1

Chapter 5
Learning Agent

 Introduction to Learning
 Machine Learning
 Phases of Machine Learning
 Applied Learning
 Learning Techniques
 Introduction to Neural Network

Compiled By Hamba A
 2
Introduction to Learning

 Learning can be described as normally a relatively permanent change that occurs in behavior as a result of
experience.
 It occurs in various regimes.
 For example, it is possible to learn to open a lock as a result of trial and error; possible to learn how to use a
word processor as a result of following particular instructions.
 The definition of Learning is as follow:
o "Learning denotes changes in a system that ... enables a system to do the same task more efficiently the next
time." By Herbert Simon.
o "Learning is constructing or modifying representations of what is being experienced." By Ryszard Michalski.
o "Learning is making useful changes in our minds." By Marvin Minsky.

Compiled By Hamba A
 3
What is Machine Learning

 The goal of machine learning is to build computer systems that can learn from their
experience and adapt to their environments.
 Obviously, learning is an important aspect or component of intelligence.
 There are both theoretical and practical reasons to support such a claim.
 In simple words, we can say that machine learning is the competency of the software to
perform a single or series of tasks intelligently without being programmed for those
activities.
 This is part of Artificial Intelligence.
 Normally, the software behaves the way the programmer programmed it; while machine
learning is going one step further by making the software capable of accomplishing
intended tasks by using statistical analysis and predictive analytics techniques.

Compiled By Hamba A
 4
Why do we want Machine Learning?

 One response to the idea of AI is to say that computers can not think because they only do
what their programmers tell them to do.
 However, it is not always easy to tell what a particular program will do, but given the
same inputs and conditions it will always produce the same outputs.
 Machine learning is a prerequisite for any mature programme of artificial intelligence.

Compiled By Hamba A
 5
Three Phases of Machine Learning

o Training: a training set of examples of correct behavior is analyzed and some representation of the newly
learnt knowledge is stored.
 This is often some form of rules.
o Validation: the rules are checked and, if necessary, additional training is given.
 Sometimes additional test data are used, but instead of using a human to validate the rules, some other
automatic knowledge based component may be used.
 The role of tester is often called the critic.
o Application: the rules are used in responding to some new situations.
 These phases may not be distinct.
 For example, there may not be an explicit validation phase; instead, the learning algorithm guarantees some
form of correctness.
 Also in some circumstances, systems learn "on the job", that is, the training and application phases overlap.

Compiled By Hamba A
 6
Learning Techniques

Rote Learning
 In this kind of learning there is no prior knowledge.
 When a computer stores a piece of data, it is performing an elementary form of learning.
 This act of storage presumably allows the program to perform better in the future.
 Examples of correct behavior are stored and when a new situation arises it is matched
with the learnt examples.
 The values are stored so that they are not re-computed later.
 One of the earliest game playing programs is checkers program.
 This program learned to play checkers well enough to beat its creator/designer.

Compiled By Hamba A
 7
Cont…

Deductive Learning
 Deductive learning works on existing facts and knowledge and deduces new knowledge from the old.
 This is best illustrated by giving an example.
 For example, assume:
A=B
B=C
Then we can deduce with much confidence that: C = A
 Arguably, deductive learning does not generate "new" knowledge at all, it simply memorizes the logical consequences of
what is known already.
 This implies that virtually all mathematical research would not be classified as learning "new" things.
 However, regardless of whether this is termed as new knowledge or not, it certainly makes the reasoning system more
efficient.
Compiled By Hamba A
 8
Cont…

Inductive Learning
 Inductive learning takes examples and generalizes rather than starting with existing knowledge.
 For example, having seen many cats, all of which have tails, one might conclude that all cats have tails.
 This is an unsound step of reasoning but it would be impossible to function without using induction to some extent.
 In many areas it is an explicit assumption.
 There is scope of error in inductive reasoning, but still it is a useful technique that has been used as the basis of
several successful systems.
 One major subclass of inductive learning is concept learning.
 This takes examples of a concept and tries to build a general description of the concept.
 Very often, the examples sare described using attribute-value pairs

Compiled By Hamba A
 9
Applied Learning

Solving Real World Problems with Learning

 We do not yet know how to make computers learn nearly as well as people learn.
 However, algorithms have been developed that are effective for certain types of learning
tasks, and many significant commercial applications have begun to appear.
 For problems such as speech recognition, algorithms based on machine learning
outperform all other approaches that have been attempted to date.
 In other emergent fields like computer vision and data mining, machine learning
algorithms are being used to recognize faces and to extract valuable information and
knowledge from large commercial databases respectively.

Compiled By Hamba A
 10
Cont…

 Some of the applications that use learning algorithms include:

o Spoken digits and word recognition
o Handwriting recognition
o Driving autonomous vehicles
o Path finders
o Intelligent homes
o Intrusion detectors
o Intelligent refrigerators, tvs, vacuum cleaners
o Computer games
o Humanoid robotics
Compiled By Hamba A
 11
General Model of Learning Agent (Pattern Recognition)

 Any given learning problem is primarily composed of three things:

o Input
o Processing unit
o Output
 The input is composed of examples that help the learner learn the underlying problem
concept.
 The processing unit is the learning agent in our focus of study.

