1 machine learning cmput 466/551 nilanjan ray department of computing science university of alberta
Post on 21-Dec-2015
217 views
TRANSCRIPT
2
Introduction
• What is machine learning (ML)?
• Taxonomy in ML
• Applications/Examples
• Related disciplines
• References
• Resources
3
What is machine learning (ML)?
• Definition of “learning” from Merriam-Webster: “To gain knowledge or understanding of or skill in by study, instruction, or experience”
• ML = Learning in machines (computers)• ML techniques are algorithms that enable the
machines to improve its performance at some task through experience
• Tasks: recognition, diagnosis, prediction, planning, data mining, robot control, and so on.
4
Why Machine Learning?
• Some tasks can be specified only by training data/examples
• Human expertise may be scarce and/or very costly
• Amount of knowledge might be too large for explicit encoding by humans
• Modeling/Hidden parameter estimation: Often only data from measurements are available
• Computational power is ever increasing• Growing data pool and storage capacity
5
A Brief History about ML
• 1950’s: – Samuel’s checker program– Rosenblatt’s perceptron
• 1960’s: – Neural network– Pattern recognition
• 1970’s:– Winston’s ARCH– Buchanan and Mitchell’s Meta-Dendral: mass spectrometry
prediction rules– Quinlan’s ID3: Chess end-game rules– Michalski’s AQ11: Soybean disease diagnosis rules– MACROPS: macro operators in block world planning
6
A Brief History about ML…
• 1980’s: Gains momentum– Learning theory– Symbolic learning algorithms– Connectionist learning algorithm– Clustering– Explanation-based learning– Knowledge guided inductive learning– Genetic algorithm
• 1990’s: Maturity– Data mining– Ensemble learning: bagging, boosting etc.– Kernel methods– Reinforcement learning– Theoretical analysis
7
Typical Taxonomy for ML
• Supervised learning
• Unsupervised learning
• Semi-supervised learning
• Reinforcement learning
8
Supervised Learning
• Training data available in the form of (input, output) pairs
• When output is continuous valued the problem is called regression; if the output is qualitative or categorical the problem is called classification
• The goal here is to estimate the output for a novel (never-seen-before) input, after learning the training data
9
A Supervised Learning Example(From [DHS] book)
• “Sorting incoming Fish on a conveyor according to species using optical sensing”
Sea bass
Species
Salmon
10
• Set up a camera and take some sample images to extract features– Length– Lightness– Width– Number and shape of fins– Position of the mouth, etc…
Problem Analysis
11
• Use a segmentation operation to isolate fishes from one another and from the background
• Information from a single fish is sent to a feature extractor whose purpose is to reduce the data by measuring certain features
• The features are passed to a classifier
Preprocessing
13
Feature for Classification
Can we select the length of the fish as a possible feature for discrimination?
15
• Adopt the lightness and add the width of the fish
Fish xT = [x1, x2]
Lightness Width
Yet Another Feature: Width
18
Unsupervised Learning
• No training data in the form of (input, output) pair is available
• Applications:– Dimensionality reduction– Data compression– Outlier detection– Classification– Segmentation/clustering– Probability density estimation– …
20
• Applications on DNA microarray– Clustering: Group genes or samples into similar
expression profiles– Bi-clustering: Subset of genes exhibiting similar
expression pattern along a subset of samples– Dimension reduction– …
Example: Unsupervised Learning (contd..)
21
Semi-supervised Learning
• Uses both labeled data (in the form (input, output) pairs) and unlabelled data for learning
• When labeling of data is a costly affair semi-supervised techniques could be very useful
• Examples: Generative models, self-training, co-training
22
Example: Semi-supervised Learning
Source: Semi-supervised literature survey by X. Zhu, Technical Report
23
Reinforcement Learning
• Reinforcement learning is the problem faced by an agent that must learn behavior through trial-and-error interactions with a dynamic environment.
• There is no teacher telling the agent wrong or right• There is critic that gives a reward / penalty for the
agent’s action• Applications:
– Robotics– Combinatorial search problems, such as games– Industrial manufacturing– Many others!
24
Example: Reinforcement Learning
Tic Tac Toe TD-Gammon
Goal Learn to play optimal game
Learn to play game at master level
States All possible board states - 9
All possible board states - 1020
Action A new X in an empty field
21 dice combinations & avg. 20 legal moves
Reinforcement Signal
+10 winning -1 for every move
that did not win
‘1 ‘ for a reward ‘0‘ for a penalty
25
Related Disciplines
• Statistics
• Artificial Intelligence
• Psychology
• Vision and Neuroscience
• Control Theory
• Signal and Image Processing
• …..
26
References and Journals• Text: The Elements of Statistical Learning by Hastie, Tibshirani, and
Friedman (book website: http://www-stat.stanford.edu/~tibs/ElemStatLearn/) • Reference books:
– Pattern Classification by Duda, Hart and Stork– Pattern Recognition and Machine Learning by C.M. Bishop– Machine Learning by T. Mitchell– Introduction to Machine Learning by E. Alpaydin
• Some related journals / associations:– Machine Learning (Kluwer). – Journal of Machine Learning Research. – Journal of AI Research (JAIR). – Data Mining and Knowledge Discovery - An International Journal. – Journal of Experimental and Theoretical Artificial Intelligence (JETAI). – Evolutionary Computation. – Artificial Life. – Fuzzy Sets and Systems – IEEE Intelligent Systems (Formerly IEEE Expert) – IEEE Transactions on Knowledge and Data Engineering – IEEE Transactions on Pattern Analysis and Machine Intelligence – IEEE Transactions on Systems, Man and Cybernetics– Journal of AI Research – Journal of Intelligent Information Systems – Journal of the American Statistical Association – Journal of the Royal Statistical Society
27
References and Journals…– Pattern Recognition – Pattern Recognition Letters – Pattern Analysis and Applications. – Computational Intelligence . – Journal of Intelligent Systems . – Annals of Mathematics and Artificial Intelligence. – IDEAL, the online scientific journal library by Academic Press. – ECCAI (European Coordinating Committee on Artificial Intelligence). – AAAI (American Association for Artificial Intelligence). – IJCAI (International Joint Conferences on Artificial Intelligence, Inc.). – ACM (Association for Computing Machinery). – Association for Uncertainty in Artificial Intelligence. – ACM SIGAR– ACM SIGMOD– American Statistical Association. – Artificial Intelligence – Artificial Intelligence in Engineering – Artificial Intelligence in Medicine – Artificial Intelligence Review – Bioinformatics – Data and Knowledge Engineering – Evolutionary Computation
28
Some Conferences & Workshops• Congress on Evolutionary Computation
• European Conference on Machine Learning and Principles and Practice of Knowledge Discovery
• The ACM SIGKDD International Conference on Knowledge Discovery and Data Mining
• National Conference on Artificial Intelligence
• Genetic and Evolutionary Computation Conference
• International Conference on Machine Learning
• Conference on Autonomous Agents and Multiagent Systems
• European Symposium on Artificial Neural Networks Advances in Computational Intelligence and Learning
• Artificial and Ambient Intelligence
• Computational Intelligene in Biomedical Engineering
• IEEE International Symposium on Approximate Dynamic Programming and Reinforcement Learning
• International Joint Conference on Artificial Intelligence