Machine learning

journeycartAI and Robotics

Oct 15, 2013 (3 years and 11 months ago)

89 views

Chapter 13


ADVANCED
INTELLIGENT
SYSTEMS

Learning Objectives


Understand machine
-
learning concepts


Learn the concepts and applications of
case
-
based systems


Understand the concepts and applications
of genetic algorithms


Understand fuzzy set theories and their
applications in designing intelligent systems

Learning Objectives


Understand the concepts and applications
of natural language processing (NLP)


Learn the concepts, advantages, and
limitations of voice technologies


Learn about integrated intelligent support
systems

Machine
-
Learning Techniques



Machine
-
learning concepts and
definitions


Machine learning



The process by which a computer learns
from experience (e.g., using programs that
can learn from historical cases)

Machine
-
Learning Techniques



Human
learning

is a combination of many
complicated cognitive processes including:


Induction (learning by example)


Deduction (specific inferences based on
generalities)


Analogy (transference)


Other special procedures related to observing or
analyzing examples

Machine
-
Learning Techniques



How learning relates to intelligent systems


Learning systems demonstrate interesting learning
behaviors


AI is not able to learn as well as humans or in the
same way that humans


Machine learning cannot be applied in a creative
way, although such systems can handle cases to
which they have never been exposed


It is not clear why learning systems succeed or fail


A common thread running through most AI
approaches to learning is the manipulation of
symbols rather than numeric information

Machine
-
Learning Techniques



Machine
-
learning methods


Supervised learning



A method of training artificial neural networks
in which sample cases are shown to the
network as input and the weights are
adjusted to minimize the error in its outputs


Unsupervised learning


A method of training artificial neural networks
in which only input stimuli are shown to the
network, which is self
-
organizing


Machine
-
Learning Techniques


Machine
-
Learning Techniques



Inductive learning


Case
-
based
reasoning


Neural computing


Genetic algorithms


Natural language
processing (NLP)


Cluster analysis


Statistical methods


Explanation
-
based
learning



A machine learning
approach that
assumes that there is
enough existing
theory to rationalize
why one instance is
or is not a
prototypical member
of a class

Machine
-
learning methods and algorithms

Case
-
Based Reasoning (CBR)


Case
-
based reasoning (CBR)



A methodology in which knowledge
and/or inferences are derived from
historical cases

Case
-
Based Reasoning (CBR)


Analogical reasoning



Determining the outcome of a problem with the use
of analogies. A procedure for drawing conclusions
about a problem by using past experience



Suppose, for example, that I am thinking about buying a new car. I'm
very likely to speak with other people who have recently bought new
cars, noting their experiences with various makes, models, and
dealers. If I discover that three of my friends have recently bought
Brand X from ABC Dealership and that all three have been delighted
with their purchases, then I will conclude by analogy that if I buy Brand
X from ABC Dealership, I will be delighted, too.



Evaluating Analogical Reasoning


Number of instances, instance variety, # of similarities,
relevance, # of dissimilarities, modesty of conclusion

Case
-
Based Reasoning (CBR)


Inductive learning



A machine learning approach in which rules
are inferred from facts or data

Case
-
Based Reasoning (CBR)


The basic idea and process of CBR


Four
-
step process

1.
Retrieve

2.
Reuse

3.
Revise

4.
Retain

Case
-
Based Reasoning (CBR)


Definition and concepts of cases in CBR


Ossified cases



Cases that have been analyzed and have no
further value


Paradigmatic cases



A case that is unique that can be maintained
to derive new knowledge for the future

Case
-
Based Reasoning (CBR)


Definition and concepts of cases in CBR


Stories


Cases with rich information and episodes.
Lessons may be derived from this kind of
cases in a case base

Case
-
Based Reasoning (CBR)

Case
-
Based Reasoning (CBR)


Benefits and usability of CBR


CBR makes learning much easier and the
recommendation more sensible

Case
-
Based Reasoning (CBR)


Advantages of using CBR


Knowledge acquisition is improved.


