ARTIFICIAL INTELLIGENCE
UNIT - IV
INDEX
Learning
• What is learning?
• Rote Learning
• Learning by taking advice
• Learning in Problem Solving
• Learning from examples
• Winston’s Learning Program
• Decision Trees
What is learning?
In general,
• Learning is a relatively lasting change in behaviour that is the result of
experience.
• It is the acquisition of information, knowledge, and skills.
• When you think of learning, it's easy to focus on formal education that
takes place during childhood and early adulthood.
• But learning is an ongoing process that takes place throughout life and
isn't confined to the classroom.
When it comes to machines..
• Learning is one of the fundamental building blocks of artificial
intelligence (AI) solutions.
• From a conceptual standpoint, learning is a process that improves the
knowledge of an AI program by making observations about its
environment.
• From a technical/mathematical standpoint, AI learning processes
focused on processing a collection of input-output pairs for a specific
function and predicts the outputs for new inputs.
• Most of the artificial intelligence(AI) basic literature identifies two
main groups of learning models: supervised and unsupervised.
However, that classification is an oversimplification of real world AI
learning models and techniques.
Rote Learning
• Rote learning is a process of memorizing specific new items as they
are encountered.
• The basic idea is simple and easy to realize with in a computer
program : Each time a new and useful piece of information is
encountered , it is stored away for future use.
• For example, an AI system might be designed to recognize faces by
extracting a variety of features (such as distance between the eyes)
from an image and searching for a match within a database of 1000
stored feature sets.
• If it finds a match, it has recognized the person; if not, it reports
“unknown person.”
• In this latter case, the person or some human operator can provide
the necessary details of this unknown person, and the system can
enter the details in its database so that next time this person is
presented to the system he or she will be correctly recognized.
• Every rote learning event increases the size of the database, and
speed of recognition (as well as accuracy if new faces are similar to
old ones) will decrease as the size of the database grows.
• Depending on the application of this system, certain new faces may
be seen once and never again, in which case rote learning wastes time
and space.
• Alternatively, this new face may be commonly seen at first but may
over time become rare and ultimately no longer seen.
• In the first case, rote learning is immediately detrimental to system
performance, and in the second case it becomes detrimental over
time.
Induction learning (Learning by example)
• Induction learning is carried out on the basis of supervised learning.
• In this learning process, a general rule is induced by the system from a
set of observed instance.
• However, class definitions can be constructed with the help of a
classification method.
For Example:
• Consider that 'ƒ' is the target function and example is a pair (x ƒ(x)),
where 'x' is input and ƒ(x) is the output function applied to 'x'.
• Given problem: Find hypothesis h such as h ≈ ƒ
• So, in the following fig-a, points (x,y) are given in plane so that y =
ƒ(x), and the task is to find a function h(x) that fits the point well.
• In fig-b, a piecewise-linear 'h'
function is given, while the
fig-c shows more
complicated 'h' function .
Both the functions agree with
the example points, but differ
with the values of 'y' assigned
to other x inputs..
• As shown in fig.(d), we have a function that apparently
ignores one of the example points, but fits others with a
simple function.
• The true/ is unknown, so there are many choices for h,
but without further knowledge, we have no way to
prefer (b), (c), or (d).
Learning by taking advice
• This type is the easiest and simple way of learning.
• In this type of learning, a programmer writes a program to give some
instructions to perform a task to the computer. Once it is learned (i.e.
programmed), the system will be able to do new things.
• Also, there can be several sources for taking advice such as humans(experts),
internet etc.
• However, this type of learning has a more necessity of inference than rote
learning.
• As the stored knowledge in knowledge base gets transformed into an
operational form, the reliability of the knowledge source is always taken into
consideration.
Automated Advice Taking
• The following steps summarise this method :
• Request-- This can be simple question asking about general advice or
more complicated by identifying shortcomings in the knowledge base and
asking for a remedy.
• Interpret-- Translate the advice into an internal representation.
• Operationalise-- Translated advice may still not be usable so this stage
seeks to provide a representation that can be used by the performance
element .
