Scribd
Upload
7 views

Unit 3

Uploaded by

Why You
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
7 views

Unit 3

Uploaded by

Why You
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 16

Introduction to Machine

Learning
Machine Learning
• Machine learning is an application of artificial intelligence that
involves algorithms and data that automatically analyze and make
decision by itself without human intervention.
Training
• In most learning problems, the task is to learn to classify inputs
according to a finite (or sometimes infinite) set of classifications.
• A learning system is provided with a set of training data, which have
been classified by hand. The system then attempts to learn from
these training data how to classify the same data (usually a relatively
easy task) and also how to classify new data that it has not seen.
Rote Learning
• The simplest way for a computer to learn from experience is simply to
learn by rote.
• Training involves storing each piece of training data and its
classification.
• A new item of data is classified by looking to see if it is stored in
memory.
• If it is, then the classification that was stored with that item is
returned. Otherwise, the method fails.
Learning Concepts
• Concept learning involves determining a mapping from a set of input
variables to a Boolean value.
• These methods are known as inductive-learning methods.
• These methods are based on the principle that if a function is found
that correctly maps a large set of training data to classifications, then
it will also correctly map unseen data.
Example
• The learning task is to determine whether driving in a particular
manner in particular road conditions is safe or not.
• A hypothesis is a vector of values for these attributes. A possible
hypothesis is,
• h1 = <slow, wind, 30ft, 0, evening, cold>

• We also want to represent in a hypothesis that we do not care what


value an attribute takes. This is represented by “?”, as in the following
hypothesis:
h2 = <fast, rain, 10ft, 2, ?, ?>
• Negative training example
• In other cases, we need to represent a hypothesis that no value of a
particular attribute will provide a positive example. We write this as
“∅”, as in the following hypothesis:
• h3 = <fast, rain, 10ft, 2, ∅, ∅>
General-to-Specific Ordering
• hg = <?, ?, ?, ?, ?, ?>
• hs = <∅, ∅, ∅, ∅, ∅, ∅>

• hg is the hypothesis that it is safe to drive regardless of the


conditions—this is the most general hypothesis.
• hs is the most specific hypothesis, which states that it is never safe to
drive, under any circumstances.
A Simple Learning Algorithm
• The algorithm uses the general-to-specific ordering of hypotheses to
search the hypothesis space for a suitable hypothesis.
• The method is as follows:
• Start with the most specific hypothesis.
• For each positive training example, determine whether each attribute in the
example is matched by the current hypothesis.
• If it is not, replace the attributes in the hypothesis with the next more general value that
does match.
Example:
<slow, wind, 30ft, 0, evening, cold>
<slow, rain, 20ft, 0, evening, warm>
<slow, snow, 30ft, 0, afternoon, cold>
Version Spaces

• Given a set of training examples (positive and negative), the set of


hypotheses that correctly map each of the training examples to its
classification is called the version space.
Candidate Elimination
• To find a set of hypothesis that is consistent with all the training
example.
• The method operates as follows:
• Two sets are maintained of hypotheses, hs and hg.
• hs is initialized as {<∅, ∅, ∅, ∅, ∅, ∅, ∅, ∅>} and
hg is initialized as {<?, ?, ?, ?, ?, ?, ?, ?>}.
• When a positive training example is encountered,
• it is compared with the hypotheses contained in hg. If any of these hypotheses does not
match the training example, it is removed from hg.
• The positive training data are then compared with the hypotheses contained in hs. If one
of these hypotheses does not match the training data, it is replaced by the set of slightly
more general hypotheses that are consistent with the data, and such that there is at
least one hypothesis in hg that is more general.
• This method is applied in reverse for negative training data.
Inductive Bias
• Inductive bias refers to the restrictions that are imposed by the
assumptions made in the learning method.
• The inductive bias of the candidate elimination algorithm is that it is
only able to classify a new piece of data if all the hypotheses
contained within its version space give the data the same
classification. Hence, the inductive bias does impose a limitation on
the learning method.
Decision-Tree Induction
• Solution of ID3 algorithm.
• Refer the class notes for solved problem on decision tree.
The Problem of Overfitting
• Overfitting usually occurs when there is noise in the training data, or
when the training data do not adequately represent the entire space
of possible data.
• If the training data do not adequately and accurately represent the
entire data set, the decision tree that is learned from it may not
match unseen data.
Reinforcement Learning
• A system that uses reinforcement learning is given a positive
reinforcement when it performs correctly and a negative
reinforcement when it performs incorrectly.
• The information that is provided to the learning system when it
performs its task correctly does not tell it why or how it performed it
correctly, simply that it did.

You might also like