Download Slides on Artificial Intelligence - Machine Learning | CSCI 5622 and more Study notes Computer Science in PDF only on Docsity!
Greg Grudic Machine Learning 1
Machine Learning
CSCI 5622
Fall 2004
Greg Grudic
Greg Grudic Machine Learning 2
Admin Stuff 1
http://www.cs.colorado.edu/~grudic/teaching/CSCI5622_ 04
Course URL:
Email: [email protected]
Phone: 303-492-
Office Hours: Tuesday and Wednesday 10:00 to 11:
Office: ECOT 525
Instructor: Professor Greg Grudic
Location: Wednesdays 3:00pm-5:30pm ECCR 108
Admin Stuff 2
- Course Textbook: The Elements of Statistical Learning, by Hastie, Tibshirani, Friedman
- Grading:
- Homework 45%
- Project 25%
- Class participation 10%
- Midterm exam 20%
- Course workload outside of class?
Greg Grudic Machine Learning 4
Admin Stuff 3
- Homework
- ā 3 coding assignments (algorithm implementation), each worth 15%.I STRONGLY encourage you to use matlab.
- However, you can use any language you want, as long as you make sure that I can run your programs!ā Currently: Java (Sun), C++ (unix), VC++
- Unless you have a very good excuse, each day your assignment is late will take 1% off whatthe assignment is worth (up to 15% per assignment).
- Midterm (Nov 17)
- Test basic knowledge of ML. It will consist of general questions on the machine learningalgorithms covered to date. You will not be required to derive algorithms or prove theorems.
- Project ā Implementation of an algorithm (not covered in class) from a recent research paper. You will (Pick by Nov 3. Due Dec 10.) be required to summarize the algorithm and evaluate it empirically.
- Class participation ā This consists of showing up for class and asking questions. Questions by email count as class participation!
Greg Grudic Machine Learning 5
Goal of the Course
- A fundamental understanding of the basic
concepts behind machine learning
- What does it mean for a machine to learn?
- You will be able to read current research
papers in ML after completing this course.
- Why is machine learning important?
- ML algorithms are at the heart of many modern computer applications
Where can ML be found?
- Marketing
- Who should a company target for advertising?
- Profiling
- Is passenger 57 likely to hijack a plane?
- User interfaces
- Making it easier to interact with a PC by anticipating what I am doing.
- Document characterization
- Searching the web for things of interest.
- Bioinformatics
- Human genome projectā¢ Which gene is responsible for the cancer that runs in my family?
- Data mining
- āData doubles every yearā, Dunham 2002
- ML algorithms are used to make sense of this data
- Economics, medical diagnosis, robotics, computer vision, manufacturing,inventory control, elevator operationā¦.
Greg Grudic Machine Learning 10
Elements of AI
Learning
Representation Reasoning
Greg Grudic Machine Learning 11
(My) Elements of AI
Representation
Learning
Why does learning encompass reasoning?
- How can I reason rationally about a world I know nothing about?
- How can an gain knowledge about a world without sampling it and learning from those samples?
- Fundamental lesson of AI (learned in the 1980ās):
- It is not possible to hand code knowledge about anything but the most trivial problem domains!
- Expert Systems: largely failed because an expert (e.g. doctor) doesnāt know how to formalize (code) what makes her an expert!
- For Example: Iām an expert on chairs but I canāt (and no one can!) write a program that identifies chairs in an image - ML techniques can!
Greg Grudic Machine Learning 13
What is Machine Learning?
- āThe goal of machine learning is to build
computer systems that can adapt and learn
from their experience.ā
- Tom Dietterich
- What does this mean?
- When are ML algorithms NOT needed?
Greg Grudic Machine Learning 14
A Generic System
ā¦ System ā¦
x 1
x 2
x N
y 1
y 2
1 2 yM
h , h , ..., hK
x = ( x 1 (^) , x 2 ,..., xN ) h = ( h 1 (^) , h 2 ,..., hK ) y = ( y 1 (^) , y 2 ,..., yK )
Input Variables:
Hidden Variables:
Output Variables:
Another Definition of Machine
Learning
- Machine Learning algorithms discover the relationships between the variables of a system (input, output and hidden) from direct samples of the system
- These algorithms originate form many fields:
- Statistics, mathematics, theoretical computer science, physics, neuroscience, etc
Greg Grudic Machine Learning 19
Two Types of Supervised Learning
- Classification
- Model output is a prediction that the input belongs to some class
- If the input is an image, the output might be chair, face, dog, boat,ā¦ etc.
