3. Types of Machine Learning
• Supervised Learning
• Labels are provided, there is a strong learning signal.
• e.g. classification, regression.
• Semi-supervised Learning.
• Only part of the data have labels.
• e.g. a child growing up.
• Reinforcement learning.
• The learning signal is a (scalar) reward and may come with a delay.
• e.g. trying to learn to play chess, a mouse in a maze.
• Unsupervised learning
• There is no direct learning signal. We are simply trying to find structure in data.
• e.g. clustering, dimensionality reduction.
3
7. Generalization
• Consider the following regression problem:
• Predict the real value on the y-axis from the real value on the x-axis.
• You are given 6 examples: {Xi,Yi}.
• What is the y-value for a new query point X* ?
X*
7
11. • Ockham’s razor: prefer the simplest hypothesis
consistent with data.
Generalization
11
12. Generalization
Learning is concerned with accurate prediction
of future data, not accurate prediction of training data.
12
Question: Design an algorithm that is perfect at predicting training data.
13. Learning as Compression
• Imagine a game where Bob needs to send a dataset to Alice.
• They are allowed to meet once before they see the data.
• The agree on a precision level (quantization level).
• Bob learns a model (red line).
• Bob sends the model parameters
(offset and slant) only once
• For every datapoint, Bob sends
-distance along line (large number)
-orthogonal distance from line (small number)
(small numbers are cheaper to encode than
large numbers)
13
15. Classification: nearest neighbor
Example: Imagine you want to classify versus
Data: 100 monkey images and 200 human images with labels what is what.
Task: Here is a new image: monkey or human?
15
16. 1 nearest neighbor
Idea:
1. Find the picture in the database which is closest your query image.
2. Check its label.
3. Declare the class of your query image to be the same as that of the
closest picture.
query
closest image
16
18. Bayes Rule(s)
Riddle: Joe goes to the doctor and tells the doctor he has a stiff neck and a rash.
The doctor is worried about meningitis and performs a test that is 80% correct, that is,
for 80% of the people that have meningitis it will turn out positive. If 1 in 100,000 people
have meningitis in the population and 1 in 1000 people will test positive (sick or not sick)
what is the probability that Joe has meningitis?
Answer: Bayes Rule.
P(meningitis | positive test) = P(positive test | meningitis ) P(meningitis) / P(positive test)
= 0.8 * 0.00001 / 0.001 = 0.008 < 1%
18
20. Bayesian Networks & Graphical Models
• Main modeling tool for modern machine learning
• Reasoning over large collections of random variables with intricate relations
20