross-validation experiments to compare various methods on a variety of benchmark datasets.  Gather data on their performance, e.g. test accuracy, training- time, testing-time.  Analyze differences for statistical significance. Theoretical  Analyze algorithms mathematically and prove theorems about their: Computational complexity Ability to fit training data Sample complexity (number of training examples needed to learn an accurate function)

17. 17 Measuring Performance Performance of the learner can be measured in one of the following ways, as suitable for the application:  Accuracy performance Number of mistakes Mean Square Error  Solution quality (length, efficiency)  Speed of performance  …

18. 18

19. 19 Curse of Dimensionality

20. 20 Curse of Dimensionality Imagine a learning task, such as recognizing printed characters. Intuitively, adding more attributes would help the learner, as more information never hurts, right? In fact, sometimes it does, due to what is called curse of dimensionality  it can be summarized as the situation where the available data may not be sufficient to compensate for the increased number of parameters that comes with increased dimensionality

21. 21 Curse of Dimensionality

22. 22 Polynomial Curve Fitting

23. 23 Sum-of-Squares Error Function

24. 24 0 th Order Polynomial

25. 25 1 st Order Polynomial

26. 26 3 rd Order Polynomial

27. 27 9 th Order Polynomial

28. 28 Over-fitting Root-Mean-Square (RMS) Error:

29. 29 Data Set Size: 9 th Order Polynomial

30. 30 Data Set Size: 9 th Order Polynomial

31. 31 Lessons Learned about Learning Learning can be viewed as using direct or indirect experience to approximate a chosen target function. Function approximation can be viewed as a search through a space of hypotheses (representations of functions) for one that best fits a set of training data. Different learning methods assume different hypothesis spaces (representation languages) and/or employ different search techniques.

32. 32 Issues in Machine Learning Training Experience  What can be the training experience (labelled samples, self-play,…) Target Function  What should we aim to learn?  What should the representation of the target function be (features, hypothesis class,…) Learning:  What learning algorithms exist for learning general target functions from specific training examples?  Which algorithms can approximate functions well (and when)?  How does noisy data influence accuracy?  How does number of training samples influence accuracy? Training Data:  How much training data is sufficient?  How does number of training examples influence accuracy?  What is the best strategy for choosing a useful next training experience? How it affects complexity? Prior Knowledge/Domain Knowledge:  When and how can prior knowledge guide the learning process? Evaluation:  What specific target functions/performance measure should the system attempt to learn? …