Transcript Machine Learning

1
‫يادگيري ماشين‬
Machine
Learning
Lecturer: A. Rabiee
[email protected]
Rabiee.iauda.ac.ir
2
‫منابع و مراجع‬
Main Reference:
- Mitchell, T. M. (1997). Machine learning. WCB.
Other References:
-
Haykin, S. S. (2009). Neural networks and learning machines (Vol. 3). Upper
Saddle River: Pearson Education.
Mitchell, T. M. (1999). Machine learning and data mining. Communications of
the ACM, 42(11), 30-36.
Anderson, J. R. (1986). Machine learning: An artificial intelligence
approach(Vol. 2). R. S. Michalski, J. G. Carbonell, & T. M. Mitchell (Eds.).
Morgan Kaufmann.
Mitchell, M. (1998). An introduction to genetic algorithms. MIT press.
Witten, I. H., & Frank, E. (2011). Data Mining: Practical machine learning
tools and techniques. Morgan Kaufmann.
Hagan, M. T., Demuth, H. B., & Beale, M. H. (1996). Neural network design(pp.
2-14). Boston: Pws Pub..
Kecman, V. (2001). Learning and soft computing: support vector machines,
neural networks, and fuzzy logic models. MIT press.
3
- ….
Course Outline
•
•
•
•
•
Chapter 1: Introduction
Chapter 3: Decision tree learning
Chapter 4: Artificial Neural Networks
Chapter 9: Genetic Algorithms
Chapter 13: Reinforcement Learning
‫ارزشيابي درس‬
Final Exam:
Mini Projects (2 to 4):
Final Project + Presentation:
Paper (optional):
50
20
30
+15
4
Chapter 1:
Introduction to Machine Learning
Table of Contents
•
•
•
•
Definition & Examples
Applications
Why ML?
ML Problems
Definition (Mitchell 1997)
• Machine Learning
– Learn from past experiences
– Improve the performances of intelligent
programs
• Definition
– A computer program is said to learn from
experience E with respect to some class
of tasks T and performance measure P, if
its performance at the tasks improves
with the experiences
Examples
• Text Classification (or spam classification)
– Task T
• Assigning texts to a set of predefined
categories
– Performance measure P
• Precision of each category
– Training experiences E (Dataset)
• A dataset of texts with their corresponding
categories
• How about Disease Diagnosis?
• How about Chess Playing?
Two phases
• Two phases of a learning process:
– Train
– Test
Example:
Classification of texts based on content
Classified text files
Text file 1
trade
Text file 2
ship
…
…
Phase 1: train
Training
Phase 2: test
New text file
Text classifier
class
Example:
Heart disease diagnosis
Database of medical records
Patient 1’s data
Absence
Patient 2’s data
Presence
…
…
Training
New patient’s
data
Disease classifier
Presence or
absence
Example: Chess Playing
Games played:
Game 1’s move list Win
Game 2’s move list Lose
…
…
Training
New matrix
representing the
current board
Strategy of
Searching and
Evaluating
Best move
Machine Learning Problems
Clustering: Grouping similar instances
Dimension Reduction: Image Compression
Regression: Tuning the angle of a robot arm
Application: Image Categorization
(two phases)
Training
Training
Images
Image
Features
Training
Labels
Classifier
Training
Trained
Classifier
Testing
Image
Features
Test Image
Trained
Classifier
Prediction
Outdoor
Feature Extraction
Training
Training
Images
Image
Features
Training
Labels
Classifier
Training
Trained
Classifier
Example: Boundary Detection
• Is this a boundary?
Training Algorithm
Training
Training
Images
Image
Features
The main aim of this course
Training
Labels
Classifier
Training
Trained
Classifier
Classifier Training
Example: A 2-class classifier
• Given some set of
features with
corresponding labels,
learn a function to
predict the labels
from the features
• Example: Credit scoring
Discriminant (model):
IF income > θ1 AND savings > θ2
THEN
low-risk
ELSE
high-risk
Different Learning Algorithms
•
•
•
•
•
•
•
•
Decision Tree Learning
Neural networks
Naïve Bayes
Genetic Algorithm
K-nearest neighbor (clustering)
Reinforcement Learning
Support Vector Machine (SVM)
…
Note
The decision to use machine learning
is more important than the choice of a
particular learning method.
Why Machine Learning Is Possible?
• Mass Storage
– More data available
• Higher Performance of Computer
– Larger memory in handling the data
– Greater computational power for
calculating and even online learning
Advantages
• Alleviate Knowledge Acquisition
Bottleneck
– Does not require knowledge engineers
– Scalable in constructing knowledge base
• Adaptive
– Adaptive to the changing conditions
– Easy in migrating to new domains
Success of Machine Learning
• Almost All the Learning Algorithms
– Text classification (Dumais et al. 1998)
– Gene or protein classification optionally
with feature engineering (Bhaskar et al.
2006)
• Reinforcement Learning
– Backgammon (Tesauro 1995)
• Learning of Sequence Labeling
– Speech recognition (Lee 1989)
– Part-of-speech tagging (Church 1988)
Datasets
• UCI Repository: http://www.ics.uci.edu/~mlearn/MLRepository.html
• UCI KDD Archive:
http://kdd.ics.uci.edu/summary.data.application.html
•
•
•
•
•
•
•
Statlib: http://lib.stat.cmu.edu/
Delve: http://www.cs.utoronto.ca/~delve/
US government free data: data.gov
US government free data (California): data.ca.gov
…. for other states and the UK data.gov.uk, as well
Stock market softwares
Weather forecasting websites
• Reuters: data set for text classification
• ….
What I will Talk about
• Machine Learning Methods
– Simple methods
– Effective methods (state of the art)
• Method Details
– Ideas
– Assumptions
– Intuitive interpretations
What I won’t Talk about
• Machine Learning Methods
– Classical, but complex and not effective
methods (e.g., complex neural networks)
– Methods not widely used
• Method Details
– Theoretical justification
– Theorem proving
What You will Learn
• Machine Learning Basics
– Methods
– Data
– Assumptions
– Ideas
• Others
– Problem solving techniques
– Extensive knowledge of modern
techniques