UNIT V

Learning
What is learning?
 Herbert Simon, 1983
Learning denotes changes in a system that enables a system to do the same task more
efficiently the next time.
 Marvin Minsky, 1986
Learning is making useful changes in the workings of our minds.
 Ryszard Michalski, 1986
Learning is constructing or modifying representations of what is being experienced.
 Mitchell, 1997
A computer program is said to learn from experience E with respect to some class of tasks
T and performance measure P, if its performance at tasks in T, as measured by P, improves
with experience E.

ARCHITECTURE OF LEARNING AGENT

 Performance Element: The Performance Element is the agent itself that acts in the
world. It takes in percepts and decides on external actions.
 Learning Element: It responsible for making improvements, takes knowledge about
performance element and some feedback, determines how to modify performance
element.
 Critic: Tells the Learning Element how agent is doing (success or failure) by comparing
with a fixed standard of performance.
 Problem Generator: Suggests problems or actions that will generate new examples or
experiences that will aid in training the system further.

Automated Taxi on city roads

 Performance Element: Consists of knowledge and procedures for driving actions.
e.g., turning, accelerating, braking are performance element on roads.
 Learning Element: Formulates goals.
e.g., learn rules for braking, accelerating, learn geography of the city.
 Critic: Observes world and passes information to learning element.
e.g. , quick right turn across three lanes of traffic, observe reaction of other drivers.
 Problem Generator: Try south city road .

PARADIGMS OF MACHINE LEARNING

 Rote Learning: Learning by memorization; One-to-one mapping from inputs to stored
representation; Association-based storage and retrieval.
 Induction: Learning from examples; A form of supervised learning, uses specific
examples to reach general conclusions; Concepts are learned from sets of labeled
instances.
 Clustering: Discovering similar group; Unsupervised, Inductive learning in which
natural classes are found for data instances, as well as ways of classifying them.
 Analogy: Determine correspondence between two different representations that come
from Inductive learning in which a system transfers knowledge from one database into
another database of a different domain.
 Discovery: Learning without the help from a teacher; Learning is both inductive and
deductive. It is deductive if it proves theorems and discovers concepts about those
theorems. It is inductive when it raises conjectures (guess). It is unsupervised, specific
goal not given.
 Genetic Algorithms: Inspired by natural evolution; In the natural world, the organisms
that are poorly suited for an environment die off, while those well-suited for it prosper.
Genetic algorithms search the space of individuals for good candidates. The "goodness"
of an individual is measured by some fitness function. Search takes place in parallel, with
many individuals in each generation.
 Reinforcement: Learning from feedback (+ve or -ve reward) given at end of a
sequence of steps. Unlike supervised learning, the reinforcement learning takes place in
an environment where the agent cannot directly compare the results of its action to
a desired result. Instead, it is given some reward or punishment that relates to its actions.
It may win or lose a game, or be told it has made a good move or a poor one. The job of
reinforcement learning is to find a successful function using these rewards.
ROTE LEARNING
 Rote learning technique avoids understanding the inner complexities but focuses on
memorizing the material so that it can be recalled by the learner exactly the way it was
read or heard.
 Learning by Memorization which avoids understanding the inner complexities the
subject that is being learned; Rote learning instead focuses on memorizing the material so
that it can be recalled by the learner exactly the way it was read or heard.
 Learning something by Repeating over and over and over again; saying the same
thing and trying to remember how to say it; it does not help us to understand; it
helps us to remember, like we learn a poem, or a song, or something like that by rote
learning.

LEARNING BY TAKING ADIVCE

 Two basic approach to advice taking:

1) Taking high-level and abstract advice, then converting into rules.
2) Developing sophisticated modules.

Taking High-Level And Abstract Advice, Then Converting Into Rules:

 Aspects of advice taking are automated by steps below:
1) Request: This can be simple question enquiring about either general or complicated
advice by identifying shortcomings in the knowledge and asking for a remedy.
2) Interpret: The advice is translated into an internal representation.
3) Operationalize: It is quite possible that translated advice may still not be usable; so this
stage aims to provide a representation that can be used by performance element.
4) Integrate : When this knowledge is added to the knowledge, care needs to be taken to
ensure that –ve side effects, such as redundancy and contradictions are avoided.
5) Evaluate: the system must evaluate new knowledge for errors, contradictions etc.
SOPHISTICATED MODULES
 Second approach for developing sophisticated modules such as knowledge base editors
and debugging tools.
 This enables expert translate to detailed role.
 Here expert system is an integral part of learning system.
Such modules are important in Expert Systems area of Artificial Intelligence.

TYPES OF LEARNING
 Supervised Learning
 Unsupervised Learning
 Reinforcement Learning

Supervised learning
 Supervised learning is the types of machine learning in which machines are trained using
well "labelled" training data, and on basis of that data, machines predict the output. The
labelled data means some input data is already tagged with the correct output.
 In supervised learning, the training data provided to the machines work as the supervisor
that teaches the machines to predict the output correctly. It applies the same concept as a
student learns in the supervision of the teacher.
 Supervised learning is a process of providing input data as well as correct output data to
the machine learning model. The aim of a supervised learning algorithm is to find a
mapping function to map the input variable(x) with the output variable(y).
 In the real-world, supervised learning can be used for Risk Assessment, Image
classification, Fraud Detection, spam filtering, etc.