- Regression
- The output has infinitely many values
- If the input is stock features, the output could be a prediction of tomorrowās stock price
y ā { c 1 (^) , c 2 ,..., cN }
y āā
Greg Grudic Machine Learning 20
Learning Classification Models
- Collect Training data
- Build Model: happy = M (feature space)
- Make a prediction
High Dimensional Feature Space
Learning Regression Models
- Collect Training data
- Build Model: stock value = M ( feature space)
- Make a prediction
Feature Space
Stock Value _ _
_ ** _ (^) *_ _
_ ** _
__*
__*
_ ** _ (^) *_ _
_ _
_ _
__*
__*
_ ** _ (^) *_ _
_ _*
__*
Greg Grudic Machine Learning 22
Examples of Supervised Learning
f ( x )
x
- Credit risk assessment : Properties of customer and proposed purchase : Approve purchase or not
f ( x )
x
- Disease diagnosis : Properties of patient (symptoms, lab tests) : Disease (or maybe, recommended therapy)
Greg Grudic Machine Learning 23
Examples of Supervised Learning
(continued)
f ( x )
x
- Face recognition : Image of person's face : Name of the person
f ( x )
x
- Automated Vehicle Driving : Image of the road : Throttle, break, and steering commands
- Situations where humans can perform the
task but can't describe how they do it
Appropriate Applications for
Supervised Learning
f ( x )
x
- Situations where there is no human expert : Bond graph for a new molecule : Predicted binding strength to AIDS protease molecule
f ( x )
x : Bitmap picture of hand-written character : Ascii code of the character
Greg Grudic Machine Learning 28
- Premise: Data is expensive to collect (e.g most experiments in biology)
- Goal: want to get the best possible model with the smallest dataset
- Active learning starts with a classifier and asks the following questions
- Where in the feature space do I need to sample next to improve my classifier the most.
High Dimensional Feature Space
Active Learning
Greg Grudic Machine Learning 29
Reinforcement Learning (RL)
Autonomous agent learns to act āoptimallyā
without human intervention
- Agent learns by stochastically interacting
with its environment, getting infrequent
rewards
- Goal: maximize infrequent reward
Reinforcement Learning
- Addresses the temporal credit assignment
problem:
- Delayed reward (HARD problem!)
- Successful RL applications:
- TD gammon (Tesauro)
- Packing containers (Moore)
- Elevator dispatch (Crites and Barto)
Greg Grudic Machine Learning 31
RL in Robotics
Goal
Robot
Obstacle
Static Navigational Feature -Hit an obstacle: get a negative reward -Reach goal: get a positive reward -Reach goal faster: get a bigger positive reward
Greg Grudic Machine Learning 32
A (simple?) Robotics Problem
Even Simple Robotic tasks are
difficult to Program
Greg Grudic Machine Learning 37
Unsupervised Learning
- Studies how input patterns can be
represented to reflect the statistical structure
of the overall collection of input patterns
- No outputs are used (unlike supervised
learning and reinforcement learning)
- unsupervised learner brings to bear prior
biases as to what aspects of the structure of
the input should be captured in the output.
Greg Grudic Machine Learning 38
Unsupervised Learning Example
High Dimensional Feature Space
- Collect Training data (e.g. consumer info)
- Build Model: things that a similar = M (feature space)
Locally Linear Embedding (LLE)
Sam T. Roweis^ From: www.cs.toronto.edu/~roweis/[email protected] Lawrence K. Sauland www.research.att.com/~lsaul/[email protected]
Conclusion
- How many people have I scared away?
- Class Schedule?
- Whoās in my class? Please email me the
survey on the class web page by the end of
the week (a text file please).
