Lecture 10 - Intro to Machine Learning Flashcards
Why is machine learning a big thing?
Extremely difficult for computers to do very simple human tasks
4 year old outperforms computer every time at distinguishing dogs from cats
What is machine learning?
Branch of AI concerned with making computers learn from experience/data
What was Lady Lovelace’s Objection?
(sure we’ve had this before but meh)
“Computers cannot be intelligent because they only do what the programmer tells them to”
What is Turing’s response to Lovelace’s objection?
We can program computers to learn and think.
What did Arthur Samuel build in 1950s? Why is it important to AI?
Checkers/Draughts playing program that improved performance the more it played and reached state championship level
Important because one of the first machine learning programs
When were the first neural network systems?
Mid 1950s-Early 1960s
What are the Key Elements of a Machine Learner?
Task
Performance Measure
Set of Examples
Representation format for examples
(Models/Algorithms for learning)
How can learning be represented as a search?
Search of the space of possible representations of a model that maximises the performance measure:
y = f (x, W)
where f = model structure, W = model parameters
What are the 4 main types of learning?
Learning to Classify - e.g. learning to categorise animals based on pictures
Learning to Predict Numeric Values (regression/approximation)
Learning to form groups - making unclassified data make sense in some way
Learning what to do next - supervised learning - e.g. playing checkers
What type of learning is decision tree induction?
Learning to Classify
What type of learning do Neural Networks undertake?
Learning to Predict or Classify
If two events are statistically independent (i.e. neither one helps you predict the other) then
P(X ^ Y) = ?
P(X) x P(Y)
If P(X ^ Y) != P(X) x P(Y) then we can conclude that
Knowledge of one event can help you predict the other
P(X|Y) can also be represented:
P(X^Y) / P(Y)