INTELLIGENT SYSTEMS 

 
Artificial intelligence 
Artificial  intelligence  (AI)  is  the  simulation of human intelligence processes by 
machines,  especially  computer  systems.  These  processes  include  learning (the 
acquisition  of  information  and  rules  for  using  the  information),  reasoning 
(using rules to reach approximate or definite conclusions) and self-correction. 
 
Natural language processing 
Natural  language  processing  (NLP)  is  a  subfield  of  computer  science, 
information  engineering,  and  artificial  intelligence  concerned  with  the 
interactions  between  computers  and  human  (natural)  languages,  in  particular 
how  to  program  computers  to  process  and  analyze  large  amounts  of  natural 
language data.  
 
Vision processing 
A  vision  processing  unit  (VPU)  is  (as  of  2018)  an  emerging  class  of 
microprocessor;  it  is  a  specific  type  of  AI  accelerator  designed  to  accelerate 
machine  vision  tasks.  A  vision  processing  unit  (VPU)  is  a  type  of 
microprocessor  aimed  at  accelerating  machine  learning  and  artificial 
intelligence  technologies.  It  is  a  specialized  processor  that  is  made  to  support 
tasks  like  image  processing,  one  of  several  specialized  chips  such  as  the  GPU 
that are generally useful in machine learning. 
 
Speech processing 
Speech processing is the study of speech signals and the processing methods of 
signals.  The  signals  are usually processed in a digital representation, so speech 
processing  can  be  regarded  as  a  special  case  of  digital  signal  processing, 
applied to speech signals. 
 
Robotics 
Robotics  deals  with  the  design,  construction,  operation,  and  use  of  robots,  as 
well  as  computer  systems  for  their  control,  sensory feedback, and information 
processing.  These  technologies  are  used  to  develop  machines  that  can 
substitute  for  humans  and  replicate  human  actions.  Robots  can  be  used  in 
many  situations  and  for  lots  of  purposes,  but  today  many  are  used  in 
dangerous environments (including bomb detection and deactivation). 
 
Expert systems 
In  artificial  intelligence,  an  expert  system  is  a  computer  system  that  emulates 
the  decision-making  ability  of  a human expert. Expert systems are designed to 
solve  complex  problems  by  reasoning  through  bodies  of  knowledge, 
represented  mainly  as  if–then  rules  rather  than  through  conventional 
procedural code. 
 
Games 

 
Search Knowledge 

 
 
Heuristic function 

A heuristic function is a function that will rank all the possible alternatives at 
any branching step in search algorithm based on the available information. It 
helps the algorithm to select the best route out of possible routes. 

State space search 

State  space  search  is  a  process  used  in  the  field  of computer science, including 
artificial  intelligence  (AI),  in  which  successive  configurations  or  states  of  an 
instance  are  considered,  with  the  intention  of  finding  a  goal  state  with  a 
desired  property.Problems  are  often  modelled  as  a  state  space,  a  set  of  states 
that  a  problem  can  be  in.  The  set  of  states  forms  a  graph  where  two  states  are 
connected  if  there  is  an  operation  that  can  be  performed  to transform the first 
state into the second. 
 
Heuristic search 
Heuristic search’ means that this search algorithm may not find the optimal 
solution to the problem. However, it will give a good solution in reasonable 
time. 

Heuristic  search  is  an  Artificial  Intelligence  search  technique  that  employs 
heuristic  for  its  moves.  Heuristic  is  a  rule  of  thumb  that  probably  leads  to  a 
solution.  Heuristics  play  a  major  role  in  search  strategies  because  of 
exponential  nature of the most problems. Heuristics help to reduce the number 
of alternatives from an exponential number to a polynomial number. 

Hill Climbing 
Hill Climbing is heuristic search used for mathematical optimization problems 
in the field of Artificial Intelligence. Given a large set of inputs and a good 
heuristic function, it tries to find a sufficiently good solution to the problem. 
This solution may not be the global optimal maximum. 
 
Features of Hill Climbing 

1. Variant of generate and test algorithm : It is a variant of generate and test 

algorithm. The generate and test algorithm is as follows : 
 
1. Generate a possible solutions. 
2. Test to see if this is the expected solution. 
3. If the solution has been found quit else go to step 1. 
 
