Machine Learning - University of Birmingham

advertisement
Machine Learning
Ata Kaban
The University of Birmingham
Machine Learning: The Module
•
•
•
•
•
•
•
•
•
What is Learning?
Decision trees
Instance-based learning
Kernel Machines
Probabilistic Models
Bayesian Learning
Learning Theory
Reinforcement Learning
Genetic Algorithms
Lectures & Tutorials
•
•
•
•
•
Lectures
Exercise classes (tutorials)
Exercise sheets given out at lectures
Solutions discussed during tutorials
Handouts are on the module’s web page:
http://www.cs.bham.ac.uk/~axk/ML_new.htm
Continuous Assessment
•
•
•
•
ML: 20% of your final mark
ML-EXTENDED: 40% of your final mark
20% will come out of class tests
For ML-Ext the other 20% will consist of
computational assignments
• It is in your interest to work through weekly
formative exercises. These will prepare you for the
class tests and for the final exam.
• Few bonus marks will be set aside for truly excellent
additional activity. I will take your ideas about what
form these should take.
Office hours
• The time for my weekly office hours is communicated on
my timetable (watch for possible changes):
http://www.cs.bham.ac.uk/~axk/timetable.html
• Location: UG32
• What office hours are and aren’t for
– Yes: ask me concrete questions to clarify something
that has not been clear to you from the lecture
– Yes: seek advice on your solutions to the given
exercises
– Yes: seek advice on further readings on related material
not covered in the lecture
– No: ask me to solve the exercises
– No: ask me to repeat a lecture
Literature
•
•
•
•
Machine Learning (Mitchell)
Reinforcement Learning … (Barto, Sutton)
Modelling the Web (Baldi, Smyth)
Support Vector Machines and Other Kernel-Based
Learning Methods (Cristianini, Shawe-Taylor)
• Artificial Intelligence … (Russell, Norvig)
• Artificial Intelligence (Rich, Knight)
• Artificial Intelligence (Winston)
Module Web Page
•
•
•
•
•
•
•
~axk
Syllabus
Handouts
Exercise sheets
Computer-based practical exercises
Links to ML resources on the web
Literature
• What is Learning?
• How can Learning be measured?
– Any change in the knowledge of a system that allows it
to perform better on subsequent tasks
– Knowledge. Hmmm... How should knowledge be
represented? We do not know how it is represented in
our own brains!
– Think for a moment about how knowledge might be
represented in a computer.
– If I told you what subjects would come up in the exam,
you might do very well. Would you do so well on
randomly chosen subjects from the syllabus? This
illustrates the difference between learning vs. the so
called ‘overfitting’ - we need to guard against the latter!
Ways humans learn things
•
•
•
•
•
•
•
•
…talking, walking, running…
Learning by mimicking, reading or being told facts
Tutoring
Being informed when one is correct
Experience
Feedback from the environment
Analogy
Comparing certain features of existing knowledge to new
problems
• Self-reflection
• Thinking things in ones own mind, deduction, discovery
Machine Learning
• Interdisciplinary field
–
–
–
–
–
–
–
–
–
Artificial intelligence
Bayesian methods
Computational complexity theory
Control theory
Information theory
Philosophy
Psychology and neurobiology
Statistics
…
Achievements of ML
• Computer programs that can:
– Recognize spoken words
– Predict recovery rates of pneumonia patients
– Detect fraudulent use of credit cards
– Drive autonomous vehicles
– Play games like backgammon – approaching
the human champion!
What is the Learning problem?
Learning = improving with experience at some task
– Improve at task T
– With respect to performance measure P
– Based on experience E
• Example: Learning to play checkers
– T: play checkers
– P: % of games won in world tournament
– E: opportunity to play against self
• Example: Learning to recognise faces
– T: recognise faces
– P: % of correct recognitions
– E: opportunity to make guesses and being told
what the truth was
• Example: Learning to find clusters in data
– T: finding clusters
– P: compactness of the groups detected
– E: opportunity to see a large set of data
Types of training experience
• Direct or indirect
• With a teacher or without a teacher
• An eternal problem: is the training
experience representative of the
performance goal? – it needs to be.
Forms of Machine Learning
• Supervised learning: uses a series of
examples with direct feedback
• Reinforcement learning: indirect
feedback, after many examples
• Unsupervised learning: no feedback
Download