|
|
|
|
|
|
|
|
|
|
Lesson#34
|
Intelligent Systems
|
|
|
|
Today’s Goals: Intelligent Systems
To become familiar with the distinguishing features of
intelligent systems with respect to other software
systems
To become able to appreciate the role of intelligent systems in
scientific, business and consumer
applications
To look at several techniques for designing intelligent systems
34.1 (Artificial) Intelligent Systems:
SW programs or SW/HW systems designed to perform
complex tasks employing
strategies that mimic
some aspect of human thought
One can debate endlessly about whether a certain system is intelligent
or not
But to my mind, the key criterion is evolution: it is intelligent if it
can learn (even if only a limited
sense) and get better with time
Not a Suitable Hammer for All Nails!
if the nature of computations required in a task is not
well understood or
there are too many exceptions to the rules
or known
algorithms are too complex or inefficient
then AI
has the potential of offering an acceptable solution
Selected Applications:
Games: Chess, SimCity
Image recognition
Medical diagnosis
Robots
Business intelligence
Sub-Categories of AI:
Expert systems
Systems that, in some limited sense, can replace an expert
Robotics
Natural language processing
Teaching computers to understand human language, spoken as well as
written
Computer vision
Selected Techniques:
Artificial neural networks
Genetic algorithms
Rule-based systems
Fuzzy logic
Many times, any one of them can solve the problem at hand, but at
others, only the right one will do.
Therefore, it is important to have some appreciation of them all .
Neural Networks:
Original inspiration was the human brain; emphasis now on
usefulness as a computational tool
Many useful NN paradigms, but scope of today's discussion limited to the
feed-forward network, the
most popular paradigm
Feed-forward Network:
It is a layered structure consisting of a number of homogeneous and
simple (but nonlinear) processing
elements
All processing is local to a processing element and is asynchronous
During training the FN is forced to adjust its parameters so that
its response to input data becomes closer
to the desired response
Based on Darwin's evolutionary principle of ‘survival of the fittest’
GAs require the ability to recognize a good solution, but not how to get
to that solution.
Genetic Algorithms (2):
The procedure:
An initial set of random solutions is ranked in terms of ability to
solve the problem at hand
The best solutions are then crossbred and mutated to form a new set
The ranking and formation of new solutions is continued until a good
enough solution is found or …
Rulebased Systems (1):
Based on the principles of the logical reasoning ability of
humans
Components of an RBS:
Rulebase
Working memory
Rule interpreter
The design process:
An RBS engineer interviews the expert to acquire the
comprehensive set of heuristics that covers the
situations that may occur in a given domain
This set is then encoded in the form of IF-THEN structures to form the
required RBS
34.2 Fuzzy Logic:
Based on the principles of the approximate reasoning faculty that
humans use when faced with linguistic
ambiguity
The inputs and outputs of a fuzzy system are precise, only the reasoning
is approximate
Parts of the knowledgebase of a fuzzy system:
Fuzzy rules
Fuzzy sets
The output of a fuzzy system is computed by using:
The MIN-MAX technique for combining fuzzy rules
The centroid method for defuzzification
Now we know about a few techniques
Let’s now consider the situation when we are given a particular problem
and asked to find an AI
solution to that problem.
How do we determine the right technique for that particular problem?
Selection of an Appropriate AI Technique:
A given problem can be solved in several ways
Even if 2 techniques produce solutions of a similar quality, matching
the right technique to a problem
can save on time & resources
Characteristics of an optimal technique:
The solution contains all of the required information
The solution meets all other necessary criteria
The solution uses all of the available (useful) knowledge
How do we determine the suitability of a particular AI technique
for a given task. We look at the task’s
requirements and then see which technique fulfils those requirements
more completely – the one which
does, is the one we use!
Here are a few aspects of the task and the techniques that we need to be
aware off …
Credit Card Issuance:
Challenge. Increase the acceptance rate of card applicants who
will turn out to be good credit risks
Inputs. Applicant's personal and financial profiles
Output. Estimated yearly loss if application is accepted
Expert knowledge. Some rules of thumb are available
Data. Profiles & loss data available for 1+ million applicants
Suitable technique?
Determination of the Optimal Drug Dosage:
Challenge. Warn the physician if she prescribes a dosage which is
either too high or too low
Inputs. Patient's medical record. Pharmaceutical drug dosage
instructions
Output. Warning along with reasons for the warning
Data. Medical records of thousands of patients. Drug dosage instructions
on dozens of medicines
Suitable technique?
Prediction of Airline Cabin Crew's Preferences:
Challenge. Predict the future base/status preferences of the
cabin crew of an airline. The predicted
preferences will be used by the airline for forecasting its staffing and
training requirements
Inputs. Crew's personal profiles. Preference history. Other data.
Output. Predicted preference card for a date one year in the future
Expert knowledge. Some rules of thumb are available
Data. Available for the last four years for 8000 crew members
Suitable technique?
The Right Technique:
Selection of the right AI technique requires intimate knowledge
about the problem as well as the
techniques under consideration
Real problems may require a combination of techniques (AI and/or nonAI)
for an optimal solution
34.3 Robotics:
Automatic machines that perform various tasks that were
previously done by humans
• Accuracy
•
Explainability
• Response
speed
•
Scalability
•
Compactness
•
Flexibility
•
Embedability
• Ease of
use
• Learning
curve
• Tolerance
for complexity
• Tolerance
for noise in
data
• Tolerance
for sparse data
•
Independence from
experts
•
Development speed
• Computing
ease
Example:
Pilot-less combat airplanes
Land-mine hunters
Autonomous vacuum-cleaners
Components: Body structure, actuators, power-source, sensors, controller
(the AI-based part)
Autonomous Web Agents:
Also known as mobile agents, softbots
Computer program that performs various actions continuously,
autonomously on behalf of their
principal!
Key component of the Semantic Web of tomorrow
Multi-agent communities are being developed in which agents meet and
represent the interests of their
principals in negotiations or collaborations.
Example:
Agents of a patient and a doctor get together to negotiate and
select a mutually agreeable time, cost
Decision Support Systems:
Interactive software designed to improve the decision-making
capability of their users
Utilize historical data, models to solve problems
The do not make decisions - just assist in the process
They provide decision-makers with information via easy to manage
reports, what-if scenarios, and
graphics
The Future?
Get ready to see robots playing a bigger role in our daily lives
Robots will gradually move out of the industrial world and into our
daily life, similar to the way
computers did in the 80’s
Decision support systems will become a bigger part of the professional
life of doctors, managers,
marketers, etc
Autonomous land, air, sea vehicles controlled from 1000’s of miles away
from the war zone
Today’s Summary:Intelligent Systems
We looked at the distinguishing features of intelligent systems
w.r.t. other software systems
We looked at the role of intelligent systems in scientific, business,
consumer and other applications
We discussed several techniques for designing intelligent systems
Next Lecture:(Data Management)
To become familiar with the issues and problems related to
data-intensive computing
To become able to appreciate data management concepts and their
evolution over the years
|
|
|
|
|
|
|
| Copyright © 2008-2013 zainbooks All Rights Reserved |
|