Hence we call Hill climbing as a variant of generate and test algorithm as it 
takes the feedback from test procedure. Then this feedback is utilized by the 
generator in deciding the next move in search space. 
2. Uses the Greedy approach : At any point in state space, the search moves in 
that direction only which optimizes the cost of function with the hope of 
finding the optimal solution at the end. 

Types of Hill Climbing 

1. Simple Hill climbing : It examines the neighboring nodes one by one and 
selects the first neighboring node which optimizes the current cost as next 
node. 
 
2. Steepest-Ascent Hill climbing : It first examines all the neighboring nodes 
and then selects the node closest to the solution state as next node. 
 
3. Stochastic hill climbing : It does not examine all the neighboring nodes 
before deciding which node to select .It just selects a neighboring node at 
random, and decides (based on the amount of improvement in that 
neighbor) whether to move to that neighbor or to examine another. 

Branch and bound search 

Depth-first branch-and-bound search is a way to combine the space saving of 
depth-first search with heuristic information. It is particularly applicable 
when many paths to a goal exist and we want an optimal path. As in A​*​ search, 
we assume that h(n) is less than or equal to the cost of a lowest-cost path from 
n to a goal node. 
The idea of a branch-and-bound search is to maintain the lowest-cost path to 
a goal found so far, and its cost. 
 
Knowledge representation 
Knowledge is the information about a domain that can be used to solve 
problems in that domain. To solve many problems requires much knowledge, 
and this knowledge must be represented in the computer. As part of designing 
a program to solve problems, we must define how the knowledge will be 
represented. A representation scheme is the form of the knowledge that is used 
in an agent. A representation of some piece of knowledge is the internal 
representation of the knowledge. A representation scheme specifies the form 
of the knowledge. A knowledge base is the representation of all of the 
knowledge that is stored by an agent.  
 

 
Predicate logic 
Whereas  propositional  logic  assumes  that  a  world  contains  facts,  predicate 
logic or first-order logic (like natural language) assumes the world contains 
–Objects:  people,  houses,  numbers,  theories,  Ronald  McDonald, 
colors,baseball games, wars, centuries… 
–Relations:  red,  round,  bogus,  prime,  multistoried...,  brother  of,  bigger  than, 
inside, part of, has color, occurred after, owns, comes between,... 
–Functions: father of, best friend, third inning of, one more than, end of… 
 
Rule based system 
A typical rule-based system has four basic components: 

● A list of rules or rule base, which is a specific type of knowledge base. 

● An  inference  engine  or  semantic  reasoner,  which  infers  information  or 
takes  action  based  on  the  interaction  of  input  and  the  rule  base.  The 
interpreter  executes  a  production  system  program  by  performing  the 
following match-resolve-act cycle. 

STRUCTURED REPRESENTATION OF KNOWLEDGE 

Representing  knowledge  using  logical  formalism,  like  predicate  logic,  has 
several  advantages.  They  can  be  combined  with  powerful  inference 
mechanisms  like  resolution,  which  makes  reasoning  with facts easy. But using 
logical  formalism  complex  structures  of  the  world,  objects  and  their 
relationships, events, sequences of events etc. can not be described easily. 
 
An  efficient  system  for  the  representation  of  structured  knowledge  in  a 
particular domain should posses the following four properties: 
 
(i)  Representational  Adequacy:- The ability to represent all kinds of knowledge 
that are needed in that domain. 
 
(ii)  Inferential  Adequacy  :-  The ability to manipulate the represented structure 
and infer new structures. 
 
(iii)  Inferential  Efficiency:-  The  ability  to  incorporate  additional  information 
into the knowledge structure that will aid the inference mechanisms. 
 
(iv)  Acquisitional  Efficiency  :-  The  ability  to  acquire  new  information  easily, 
either by direct insertion or by program control. 
 
Semantic network 
A  semantic  network  is  a  graphic  notation  for  representing  knowledge  in 
patterns  of  interconnected  nodes.  Semantic  networks  became  popular  in 
artificial  intelligence  and  natural  language  processing  only  because  it 
represents  knowledge  or  supports  reasoning.  These  act  as  another  alternative 
for predicate logic in a form of knowledge representation. 