System development time is faster


Existing data and knowledge are leveraged


Complete formalized domain knowledge is not
required


Experts feel better discussing concrete cases


Explanation becomes easier


Acquisition of new cases is easy


Learning can occur from both successes and failures

Case
-
Based Reasoning (CBR)


Uses, issues, and applications of CBR


Applications


CBR in electronic commerce


WWW and information search


Planning and control


Design


Reuse


Diagnosis


Reasoning

Case
-
Based Reasoning (CBR)


Uses, issues, and applications of CBR


Implementation issues for designers


What makes up a case? How can we represent
case memory?


Automatic case
-
adaptation rules can be very
complex


How is memory organized? What are the indexing
rules?


The quality of the results is heavily dependent on
the indexes used

Case
-
Based Reasoning (CBR)


Implementation issues for designers


How does memory function in relevant information
retrieval?


How can we perform efficient searching (i.e.,
knowledge navigation) of the cases?


How can we organize the cases?


How can we design the distributed storage of cases?


How can we adapt old solutions to new problems?
Can we simply adapt the memory for efficient
querying, depending on context? What are the
similarity metrics and the modification rules?

Case
-
Based Reasoning (CBR)


Implementation issues for designers


How can we factor errors out of the original cases?


How can we learn from mistakes? That is, how can
we repair and update the case base?


The case base may need to be expanded as the
domain model evolves, yet much analysis of the
domain may be postponed.


How can we integrate CBR with other knowledge
representations and inferencing mechanisms?


Are there better pattern
-
matching methods than the
ones we currently use?


Are there alternative retrieval systems that match
the CBR schema?

Case
-
Based Reasoning (CBR)


Success factors for CBR systems

1.
Determine specific business objectives

2.
Understand your end users and customers

3.
Design the system appropriately

4.
Plan an ongoing knowledge
-
management
process

5.
Establish achievable returns on investment
(ROI) and measurable metrics

6.
Plan and execute a customer
-
access strategy

7.
Expand knowledge generation and access
across the enterprise

Genetic Algorithm Fundamentals



Genetic algorithms (GAs)



Software programs that learn in an
evolutionary manner similar to the way
biological systems evolve

Genetic Algorithm Fundamentals



Genetic algorithm process and terminology


Chromosome


A candidate solution for a genetic algorithm


Reproduction


The creation of new generations of improved
solutions with the use of a genetic algorithm

Genetic Algorithm Fundamentals



Genetic algorithm process and terminology


Crossover


The combining of parts of two superior
solutions by a genetic algorithm in an attempt
to produce an even better solution


Mutation


A genetic operator that causes a random
change in a potential solution

Genetic Algorithm Fundamentals

Genetic Algorithm Fundamentals



A few parameters must be set for the genetic
algorithm


Number of initial solutions to generate


Number of offspring to generate


Number of parents and offspring to keep for the
next generation


Mutation probability (very low)


Genetic Algorithm Fundamentals



Limitations of genetic algorithms


Not all problems can be framed in the mathematical
manner that genetic algorithms demand


Development of a genetic algorithm and
interpretation of the results requires an expert who
has both the programming and
statistical/mathematical skills demanded by the
genetic algorithm technology in use


In some situations, the “genes” from a few
comparatively highly fit (but not optimal) individuals
may come to dominate the population, causing it to
converge on a local maximum

Genetic Algorithm Fundamentals



Limitations of genetic algorithms


Most genetic algorithms rely on random number
generators that produce different results each time
the model runs


Locating good variables that work for a particular
problem is difficult


Selecting methods by which to evolve the system
requires thought and evaluation

Developing Genetic
Algorithm Applications



GAs are a type of machine learning for
representing and solving complex
problems

Developing Genetic
Algorithm Applications



Dynamic process control


Induction of optimization
of rules


Discovery of new
connectivity topologies
(e.g., neural computing
connections, i.e., neural
network design)


Simulation of biological
models of behavior and
evolution


Complex design of
engineering structures


Pattern recognition


Scheduling


Transportation and
routing


Layout and circuit design


Telecommunication


Graph
-
based problems

Applications of GAs include:

Fuzzy Logic Fundamentals



Fuzzy logic



Logically consistent ways of reasoning
that can cope with uncertain or partial
information; characteristic of human
thinking and many expert systems.


