lets deal with neural networks

Replies

• 

  PraveenKumar Purushothaman

  I am interested but I have no idea about it... 😔
• 

  Ankita Katdare

  I think we can all learn a lot from this thread.
  Where to start?
• 

  Adivy

  Neural networks are computational models inspired by the real neurons.
  There are over 100 billion neurons in our brain.Ofcourse everyone must have heard about neurons but have you ever thought of what makes them different from other cells??Why are they so important?(I know some of you may know the answer but still).
  There are two things that make them different:-
  Firstly their special structure i.e that is apart from normal cell components they have these dendrites(number of arms you can say ).
  Secondly they are electrically active.
  Neurons have resting as well as active states.They connect to other neurons via joints called Synapse.It is like a neuron with its dendrites connected to dendrites of other neurons at synapses.
  This was all biology stuff.So coming to how we can model it,its like simply capturing the behaviour of a neuron into mathematical model.
  This mathematical model is called McCulloch-Pitts Model.Its is like a circle representing a neuron with inputs x1,x2,..xi,..,xn which are the information coming from different dendrites of the neuron and an output y which is a function of inputs with their weights(w).So the equation is somewhat like this:-
  y=g((summation of wi*xi)-b )
  Neural netwotk(NN) is a parametric model that can model arbitrary i/o relationships.It can learn i/o relationships from sample data.
  Applications of NN in which i was interested was Optical character recognition(OCR) which is basically a DSP application.There are many more applications like in VLSI,contol &instrumentation,etc.
  There are varieties of NN:-
  Multilayer Perceptions/Backpropagation networks,
  Radial Basis Function Networks (I know only this one because it was involved in OCR),
  Hopfield network,Recurrent networks,Modular networks etc.
  
  Guys this is what i know about neural networks.I didnt particularly study this as a subject but i have used it in studyin or doing some OCR techniques.Any mistakes please do correct me.
• 

  Ankita Katdare

  @Adivy: That was enlightening! 😀
  What are the other major applications of neural networks?
• 

  Adivy

  @Abrakadabra:😀 Like I said it was a very little part of what I studied in DSP Application,so I just shared what I knew.I dont have deep knowledge on the topic.And about the application there are many like:-
  Pattern recognition or identifying patterns or trends in data(One of the main apps),
  Robotics,detection and tracking of moving items ,motion detection ,etc
  Its widely used in Artificial intelligence.
  What is interesting is their ability to learn by example unlike conventional computers which are algorithmic.
  So we can say NN=Learning by example😀Like how we grow up taking others as example.
• 

  narayana murthy

  @adivy: impressive this is what i love to say next another model is introduced called perceptron model which brought a great revolution in neural networks this was given as like after perceptron and before perceptron
• 

  ISHAN TOPRE

  Adivy: That is good. But I read a while ago about how our eyes scan a particular object. Does neural network concerns that too? I mean we have scanners at many places. 😀
• 

  narayana murthy

  @ishu: hahaha silly question as our brain is made of completely with neurons how can can eyes scan without presence of neuron
  our retina takes the image in reverse like it look like in down to up and it goes to brain in exact way by the neuron networks
  for this you can refer books about brain
• 

  narayana murthy

  i am free now so i am posting something about this topic
  actually this was started in around 1943 as our cean said it was mp model later some days perceptron model was introduced to overcome drawbacks of mp model and it is an revolutionary model
  so our artificial networks was made into 2 phases like before perceptron and after perceptron
  but it got a drawback of complexity for higher applications
  later Minsky and Papert was announced that this perceptron model is not applicable for complex applications this made the research on neural network stopped for about 20 years
  and now a days we are using some neural network by upgrading perceptron model
  here i have seeded some information and here some images about neurons and information transfers
  
  
  
  
  
  
  
  
• 

  narayana murthy

  i have more information about brain but this is not fair to post that stuff in this thread because i am started this only to discuss about artificial intelligence and neuron artificial models
  so make it more informative regarding it and let discuss about them and if any one needed really wanted to know about please send me vm i will start a thread about brain
• 

  Adivy

  @Narayana Murthy thankyou 😀I didnt know about after perceptron and before perceptron thing .That was a new information.Can you listout some of the drawbacks of mp model and how those are overcome in the Perceptron model?
  
