Artificial Intelligence


<< Advanced Topics
Artificial Intelligence (CS607)
Search though the internet and read about intersting happeneing and reseach
going on around the globe in the are of clustering.
The above link might be useful to explore knowledge about clustering.
10 Conclusion
We have now come to the end of this course and we have tried to cover all the
core technologies of AI at the basic level. We hope that the set of topics we have
studied so far can give you the essential base to work into specialized, cutting-
edge areas of AI.
Let us recap what have we studied and concluded so far. The list of major topics
that we covered in the course is:
Introduction to intelligence and AI
Classical problem solving
Genetic algorithms
Knowledge representation and reasoning
Expert systems
Fuzzy systems
Advanced topics
Let us review each of them very briefly.
10.1 Intelligence and AI
Intelligence is defined by some characteristics that are common in different
intelligent species, including problem solving, uncertainty handling, planning,
perception, information processing, recognition, etc.
AI is classified differently by two major schools of thought. One school classifies
AI as study of systems that think like humans i.e. strong AI and the other
classifies AI as study of systems that act like humans i.e. weak AI. Most of the
techniques prevalent today are counted in the latter classification.
10.2 Problem solving
Many people view AI as nothing but problem solving. Early work in AI was done
around the generic concept of problem solving, starting with the basic technique
of generate and test. Although such classical problem solving did not get
extraordinary success but still it provided a conceptual backbone for almost each
approach to the systematic exploration of alternatives.
The basic technique used in classical problem solving is searching. There are
several algorithms for searching for problem solving, including BFS, DFS, hill
climbing, beam search, A* etc. broadly categorized on the basis of completeness,
optimality and informed ness. A special branch of problem solving through
Artificial Intelligence (CS607)
searching involved adversarial problems like classical two-player games, handled
in classical problem solving by adversarial search algorithms like Minimax.
10.3 Genetic Algorithms
Genetic algorithms is a modern advancement to the hill climbing search based
problem solving. Genetic algorithms are inspired by the biological theory of
evolution and provide facilities of parallel search agents using collaborative hill
climbing. We have seen that many otherwise difficult problems to solve through
classical  programming  or  blind  search  techniques  are  easily  but
undeterministically solved using genetic algorithms.
At this point we introduced the cycle of AI to set base for systematic approach to
study contemporary techniques in AI.
10.4 Knowledge representation and reasoning
Reasoning has been presented by most researchers in AI as the core ability of an
intelligent being. By nature, reasoning is tightly coupled with knowledge
representation i.e. the reasoning process must exactly know how the knowledge
is kept to manipulate and extract new knowledge from it.
As we are yet to decode the exact representation of knowledge in natural
intelligent beings like humans, we have based our knowledge representation and
hence reasoning on man-made logical representation namely logic i.e. predicate
logic and family.
10.5 Expert systems
The first breakthrough successful application of AI came from the subject of
knowledge representation and reasoning and was name expert systems. Based
on its components i.e. knowledge base, inference and working memory, expert
systems have been successfully applied to diagnosis, interpretation, prescription,
design, planning, simulations, etc.
10.6 Fuzzy systems
Predicate logic and the classical and successful expert systems were limited in
that they could only deal with perfect boolean logic alone. Fuzzy logic provided
the new base of knowledge and logic representation to capture uncertain
information and thus fuzzy reasoning systems were developed. Just like expert
systems, fuzzy systems have almost recently found exceptional success and are
one of the most used AI systems of today, with applications ranging from self-
focusing cameras to automatic intelligent stock trading systems.
10.7 Learning
Having covered the core intelligence characteristic of reasoning, we shifted to the
other major half contributed to AI i.e. learning or formally machine learning. The
KRR and fuzzy systems perform remarkably but they cannot add or improve their
Artificial Intelligence (CS607)
knowledge at all, and that is where learning was felt essential i.e. the ability of
knowledge based systems to improve through experience.
Learning has been categorized into rote, inductive and deductive learning. Out of
these all almost all the prevalent learning techniques are attributed to inductive
learning, including concept learning, decision tree learning and neural networks.
10.8 Planning
In the end we have studied a rather specialized part of AI namely planning.
Planning is basically advancement to problem solving in which concepts of KRR
are fused with the knowledge of classical problem solving to construct advanced
systems to solve reasonably complex real world problems with multiple,
interrelated and unrelated goals. We have learned that using predicate logic and
regression, problems could be elegantly solved which would have been
nightmare for machines in case of classical problem solving approach.
10.9 Advanced Topics
You have been given just a hint of where the field of AI is moving by mentioning
some of the exciting areas of AI of today including vision, robotics, soft-computing
and clustering. Of these we saw robotics as the most comprehensive field in
which the other topics like vision can be considered as a sub-part.
Now, it's up to you to take these thoughts and directions along with the basics
and move forward into advanced study and true application of the field of Artificial