• Integrate-- When knowledge is added to the knowledge base care must be
taken so that bad side-effects are avoided . E.g. Introduction of
redundancy and contradictions .
• Evaluate-- The system must assess the new knowledge for errors,
contradictions etc
Learning in problem solving
• Cognitive processing aimed at figuring out how to achieve a goal is
called problem solving.
• In problem solving, the problem solver seeks to devise a method for
transforming a problem from its current state into a desired state when
a solution is not immediately obvious to the problem solver.
(1) problem solving is cognitive–that is, it occurs internally in the mind
(or cognitive system) and must be inferred indirectly from behaviour;
(2) problem solving is a process–it involves the manipulation of
knowledge representations (or carrying out mental computations); and
(3) problem solving is directed–it is guided by the goals of the problem
solver.
• A problem occurs when a problem solver has a goal but initially does
not know how to achieve the goal.
• This definition has three parts:
(1) the current state–the problem begins in a given state;
(2) the goal state–the problem solver wants the problem to be in a
different state, and problem solving is required to transform the problem
from the current (or given) state into the goal state, and
(3) obstacles–the problem solver does not know the correct solution and
an effective solution method is not obvious to the problem solver.
• Learning to problem solve requires acquiring multiple forms of knowledge.
• Problem solving is viewed as a search of a state-space formulation of a
problem.
• With this formalism, operators are applied to states to transit from the
initial state to the goal state.
• The learning task to acquire knowledge of the state-space to guide search.
• In particular, three forms of knowledge are required: why each operator is
useful, when to apply each operator, and what each operator does.
• Episodic learning, perturbation and relational models form an integrated
approach for learning problem solving.
• The approach demonstrates self-teaching by reasoned experimentation.
• Episodic learning is a technique for learning why individual
operators are useful in a solution path.
• Episodic learning acquires generalized operator sequences which
achieve the goal state.
• This is done by backing-up state evaluation and learning sub-goals in
the state-space.
• Perturbation is a technique for learning when individual operators are
useful.
• Perturbation guides the generalization process to discover minimally-
constrained preconditions for useful operator applications.
• This done by experimentation, thereby reducing the teacher's role in
the learning process.
• Learning relational models is a technique for discovering what
individual operators do.
• Relational models are an explicit representation of the transformation
performed by operators.
• This representation enables the learning element to reason with
operator semantics to guide further learning.
A Blocks World Learning Example -- Winston (1975)
• The goal is to construct representation of the definitions of concepts in
this domain.
• Concepts such a house - brick (rectangular block) with a wedge (triangular
block) suitably placed on top of it, tent - 2 wedges touching side by side, or
an arch - two non-touching bricks supporting a third wedge or brick, were
learned.
• The idea of near miss objects -- similar to actual instances was introduced.
• Input was a line drawing of a blocks world structure.
• Input processed to produce a semantic net representation of the
structural description of the object.
• Links in network include left-of, right-of, does-not-marry, supported-by,
has-part, and isa.
• The marry relation is important -- two objects with a common touching
edge are said to marry. Marrying is assumed unless does-not-marry stated.
• There are three basic steps to the problem of concept formulation:
• Select one know instance of the concept. Call this the concept definition.
• Examine definitions of other known instance of the concept. Generalise
the definition to include them.
• Examine descriptions of near misses. Restrict the definition to exclude
these.
• Both steps 2 and 3 rely on comparison and both similarities and
differences need to be identified.
Version Spaces
• Structural concept learning systems are not without their problems.
• The biggest problem is that the teacher must guide the system through
carefully chosen sequences of examples.
• Winston's program the order of the process is important since new links
are added as and when now knowledge is gathered.
• The concept of version spaces aims is insensitive to order of the example
presented.
• To do this instead of evolving a single concept description a set of
possible descriptions are maintained.
• As new examples are presented the set evolves as a process of new
instances and near misses.
• We will assume that each slot in a version space description is made
up of a set of predicates that do not negate other predicates in the set --
positive literals.
• Indeed we can represent a description as a frame bases representation
with several slots or indeed use a more general representation.
• Mitchell's candidate elimination algorithm is the best known
algorithm for working on version spaces datasets.
Decision Trees