  PS:Can anyone help me out regarding adding images or drawings in a post??
• 

  narayana murthy

  as per my knowledge mp model is totally based on neural logic of the brain see if it is a simple app we can successfully use it but if it is a large app then the number of neurons are increasing so complexity increases
  so rosenblast created a theory called perceptron theory it is done by using a logic of threshold logic like
  X1--w1-------------------------------->
  

  T----->Y=output​

  X2--w2-------------------------------->
  here the condition is if X1*w1+X2*w2 if X1*w1+X2*w2>=T then o/p will be 1
  by using this we design weights as X1 and X2 are inputs T is threshold
  then many learning rules and models have been existed but perceptron is remained as same so neural network research was made into 2 phases
• 

  Adivy

  Here I have a simple question:-
  How are these neuron models and artificial intelligence related?could anyone explain with examples?
• 

  narayana murthy

  @adivy: according to me all the artificial intelligence is done on basis of neuron logic
  see the main drawback of our machines is it can't take decisions so we are designing some machines which can think actually this was started by the scientists in mp model as we discussed in above
  our neuron functions have some logic's see our brain acts according to its training if we design a circuit which can be trained so they can also take decisions so threshold logic is designed by rosenblast in perceptron model
  this makes the sense of artificial intelligence
• 

  narayana murthy

  hey friends i am interested to add a point in this thread
  this is about types of perceptron depends on this operation
  they are
  1.discreate perceptron
  2.continous perceptron
  3.multi category perceptron
  lets see with discreate perceptron
  discreate perceptron operates in bipolar binary or unipolar binary i.e.., unipolar means 0's and 1's for bipolar -1 and +1
  we change the weights depending on the o/p obtained
  let me explain in detail
  as i said in another post weights and inputs gives o/p if it is supervised learning we change weights depending on error obtained as we assume 'd' as actual o/p and o as obtained o/p then error is d-o
  as depending on various learning rules we change weights
  next we are going to continous perceptron
  in the continous perceptron we are having the continous function as a threshold function
  that will be only difference
  next the multi category perceptron is having both the functions
• 

  Adivy

  @NM:

  this is about types of perceptron depends on this operation

  We were talking about the models right?Do you mean types of Perceptron models?and What is this operation I didnt understand?
  About Discreate perceptron,are the weights which take different values according to the type I mean bipolar or unipolar ?Is it like their values are assigned either 0,-1 and 1 depending upon the rules?
• 

  narayana murthy

  Adivy

  @NM:
  We were talking about the models right?Do you mean types of Perceptron models?and What is this operation I didnt understand?
  About Discreate perceptron,are the weights which take different values according to the type I mean bipolar or unipolar ?Is it like their values are assigned either 0,-1 and 1 depending upon the rules?

  sorry for confusing you
  i mean to say is it works on binary networks i.e..., the threshold unit will be binary function in this model
• 

  Adivy

  Just found another AI application: checkout #-Link-Snipped-#
• 

  narayana murthy

  KALYANpwn

  HAI NARAYANA MURTHY :- )
  
  please can you suggest me a text book for this topic >>Perceptron models; Discrete, continuous and multi- category...AND training algorithms also!!
  i've reffered 2 to 3 text books i didn't got full information, but i've got some idea about this topic which is posted by you NM >>30th July 2011 08:46 PM
  
  please help me sir
  
  Thnks and regards
  KALYANpwn
  
  

  i am refering to zurada
  if you want you can check it or if you want about biological nn's its nice to refer simen haykin
• 

  KALYANpwn

  plz tel me i did not find kolomogorov theorem from 4th unit >> can yu knw anyone plz tel me?
• 

  Manish Goyal

  narayana murthy

  @adivy: according to me all the artificial intelligence is done on basis of neuron logic
  see the main drawback of our machines is it can't take decisions so we are designing some machines which can think actually this was started by the scientists in mp model as we discussed in above
  our neuron functions have some logic's see our brain acts according to its training if we design a circuit which can be trained so they can also take decisions so threshold logic is designed by rosenblast in perceptron model
  this makes the sense of artificial intelligence

