Downloaded 88 times
Uploaded byRicardo Wendell Rodrigues da Silveira
Machine learning
This document provides an overview of machine learning techniques including artificial neural networks, clustering, genetic algorithms, and reinforcement learning. It discusses how machines can learn through supervised and unsupervised methods, using techniques from statistics, brain modeling, and more. Specific algorithms covered include backpropagation for training neural networks, k-means clustering, genetic algorithms that represent solutions as chromosomes, and reinforcement learning approaches like Markov decision processes. The goal is to explain how different machine learning methods can allow computers to learn without being explicitly programmed.
Machine learning
- 1.
- 2. 2 Artificial Intelligence & MachineLearning Artificial Neural Networks Clustering Genetic Algorithms Reinforcement Learning Q&A Agenda
- 3.
- 4.
- 5. What’s involved inintelligence? ü Ability to interact with the real world ü Reasoning and planning ü Learning and adaptation 5
- 6. What’s involved inintelligence? ü Ability to interact with the real world ü Reasoning and planning ü Learning and adaptation 6
- 7. “To learn isto gain knowledge, or understanding of, or skill in, by study, instruction, or experience” 7
- 8. Machine Learning Field ofstudy that gives computers the ability to learn without being explicitly programmed 8
- 9.
- 10. Why? Why should machineshave to learn? 10
- 11. Why? Why should machineshave to learn? Why not design machines to perform as desired in the first place? 11
- 12. Some reasons… There aretasks that cannot be defined well except by example It is possible that hidden among large piles of data are important relationships and correlations 12
- 13.
- 14.
- 15.
- 16.
- 17. Ways of addressthese problems… ü Statistics ü Brain Models ü Adaptive Control ü Evolutionary Models ü Psychological Models 17
- 18. Ways of addressthese problems… ü Statistics ü Brain Models ü Adaptive Control ü Evolutionary Models ü Psychological Models 18
- 19.
- 20.
- 21.
- 22.
- 23. How do wetrain it? 23
- 24. Backpropagation Algorithm initialize the weights in the network do for each example e in the training set O = neural-‐net-‐output(network, i) T = teacher output for e compute error (T -‐ O) at the output units compute delta_wh for all weights from hidden layer to output layer compute delta_wi for all weights from input layer to hidden layer update the weights in the network until all examples classified correctly or stopping criterion satisfied return the network 24
- 25.
- 26. Drawbacks ANNs with manyhidden layers can solve difficult problems… but are very hard to train! 26
- 27.
- 28.
- 29.
- 30.
- 31. Many applications Medical imaging Marketresearch Climatology Social networks analysis Recommender systems 31
- 32.
- 33. Given a population,how can we find the best group of individuals that satisfy some criteria? 33
- 34. Given a population,how can we find the best group of individuals that satisfy some criteria? 34
- 35. Typical requirements a geneticrepresentation of the solution domain a fitness function to evaluate the solution domain 35
- 36. Genetic representation Each solutionis represented as a “chromosome” Each solution has a fitness value 36
- 37.
- 38. Example: Mona Lisafrom 1500 characters 38 http://www.youtube.com/watch?v=TManzvC9pi8&NR
- 39.
- 40. Main concepts Inspired bybehaviorist psychology Learning by interacting with the environment 40 Suited for problems which include a long- term versus short-term reward trade-off
- 41. How to teacha computer to play games? 41
- 42.
- 43. How to flippancakes? 43 http://www.youtube.com/watch?v=W_gxLKSsSIE
- 44.
- 45. Conclusion There are manytechniques… Results are heavily influenced by input representation It’s a math-heavy field! 45
- 46.
- 47.
Editor's Notes
- #10 Several reasons: -
- #11 Several reasons: -
- #12 Several reasons: -
- #16 Supervided: Classification, Regression, Prediction
- #17 Unsupervisioned: taxonomic problems in which it is desired to invent ways to classify data into meaningful categories.
- #22 Also called perceptronsLinear aproximators
- #29 Usually unsupervisedTry to map high dimensional data to lower dimensions