Beruflich Dokumente
Kultur Dokumente
Ning Xiong
Mälardalen University
1
Agenda
• What is machine learning?
2
Part 1: What is Machine learning
3
Inspiration from Human
4
What is Machine Learning
7
Learning Problems
The nature of experience causes different
kinds of learning problems
• Supervised Learning
• Unsupervised Learning
• Reinforcement Learning
8
Supervised Learning
• A teacher available to give desired response
• Direct experience represented by a set of input-
output examples (xi,yi)
• minimize the error between the actual response of
the learner and the desired response
desired
state x response
Environment Teacher
actual
+
response
S
Learning -
system
error signal 9
Unsupervised Learning
10
Unsupervised Learning: Clustering
Objects similar in their attributes are clustered in
the same group
Cluster 1
Outlier
x
Cluster 2
11
Reinforcement Learning
• Aim to obtain optimal action strategy without teacher
• learning through interaction with the environment
• exploration of states and actions
• Indirect experience in form of delayed reward signal
(temporal credit assignment problem)
• goal: maximize accumulated rewards for the long term
Environment
Delayed reward/penalty signal
state
Learning
system
12
action
Credit Assignment Problem
Assigning credit or blame for the overall outcomes
(delayed reward) to each of the internal decisions
made by the learning machine which contributed to
the overall outcomes
s0 s1 s2 s3 Credit
a1 a2 a3
13
Overview of learning techniques
14
Linear Regression (Supervised)
Build a linear function to model the relation of dependent
variable Y with its relevant variables
å (Y )
n n
å
2
min i - Yˆi = e
ˆ 2
i
i =1 i =1
15
Logistic Regression (Supervised)
• Build a linear surface to separate training examples into two classes
β0 + β1 X
e
• Convert the linear surface into class probability: π= β0 + β1 X
1+ e
• Finding parameters to minimize the misclassifications, i.e. log loss function
+
+
+
+ -
- -
16
Support Vector Machine (supervised)
17
Problem with Linear SVM
=-1
=+1
Φ: x → φ(x)
19
Artificial Neural Network
To directly represent complex nonlinear relation, we can build
an artificial neural network consisting of many units.
Output data
Input data
20
Learning of ANN
22
Reading Assignment
23