  We had a separate subject on this topic in last sem , it was quite interesting to learn it
  
  The concept of neural network is far above than artificial intelligence , these are commonly used for weather forecasting, sales forecasting etc
  
  Looking forward to learn more from this thread
• 

  narayana murthy

  @goyal: my faculty already told that we are only learning basics in this subject
  so you may be correct
  @kalyan: that theorem is not used mostly in practice so many of books dont give that i believe my faculty told that in notes
  so i will post as soon as possible
• 

  KALYANpwn

  //@kalyan: that theorem is not used mostly in practice so many of books dont give that i believe my faculty told that in notes
  so i will post as soon as possible//
  
  #-Link-Snipped-# >> YOU ROCK dude :- )
  
  TFS!! waiting for ur post
  
• 

  narayana murthy

  @kalyan: it is applicable for multi layer feed forward n/w which has m inputs and 2m+1 outputs
  the formula of activation is a big one
  check here for more details about it
  Static and Dynamic Neural Networks: From Fundamentals to Advanced Theory - Madan Gupta, Liang Jin, Noriyasu Homma - Google Books
• 

  KALYANpwn

  @narayana my frnd:- )
  
  THNK YOU SO MUCH NARAYANA 😀 you rock dude
  but problem is how can i download books from google books😔😔.. i tried so many times...i get frustrated frnd>> plz show me alternate way
• 

  narayana murthy

  @kalyan: you can't download directly from google books try on websites for Ebooks so you can have a little chances of them
• 

  KALYANpwn

  #-Link-Snipped-# >> YOU ROCK dude :- )
  
  @@ can you suggest me any website name or links>> plz of dont mine help me :- ) i tried but pichi pichi links are open avutunaye frnd 😔
  
  thanks frnd :- )
  
• 

  KALYANpwn

  #-Link-Snipped-# >> YOU ROCK dude :- )
  
  //@@kalyan: you can't download directly from google books try on websites for Ebooks so you can have a little chances of them ///
  
  hm if you dont mine can you suggest me a website for our electrical books easy download..really i dont know tht's why i'm asking frnd>> please do needfull
  
  Thnkas in adnvc :- )
  UR MY FRND>>
• 

  narayana murthy

  hey friends i am back with more information soon associative memories will be posted if i get some time
• 

  narayana murthy

  k friends i think i got enough time
  here is it before going to ann memories i want to differenciate ann's and computer memories
  our computers has a limited space of memory but a human memory is unlimited storage of memory
  if we able to make perfect ann then that will be of high amount of storage
  and coming to processing a computer can make a serial computation but a brain and ann's can make a parallel computation by remembering them so we can make process much faster so we can use them much in pattern recognition
  while coming to ann memories there are 2 types of memories
  auto associative and hetro associative
  the recognition of auto associative is good because if a signal is coming with noice
  this memory can check it and give relavent one
  but hetro gives completely different one
  on this there is another given hop it is known as hop field associative memory
  more information will be given soon
• 

  parth 127

  can u explain me all models that come in neural network
• 

  narayana murthy

  parth 127

  can u explain me all models that come in neural network

  what type of models do you mean training algorithms or design models
  if design is concern base for all the models are perceptron
  all the others are modification of perceptron
  and if you want anything on specific specify clearly
• 

  parth 127

  i need to understand hopfield and pitts model designing and also training
• 

  narayana murthy

  i think thats not hopfiled and pitts model it is mcculloch and pitts anyway hopfield network is totally depending on auto associative memory refer to following site its given total information about hopfield network #-Link-Snipped-#
  coming to mcculloch and pitts model it is depending on weight and threshold logics i think i have given about that in this thread
• 

  KALYANpwn

  CAN YOU Explain how Back Propagation Network is used as dierentiator??
  NM i want ANSWER.please
• 

  narayana murthy

  KALYANpwn

  CAN YOU Explain how Back Propagation Network is used as dierentiator??
  NM i want ANSWER.please

  sorry buddy i dont know much about Back propagation network i know only how it works
  try in books and please post here
  by the way what is dierentiator
• 

  narayana murthy

  hey friends
  today i have noticed something in text called artificial neural networks by zurada
  if anyone interested or having that text please refer page no 9 and 8th chapter
  8th chapter consists of all the apps of neural networks they are really interesting check that friends
• 

