Weight training calculations - artificial intelligence, Computer Engineering

Weight Training Calculations -Artificial intelligence:

Because we have more weights in our network than in perceptrons, first we have to introduce the notation: wij to denote the weight between unit i and unit j. As with perceptrons, we will calculate a value Δij to add up to each weight in the network afterwards an example has been tried. To calculate the weight changes for a specific example, E, first we begin with the information regarding how the network would perform for E. That's, we write down the target values ti(E) that each output unit Oi  would produce for E. Note that, for categorization problems, ti(E) will be 0  for  all  the  output  units  except  1,  which  is  the  unit  associated  with  the  right categorisation for E. For that unit, ti(E) will be 1.

736_Weight Training Calculations.png

Next, example E is propagated through the network so we may record all the observed values oi(E) for the output nodes Oi. At the same time, we record all the calculated values hi (E) for the hidden nodes. For each output unit Ok, then, we calculate its error term as follows:

1966_Weight Training Calculations1.png

The error terms from the output units are utilized to calculate error terms for the hidden units. In actual fact, this method gets its name because we propagate this information backwards through the network. For each hidden unit Hk, we calculate the error term in following manner:

In English language, this means that we take the error term for the entire output unit and multiply it by the weight from hidden unit Hk to the output unit. Then we add all these together and multiply the sum by hk(E)*(1 - hk(E)).

Having calculated all the error values connected with each unit (hidden and output), now we may transfer this information into the weight changes Δij between units i and j. The calculation is as following: for weights wij between input unit Ii and hidden unit Hj, we add on:

[Remembering that xi  is the input to the i-th input node i.e. E; that η is a small value known as the learning rate and that δHj is the error value we calculated for hidden node Hj utilizing the formula above].

For weights wij among hidden unit Hi and output unit Oj, we add on:

2491_Weight Training Calculations2.png

[Remembering that hi (E) is the output from hidden node Hi when example E is propagated through the network and that δOj is the error value we calculated for output node Oj utilizing the formula above].

2128_Weight Training Calculations3.png

Each alteration Δ is added to the weights and this concludes the calculation i.e. E. The next instance is then used to tweak the weights further. As with perceptrons, the learning speed is used to ensure that the weights are just moved a small distance for each particular example, so that the training for earlier examples is not lost. Note down that the mathematical derivation for the above calculations is based on derivative of σ that we discussed above. For total description of this, see chapter 4 of Tom Mitchell's book "Machine Learning".

Posted Date: 10/2/2012 6:31:48 AM | Location : United States







Related Discussions:- Weight training calculations - artificial intelligence, Assignment Help, Ask Question on Weight training calculations - artificial intelligence, Get Answer, Expert's Help, Weight training calculations - artificial intelligence Discussions

Write discussion on Weight training calculations - artificial intelligence
Your posts are moderated
Related Questions
er table for hospital management system

Q. Library of functions of parallel virtual machine? PVM offers a library of functions libpvm3.a, that application programmer calls. Every function has some specific effect in

draw input and output charectoristics of BJT and justify CE configuration provides large current amplification

Discuss about Translation table encryption technique to ensure secured transactions onto the net? Translation table: In this process each chunk of data is used like an of

What is application analysis? The purpose of analysis is to understand the problem so that a correct design can be constructed. The application analysis focuses on major applic

The history file, history.txt, is the branch history of a run of the gcc compiler (the first field is the address of the branch instruction, the second field is the target branch a

The Function code currently active is ascertained by what Variable? The function code at present active in  a Program can be ascertained from the SY-UCOMM  Variable.

1)   Discuss various types of persuasive communication you might be needed to write or present in your professional and personal life. 2)    Describe two ways to organize a resume

What is the meaning of Proper programming Proper programming of the ports of the company's Web server through detection of IP addresses could be an excellent strategy or solut

Q. What do you mean by True Complement Method? Explain in detail. Q. Show IEEE format for (12.75)10. Q. What are the various ways to represent Negative Numbers in computer sy