Fuzzy sets



A set theory approach in which set
membership is less precise than having
objects strictly in or out of the set

Fuzzy Logic Fundamentals


Fuzzy Logic Fundamentals



Fuzzy logic applications in manufacturing and
management


Selection of stocks to purchase (e.g., the Japanese
Nikkei stock exchange)


Retrieval of data (because fuzzy logic can find data
quickly)


Inspection of beverage cans for printing defects


Matching of golf clubs to customers’ swings


Risk assessment


Control of the amount of oxygen in cement kilns


Accuracy and speed increases in industrial quality
-
control applications


Sorting problems in multidimensional spaces

Fuzzy Logic Fundamentals



Fuzzy logic applications in manufacturing and
management


Enhancement of models involving queuing (i.e.,
waiting lines)


Managerial decision support applications


Project selection


Environmental control building


Control of the motion of trains


Paper mill automation


Space shuttle vehicle orbiting


Regulation of water temperature in shower heads

Natural Language
Processing (NLP)



Natural language processing (NLP)



Using a natural language processor to
interface with a computer
-
based system


Two types of NLP


Natural language understanding


Natural language generation

Natural Language
Processing (NLP)



Some problems that make NLP difficult


Word boundary detection


Word sense disambiguation


Syntactic ambiguity


Imperfect or irregular input


Speech acts and plans

Natural Language
Processing (NLP)



The current NLP technology


Search and information retrieval


A person enters a certain phrase, word, or
sentence on which to search the Internet or some
database, and NLP is then used to construct the
best query possible

Natural Language
Processing (NLP)



Applications of NLP


Human

computer interfaces


Abstracting and summarizing text


Analyzing grammar


Understanding speech

Natural Language
Processing (NLP)



Applications of NLP


Front ends for other software packages

querying a database that allows the user to
operate the applications programs with everyday
language


Text mining


FAQs and query answering

Natural Language
Processing (NLP)



Machine translation


Translation of content to other languages


Criteria used to assess machine translation

1.
Intelligibility

2.
Accuracy

3.
Speed

Voice Technologies



Voice technologies fall into three broad
categories:


Voice (or speech) recognition


Voice (or speech) understanding


Text
-
to
-
voice (or voice synthesis)

Voice Technologies



Voice (speech) recognition


Translation of the human voice into individual
words and sentences understandable by a
computer


Speech understanding


An area of AI research that attempts to allow
computers to recognize words or phrases of
human speech

Voice Technologies



Advantages of voice technologies

1.
Ease of access

2.
Speed

3.
Manual freedom

4.
Remote access

5.
Accuracy

6.
Communicating while driving

7.
Quick selection

8.
Security

9.
Cost benefit

Voice Technologies



Limitations of speech recognition and
understanding


Inability to recognize long sentences, or the
excessive length of time needed to accomplish
that understanding


High cost


Speech may need to be combined with keyboard
entry, which slows communication

Voice Technologies



Voice synthesis


The technology by which computers
convert text
-
to
-
voice (speak)


A text
-
to
-
speech system is composed of two
parts:


Front end takes input in the form of text and
outputs a symbolic linguistic representation


Back end takes the symbolic linguistic
representation as input and outputs the
synthesized speech waveform

Voice Technologies



Voice technology applications


Call center


Contact of customer care center


Computer/telephone integration (CTI)


Interactive voice response (IVR)


Voice portal


Voice over IP (VoIP)

Voice Technologies



Voice portals


Web sites, usually portals, with audio
interfaces

Developing

Integrated Advanced Systems



Fuzzy neural networks


Fuzzification


A process that converts an accurate number
into a fuzzy description, such as converting
from an exact age into young or old


Defuzzification


Creating a crisp solution from a fuzzy logic
solution

Developing

Integrated Advanced Systems


Developing

Integrated Advanced Systems


Developing

Integrated Advanced Systems



Genetic algorithms and neural networks


The genetic learning method can perform rule
discovery in large databases, with the rules fed into
a conventional ES or some other intelligent system


To integrate genetic algorithms with neural network
models use a genetic algorithm to search for
potential weights associated with network
connections


A good genetic learning method can significantly
reduce the time and effort needed to find the optimal
neural network model