  KALYANpwn

  What are kohonen’s self organizing maps
  nm tel me>>
• 

  parth 127

  i think this may help you to understand kohenen well
  
  #-Link-Snipped-#
• 

  parth 127

  ya the neural network is a purely human based network its a similar network of human brain
• 

  KALYANpwn

  Tfs 😀 frnd
• 

  KALYANpwn

  HELO HELP ME OUT> what is fuzzy logic and fuzzy set and crisp set? what is the difference b/w nn and fuzzy?
  
  
  --
  thnks in advnc
• 

  narayana murthy

  KALYANpwn

  HELO HELP ME OUT> what is fuzzy logic and fuzzy set and crisp set? what is the difference b/w nn and fuzzy?
  
  
  --
  thnks in advnc

  actually fuzzy set is a set which has the membership values for a set
  the main diff can be explained with an example
  see these sets
  A={(1,2,3,4)}
  A1={(1,.2),.......,(4,.9)}
  the first set A is called as a crisp set and A1 is a fuzzy set
  where 2nd element is known as membership values of the 1st element in the set
  coming to logic
  fuzzy logic is the logic which is designed with members in the relations or the properties of a object which is unknown
  there is nothing comparing with nn and fuzzy logic
  nn is a network connecting and training for a specified function
  fuzzy logic is a logic for a given network to design the possibilities of occur like load forecasting
  hope i am clear
• 

  rashidsohail

  I said it was a real younger thing of what I unnatural in DSP Usage,so I meet common what I knew.I dont hit bottomless knowledge on the issue.And nigh the use there are numerous equivalent:-
  Ornamentation recognition or identifying patterns or trends in data(One of the important apps),
  Robotics,espial and pursuit of kinetic items ,happening perception ,etc
  Its widely used in Near info.
• 

  shenglu

  I mean, according to bipolar or unipolar? It is like their values ​​are 0, -1 and 1 is assigned depending on the type of rule, to take different values ​​of weight?
• 

  narayana murthy

  shenglu

  I mean, according to bipolar or unipolar? It is like their values ​​are 0, -1 and 1 is assigned depending on the type of rule, to take different values ​​of weight?

  sorry i didn't understand your doubt please clear your question
• 

  Rex 876

  I am interested but I have no idea about it... 😔 ​
• 

  narayana murthy

  Rex 876

  I am interested but I have no idea about it... 😔 ​

  hehehe your on last buddy almost all topics are cleared check all the post you can get an idea
• 

  narayana murthy

  hey i have listened that neuron is made of an open loop opamp if it is so why we are so lagging in this concept anyone can explain the reasons please
• 

  narayana murthy

  can i close this thread or anyone want to take or give more information
  just give me a suggestion
  if no one is interested i will close this thread
• 

  Jeffrey Arulraj

  Yo pal once in a while an age old thread is bumped up and we find treasures hidden there pls drop the idea of closing this thread It will be a great help
  
  PS: I am new to this and it is going to take me a couple of days to post a valid doubt that crosses my mind after reading the post here in this thread
• 

  narayana murthy

  ok no problem buddy
  if anyone interested to continue go on with your doubts and i am always here to help and to gain knowledge by getting some posts on this
• 

  soni.sapan

  If some one want to know about it via video lecture then check it out
  Coursera
  It is very interesting way to learn
• 

  ahmed sarfraz

  neural network or i can typically say artificial neural networks (ANN) are a very important part of artificial intelligence and DSP. they are and can be used in any if the fields including electrical. mechanical, civil, computer sciences, etc or any biological field including cardiology, neurology, dermatology, optics, etc. these also find application in field of machines, control sections, rocket science, aerospace engineering, automation, etc
• 

  ahmed sarfraz

  even i have done a good research in this topic and have even tried weird thing with NN like controlling the speed of a DC motor, maintaining library records automatically, finding out particular patterns, and making different and desired outputs no matter what kind of inputs we give.