School of EECS Washington State University CptS 570 - Machine - PowerPoint PPT Presentation

CptS 570 – Machine Learning School of EECS Washington State University CptS 570 - Machine Learning 1

 Course overview  What is machine learning?  Why do machine learning?  Applications  Approaches  Resources CptS 570 - Machine Learning 2

 Objectives ◦ Knowledge of machine learning (ML) foundations, paradigms and algorithms ◦ Techniques for evaluating ML algorithms ◦ Practical experience using ML algorithms ◦ Current ML issues  Website ◦ www.eecs.wsu.edu/~holder/courses/CptS570.html CptS 570 - Machine Learning 3

 Assignments ◦ Readings ◦ Six (6) homeworks (40%) ◦ Two (2) exams (20%) ◦ Project (20%) ◦ Presentation (10%) ◦ Critiques and class participation (10%) CptS 570 - Machine Learning 4

 Textbook ◦ Ethem Alpaydin (2010). Introduction to Machine Learning, Second Edition. MIT Press. ◦ www.cmpe.boun.edu.tr/~ethem/i2ml2e CptS 570 - Machine Learning 5

 What is learning?  What is machine learning? CptS 570 - Machine Learning 6

 Webster ◦ Gain knowledge or understanding of or Sleep learning? skill in by study, instruction or experience (www.links999.net) ◦ Memorize ◦ Synonym: Discovery  Obtain knowledge of for the first time  May imply acquiring knowledge with little effort or conscious intention (as by simply being told) or it may imply study and practice ◦ Knowledge  Knowing something with familiarity gained through experience or association  Facts or ideas acquired by study, investigation, observation, or experience CptS 570 - Machine Learning 7

 Herbert Simon (1970) ◦ Any process by which a system improves its performance.  Tom Mitchell (1990) ◦ A computer program that improves its performance at some task through experience.  Ethem Alpaydin (2010) ◦ Programming computers to optimize a performance criterion using example data or past experience. CptS 570 - Machine Learning 8

 How is knowledge represented?  How is experience represented?  What is the performance measure?  Knowledge acquisition vs. skill acquisition  Is deduction learning? CptS 570 - Machine Learning 9

 Automated knowledge engineering ◦ Expertise is scarce ◦ Codification of expertise is difficult ◦ Expertise frequently consists of a set of test cases ◦ Data from measurements, but no information or knowledge  Only one computer has to learn, then copy  Discover new knowledge  Understand human learning  Systems need to adapt to unknown, dynamic environments CptS 570 - Machine Learning 10

 Patient cases  [medical knowledge]  automated (better?) diagnosis  Autonomous driving  Speech recognition  Recommendations (Amazon, Netflix)  Prediction (business, financial, environment, health, energy, …)  Fraud/intrusion detection CptS 570 - Machine Learning 11

 Statistics  Pattern recognition  Signal processing  Control  Artificial intelligence  Data mining  Neuroscience  Cognitive science  Psychology CptS 570 - Machine Learning 12

 Supervised Learning ◦ Classification ◦ Regression  Unsupervised Learning ◦ Clustering  Reinforcement Learning CptS 570 - Machine Learning 13

D Default e b Good Status t Income CptS 570 - Machine Learning 14

Default Good Status D e b t If Income < t Then Default t Income CptS 570 - Machine Learning 15

No Loan Default Good Status D e b t if Debt < a*Income + b then Loan else No Loan Loan Income CptS 570 - Machine Learning 16

Cluster 2 Cluster 1 D e Categories b 1) Debt exceeds t Income 2) High Debt, Cluster 3 High Income 3) Low Debt Income CptS 570 - Machine Learning 17

No Loan Debt<50 yes no no D e Income Income b t 50- 50 >100 >100 <50 50- 50 <50 >100 >100 100 100 100 100 Loan NO YES YES NO NO YES Income CptS 570 - Machine Learning 18

Input Hidden Outpu put Layer Layer Layer No Loan 0.123 23 0.117 17 Debt Loan D e No No Income me b Loan t 0.203 03 0.545 45 Loan Income CptS 570 - Machine Learning 19

 Evaluation ◦ Which learning approach is better  Theoretical bounds ◦ What is and is not learnable  Scalability ◦ Learning from massive datasets CptS 570 - Machine Learning 20

 Software ◦ Weka (www.cs.waikato.ac.nz/~ml/weka) ◦ Machine learning open-source software (mloss.org)  Data ◦ UCI ML Repository (archive.ics.uci.edu/ml) ◦ UCI KDD Repository (kdd.ics.uci.edu) ◦ Challenges: KDD- Cup, Netflix, … CptS 570 - Machine Learning 21

 Conferences ◦ International Conference on Machine Learning (ICML) ◦ Knowledge Discovery and Data Mining (KDD) ◦ IEEE Conference on Data Mining (ICDM) ◦ SIAM Data Mining Conference (SDM) ◦ Association for the Advancement of AI (AAAI) ◦ International Joint Conference on AI (IJCAI) ◦ Many more … CptS 570 - Machine Learning 22

 Journals ◦ Machine Learning Journal ◦ Journal of Machine Learning Research ◦ Data Mining and Knowledge Discovery ◦ Many more …  WWW ◦ www.kdnuggets.com (subscribe!) CptS 570 - Machine Learning 23

 Machine learning is a computational process that improves performance based on experience.  Numerous successful methods  Maturing theory  Open and active research area CptS 570 - Machine Learning 24