Solving the problems

Solving the problems
1) We found the knowledge from the report of Manchester PHD.
The chromosome consists n bits delays and n bits weights .
For example 4bits delays and 4bits weights

delays_encoded(i,:)=Population(chromosome,[gene:gene+3]);

weights_encoded(i,:)=Population(chromosome,[gene+4:gene+7]); 

gene=gene+8

2) 4 bits delays  

0000    1               

0001    2              

0010    3

0100    4

1000    5

0011    6

0101    7

0110    8

1001    9

1010    10

1100    11

1011    12

1101    13

1110    14

 0111   15

 1111   16         

3) 4 bits weights  

 0000   8             

0001    7              

0010    6

0100    5

1000    4

0011    3

0101    2

0110    1

1001    0

1010    -1

1100    -2

1011    -3

1101    -4

1110    -5

 0111   -6

 1111   -7

4) we found we should also change the delay and weight value in the XOR train code. So code will be successful.



5)

This figure we found came from the report of PHD. We thought  was very important. So we show this figure.


                

Run the code and The troubles we got

The supervisor Marsland promised he would give us the code. We should change the parameters in codes and get the results according to his notes.  We got the code in 19th Feb 2013. The code was 3 delays and 3 weights. We run it and found it successfully. The figure will be shown in below.

3 delays  3 weights

In the next, we tried to change the codes. But we come across some problems.

1) how to set delay encode

2) how to set weight encode

3) how to use limited precision to show the delay and weight.

4) After changing these datas, the code was still not successful. 

5) There are many codes we do not know the meaning.

Some hints for the project

After a week, we all thought we could not understand the project if our supervisor Marsland did not help us. We seek the our supervisor by the email. Fortunately, our supervisor promised us having a meeting in 18 Feb 2013. In this meeting, our supervised  told us some knowledge. We could learn much and understand better because we had learned a bit of these knowledge.

Our tutorial give us four instructions.

The first instruction is neural network. We have talked it before. But there still have some hints. 

1) connection = synapse

2)connection strength =synaptic weight

3) artificial neural network can be "trained" by synaptic weight to a ideal output for a pattern of input.

The second instruction is spiking neural network. In this part, we know the architecture of spiking neural network and all kinds of  figures about weight, delay and output spike.

The third instruction is limited precision. We know how to express the limited precision in the Matlab. 

For example, 3 bits for delays    e.g. 1ms------------8ms

                                                                     1,2,3,4,5,6,7,8  

                          3 bits for weights  e.g. allowed values can be ( -3,-2,-1,0,1,2,3,4)

                                                                                                     or (-1.5,-1,-0.5,0,0.5,1,1.5,2)

The fourth instruction is XOR problem and the Genetic algorithm.  The part was neglected because of the limit of time. But it dose not mean we can not finish our project.

the buildup model -----------preliminary understanding

  After we inquired many relevant materials, we found there had numerous scientists were dedicate to the study neural network from 1943 to 1990. In term of the time, the neural network had been completed perfectly step by step. So we can select the most simple model to interpret the problems.

Figure 1 a simple artificial network net

  Figure 1 is a simple artificial network net with two input neurons (x1, x2) and one input neurons (y). The inner connected weights are given by (w1, w2). It is the simple and no-branch neural network net.

  In the next, we should find the biological neural network net. Then we compare these two neural network each other.

Figure 2 a biological neuron

  After we compared each other, we can get the following figure.

Figure 3 association of biological network with artificial network

According to these materials, we think we can put the data of reality transforming the data of MATLAB.

Year 2 project ‘Limited precision spiking neural networks using MATLAB’

 ‘Limited precision spiking neural networks using MATLAB’ is year 2 mini project in liverpool university . It will be done by GANG. HU and JIN.BAI in the next several  weeks. And our supervisor is JOHN MARSLAND who will give us guidance and help.
  We feel it is the difficult question when we firstly face it. The reason is divided into two parts. Firstly,  this project relates to many diverse fields of knowledge. For example, if we can control biological knowledge , mathematical knowledge and physical knowledge, then they all can help us better understand the biological neural net transforming into artificial neural net. Secondly, we should use the MATLAB to  simulate it. It is difficult to us the data of reality world turning into the data  in the MATLAB.
  Facing to these problems, we decide to solve the first problem.  We all think we should have a model of the neural networks to complete the project in the first. It is the key to the project.