Compiled By Hamba A
 12
Symbol Based Learning

 Ours is a world of symbols.

 We use symbolic interpretations to understand the world around us.
 For instance, if we saw a ship, and were to tell a friend about its size, we will not say that
we saw a 254.756 meters long ship, instead we’d say that we saw a ‘huge’ ship about the
size of ‘Eiffel tower’.
 And our friend would understand the relationship between the size of the ship and its
hugeness with the analogies of the symbolic information associated with the two words
used: ‘huge’ and ‘Eiffel tower’.
 Similarly, the techniques we are to learn now use symbols to represent knowledge and
information.

Compiled By Hamba A
 13
Introduction to Neural Network

 In the area of Computer Science a Neural Network is a system of programs and data
structures that approximates the operation of the human brain.
 A neural network usually involves a large number of processors operating in parallel, each
with its own small sphere of knowledge and access to data in its local memory.
 Typically, a neural network is initially "trained" or fed large amounts of data and rules
about data relationships (for example, "A grandfather is older than a person's father").
 A program can then tell the network how to behave in response to an external stimulus
(for example, to input from a computer user who is interacting with the network) or can
initiate activity on its own (within the limits of its access to the external world).

Compiled By Hamba A
 14
Cont…

 In making determinations, neural networks use several principles, including gradient based training, fuzzy logic, genetic
algorithms, and Bayesian methods.
 Neural networks are sometimes described in terms of knowledge layers, with, in general, more complex networks having
deeper layers.
 Current applications of neural networks include:
o Oil exploration data analysis.
o Weather prediction.
o The interpretation of nucleotide sequences in biology labs.
o The exploration of models of thinking and consciousness.
 The term neural network was traditionally used to refer to a network or circuit of biological neurons.
 The modern usage of the term often refers to artificial neural networks, which are composed of artificial neurons or nodes.

Compiled By Hamba A
 15
Cont…

 Thus the term has two distinct usages:

o Biological neural networks :are made up of real biological neurons that are connected or functionally related
in a nervous system.
 In the field of neuroscience, they are often identified as groups of neurons that perform a specific physiological
function in laboratory analysis.
o Artificial neural networks: are composed of interconnecting artificial neurons (programming constructs that
mimic the properties of biological neurons).
 Artificial neural networks may either be used to gain an understanding of biological neural networks, or for
solving artificial intelligence problems without necessarily creating a model of a real biological system.
 The real, biological nervous system is highly complex.
 Artificial neural network algorithms attempt to abstract this complexity and focus on what may hypothetically
matter most from an information processing point of view.
Compiled By Hamba A
 16
Cont…

Comparison between Human and Computer

Compiled By Hamba A
 17
Directions and Classifications of Neural Network

 The Neural Networks can roughly be classified into four different types of orientations which includes its applications to real-world
problems.
o In Cognitive science/Artificial intelligence, the interest is in modeling intelligent behavior.
o The interest in neural networks is mostly to overcome the problems and pitfalls of classical and symbolic methods of modeling
intelligence.
o Neurobiological modeling has the goal to develop models of biological neurons.
 Here, the exact properties of the neurons play an essential role.
 In most of these models there is a level of activation of an individual neuron.
o Scientific modeling uses neural networks as modeling tools.
 In physics, psychology, and sociology neural networks have been successfully applied.
o Computer science views neural networks as an interesting class of algorithms that has properties like noise and fault tolerance, and
generalization ability– that make them suited for application to real-world problems.

Compiled By Hamba A
 18
Applications of Neural Network

 Neural Networks are now applied in many areas of science.

 Here are some few examples:
o Optimization: Neural networks have been applied to almost many kind of optimization problem.
 Conversely, neural network learning can often be conceived as an optimization problem in which it will minimize a kind of
error function.
o Control: Many complex control problems have been solved by neural networks.
 They are especially popular for robot control. Such as autonomous robots - like humanoids, that have to operate in real world
environments that are characterized by higher level rapid change.
 Since biological neural networks have evolved for precisely these kinds of conditions, they are well suited for such types of
tasks.
 Also, because in the real world generalization is crucial, neural networks is often the tool of choice for systems, in particular
robots, having interacted with physical environments.
Compiled By Hamba A
 19
Cont…

o Signal Processing: Neural networks have been used to distinguish mines from rocks using sonar signals, to detect
sun eruptions, and to process speech signals.
 Speech processing techniques and statistical approaches involving hidden models are sometimes combined.
o Pattern Recognition: Neural networks have been widely used for pattern recognition purposes, from face
recognition, to recognition of tumors in various types of scans, to identification of plastic explosives in luggage of
aircraft passengers (which yield a particular gamma radiation patterns when subjected to a stream of thermal
neurons), to recognition of hand-written zip-codes.
o Stock market prediction: The dream of every mathematician is to develop methods for predicting the
development of stock prices.
 Neural networks, in combination with other methods, are often used in this area. However, it is an open question
whether they have been really successful.

Compiled By Hamba A
 20

Thank You!!!.

Compiled By Hamba A