Class-family car(what cars are family cars) +ve or –ve cars:

 Attributes necessary for this evaluation may be or say price P, seat capacity S, Engine E.
 Our survey deals with whether a given car is family car or not.
 We extract important survey information from domain experts.
We represent each car by ordered pair(X,label)
Where x=(price,seat,engine)
Label=label type ‘1’ or ‘0’.
‘1’for positive example.
‘0’ for negative example.
 If the training set contains ‘N’ such examples.
 It is given by
X= {(Xk , labelk ),1<=K<=N}
Here, Xk =tuple (price,seat,engine ) of the Kth car
 A 3-d graph can be plotted with each Kth instance and its label ( ‘0’ or ‘1’) as data point.
 Certain ranges can also be imposed on the these features such as
P1<=Price<=P2
S1<=Seat<=S2
E1<=engine<=E2
 These conditions fix hypothesis ‘H’ class, from which region ‘C’ containing positive
examples data points in the space can be drawn.
 Each tuple of pairs ((p1,p2),(s1,s2),(e1,e2)) defines one hypothesis ‘h’ from ‘H’ and
choose best hypothesis.
 The learning algorithm should then find out a particular hypothesis h£H that approx. ‘C’
as close as possible.
 All 6 parameters (P1,P2,E1,E2,S1,S2) from training set to include all positive examples
& no –ve examples.
 Now for instance ‘x’ defined as h(x)=1
 If ‘h’ classified ‘x’ as positive example and ‘o’ otherwise.

UNSUPERVISED LEARNING
 In this a typical task of classification or regression using a set of examples prepared by
humans.
 In this learning there is no information available about the correct outputs.
 Here learners predict future precepts based up on present ones.
 This learning includes clustering and discovery.
 Clustering classes are formed for data instances and the method of classifying them are
devised.
 Discovery specific goal is not given, it is based on deductive and inductive learning's.
 Inductive learning involves extracting rules and patterns from given data.
 Deductive learning involves deducing new knowledge from already existing knowledge.
 So, deductive is related to discovery and inductive is formulation of opinion using
incomplete information.
 Unsupervised learning algorithms among neural networks models self organizing map
and ART(Adaptive Resonance theory)
 The goal of learning is the self organizing map is to discover some underlying structure
of data and cause different parts of the network to respond. Similarly to certain input
patterns, these systems adapt over a number iterations using feedback process.
 ART the number of clusters can be varies according to the problem size, this model
allows users to control the degree of similarity between members of the same clusters
with the help of a user defined constant vigilance process.

REINFORCEMENT LEARNING
 In reinforcement learning the decision making system (also known as agent) receives
rewards or feedback (+ve or –ve) for its action at the end of a sequence of steps.
 It is required to assign rewards to step while solving the credit assignment problem.
 This problem determines which steps should receive credit or blame for the final result.
 Reinforcement learning takes place in an environment where the agent cannot directly
compare the results of its action to a desired result.
 It may cause system to win or lose a game, inform a system that it has a good move or
poor move.
 The primary task of reinforcement learning is to obtain a successful function using these
rewards.
 Basic kinds of information to learn by the system:
1) Utility function: the agent learns the utility of being in various states and choose
actions in order to maximize the expected utility of their outcomes. This require the
agent to keep a model of the environment.
2) Action-Value Function: The agent learns an action value function giving the utility of
performing an action in a given state. This is called Q-learning, it is model free
approach.
Now we can draw inference, here the process of learning is happening with a critic.
Here system works based on credit assignment problem. It learns the internal reward
mechanism and uses it to take local actions in order to maximize it.
NEURAL NETWORK BASED LEARNING
 A neural network is a computational model similar to human brain.
 It consists of several simple units (neurons) that work in parallel, with no central control.
 Connection between these working units are weighted and these weights can be modified
by learning method.
 Here Data structure consists of a set of neurons connected by weighted links.
 Feed-Forward network is a kind of supervised learning, used in this concept.
 The training in this network is of 2 parts
An Input vectorà input neuron
An output vectorà Output neuron.
 In the process learning, the input vector is provided at the input layer of the system along
with desired responses, one for each neuron, at the output layer.
 When a forward pass is done, the errors and discrepancies between the desired and actual
responses for each neuron in the output layer are obtained.

INDUCTIVE AND DEDUCTIVE LEARNING

Basis for Deductive Reasoning Inductive Reasoning

comparison
Definition Deductive reasoning is the form of valid Inductive reasoning arrives at a
reasoning, to deduce new information or conclusion by the process of
conclusion from known related facts and generalization using specific facts or
information. data.
Approach Deductive reasoning follows a top-down Inductive reasoning follows a
approach. bottom-up approach.
Starts from Deductive reasoning starts from Premises. Inductive reasoning starts from the
Conclusion.
Validity In deductive reasoning conclusion must be In inductive reasoning, the truth of
true if the premises are true. premises does not guarantee the
truth of conclusions.
Usage Use of deductive reasoning is difficult, as we Use of inductive reasoning is fast
 need facts which must be true. and easy, as we need evidence
instead of true facts. We often use it
in our daily life.
Process Theory→hypothesis→patterns→confirmation. Observations-
→patterns→hypothesis→Theory.
Argument In deductive reasoning, arguments may be In inductive reasoning, arguments
valid or invalid. may be weak or strong.