1
Background
1.1
Why Machines should learn
1.1.1
AI and machine learning
The ability of the human mind has marvelled scientists and philosophers alike for many
centuries and has continuously inspired our quest for intelligent machines. The famous
Dartmout
4
Probabilistic regression and maximum
likelihood
4.1
Probabilistic motion models
We are now ready for formalize supervised learning and to demonstrate this with
a probabilistic motion model. In supervised learning we consider training data that
consist o
5
Probabilistic reasoning and Bayes
ltering
5.1
Multivariate generative models and probabilistic
reasoning
5.1.1
F
Graphical models
A
We have so far only considered very simple hypothesis appropriate for the low
dimensional data given in the above example
3
Probability theory and motion/sensor
models
A major milestone for modern approaches to machine learning and robotics is the
acknowledgement of our limited knowledge about the world and the unreliability of
sensors and actuators. It is then only natural
9
Reinforcement Learning
As discussed above, a basic form of supervised learning is function approximation,
relating input vectors to output vectors, or, more generally, nding density functions
p(y, x) from examples (x(i) , y(i) ). However, in many applic
2
Sensing, acting and control
This is a busy chapter where we review some fundamental techniques for robotics.
We will learn how to acquire images from a webcam and to lter the image in order
to look for specic items. We will then explain how to use a Leg
6
Umbiquitous Learning Machines
In this chapter we discuss some methods that are widely used for machine learning applications. In the previous discussions we always assumed very specialized hypothesis
functions for particular problems. However, nding an
Machine Learning - Assignment 7
FacultyofComputerScience,DalhousieUniversity
6050UniversityAve,Halifax,NovaScotia,Canada,B3H1W5
tt @cs.dal.ca
Abstract
This document presents how we used Machine Learning, Convolution and Edge Filters to solve the
problem o