What is Fuzzy Logic? 

Fuzzy  Logic  (FL)  is  a  method  of  reasoning  that  resembles  human  reasoning. 
The  approach  of  FL  imitates  the  way  of  decision  making  in  humans  that 
involves all intermediate possibilities between digital values YES and NO. 
The  conventional  logic  block  that  a  computer  can  understand  takes  precise 
input  and  produces  a  definite  output  as  TRUE  or  FALSE,  which is equivalent to 
human’s YES or NO. 
The  inventor  of  fuzzy  logic,  Lotfi  Zadeh,  observed  that  unlike  computers,  the 
human  decision  making  includes  a  range  of  possibilities  between  YES  and  NO, 
such as − 
 
CERTAINLY 
YES 

POSSIBLY YES 

CANNOT SAY 

POSSIBLY NO 

CERTAINLY NO 
 
The  fuzzy  logic  works  on  the  levels  of  possibilities  of  input  to  achieve  the 
definite output. 
Why Fuzzy Logic? 
Fuzzy logic is useful for commercial and practical purposes. 

● It can control machines and consumer products. 

● It may not give accurate reasoning, but acceptable reasoning. 
● Fuzzy logic helps to deal with the uncertainty in engineering. 

Probabilistic Reasoning in Intelligent Systems 

Probabilistic  Reasoning  in  Intelligent  Systems  is  a  complete  and  accessible 
account  of  the  theoretical  foundations  and  computational  methods  that 
underlie plausible reasoning under uncertainty. 
 
Neural network 
An  Artificial  Neural  Network  (ANN)  is  an  information  processing  paradigm 
that  is  inspired  by  the  way  biological  nervous  systems,  such  as  the  brain, 
process information. The key element of this paradigm is the novel structure of 
the  information  processing  system.  It  is  composed  of  a large number of highly 
interconnected  processing  elements  (neurones)  working  in  unison  to  solve 
specific  problems.  ANNs,  like  people,  learn  by  example.  An  ANN  is  configured 
for  a  specific  application,  such  as  pattern  recognition  or  data  classification, 
through  a  learning  process.  Learning  in  biological  systems  involves 
adjustments to the synaptic connections that exist between the neurons. 
 
Genetic Algorithm 
The  genetic  algorithm  is  a  method  for  solving  both  constrained  and 
unconstrained  optimization  problems  that  is  based  on  natural  selection.  The 
genetic  algorithm  repeatedly  modifies  a  population  of  individual  solutions.  At 
each step, the genetic algorithm selects individuals at random from the current 
population  to  be  parents  and  uses  them  to  produce  the  children  for  the  next 
generation.  Over  successive  generations,  the  population  "evolves"  toward  an 
optimal solution.  
The  genetic  algorithm  uses  three  main  types  of  rules  at  each  step to create the 
next generation from the current population: 

● Selection  rules  select  the  individuals,  called  parents,  that  contribute  to 
the population at the next generation. 
● Crossover  rules  combine  two  parents  to  form  children  for  the  next 
generation. 
● Mutation  rules  apply  random  changes  to  individual  parents  to  form 
children. 

Inductive Learning / Learning by induction 

This involves the process of learning by example  
--  where  a  system  tries  to  induce  a  general  rule  from  a  set  of  observed 
instances. 
This involves classification  
--  assigning,  to  a  particular  input,  the  name  of  a  class  to  which  it  belongs. 
Classification is important to many problem solving tasks. 
A learning system has to be capable of evolving its own class descriptions: 

● Initial class definitions may not be adequate. 

● The world may not be well understood or rapidly changing. 

Automated Machine Learning / Learning by automation 

The  task  of  constructing  class  definitions  is  called  induction  or  concept 
learning.  Automated  machine  learning  (AutoML)  represents  a  fundamental 
shift  in  the  way  organizations  of  all  sizes  approach  machine  learning  and data 
science. Applying traditional machine learning methods to real-world business 
problems  is  time-consuming,  resource-intensive,  and  challenging.  It  requires 
experts  in  the  several  disciplines,  including  data  scientists  – some of the most 
sought-after professionals in the job market right now.