12.07.2015 Views

Neural Networks - Algorithms, Applications,and ... - Csbdu.in

Neural Networks - Algorithms, Applications,and ... - Csbdu.in

Neural Networks - Algorithms, Applications,and ... - Csbdu.in

SHOW MORE
SHOW LESS
  • No tags were found...

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

1.3 ANS Simulation 37not know (or care) whether the connection list implements a fully <strong>in</strong>terconnectednetwork or a sparsely connected network. It simply follows the list to the end,then moves on to the next unit <strong>and</strong> repeats the process.1.3.5 Extension of ANS Data StructuresNow that we have def<strong>in</strong>ed two possible structures for perform<strong>in</strong>g the <strong>in</strong>putcomputations at each node <strong>in</strong> the network, we can extend these basic structuresto implement an entire network. S<strong>in</strong>ce the array structure tends to be moreefficient for comput<strong>in</strong>g <strong>in</strong>put values at run time on most computers, we willimplement the connection weights <strong>and</strong> node outputs as dynamically allocatedarrays. Similarly, any additional parameters required by the different networks<strong>and</strong> associated with <strong>in</strong>dividual connections will also be modeled as arrays thatcoexist with the connection-weights arrays.Now we must provide a higher-level structure to enable us to access thevarious <strong>in</strong>stances of these arrays <strong>in</strong> a logical <strong>and</strong> efficient manner. We can easilycreate an adequate model for our <strong>in</strong>tegrated network structure if we adopt a fewassumptions about how <strong>in</strong>formation is processed <strong>in</strong> a "st<strong>and</strong>ard" neural network:• Units <strong>in</strong> the network can always be coerced <strong>in</strong>to layers of units hav<strong>in</strong>gsimilar characteristics, even if there is only one unit <strong>in</strong> some layers.• All units <strong>in</strong> any layer must be processed completely before the CPU canbeg<strong>in</strong> simulat<strong>in</strong>g units <strong>in</strong> any other layer.• The number of layers that our network simulator will support is <strong>in</strong>def<strong>in</strong>ite,limited only by the amount of memory available.• The process<strong>in</strong>g done at each layer will usually <strong>in</strong>volve the <strong>in</strong>put connectionsto a node, <strong>and</strong> will only rarely <strong>in</strong>volve output connections from a node (seeChapter 3 for an exception to this assumption).Based on these assumptions, let us presume that the layer will be the networkstructure that b<strong>in</strong>ds the units together. Then, a layer will consist of arecord that conta<strong>in</strong>s po<strong>in</strong>ters to the various arrays that store the <strong>in</strong>formationabout the nodes on that layer. Such a layer model is presented <strong>in</strong> Figure 1.22.Notice that, whereas the layer record will locate the node output array directly,the connection arrays are accessed <strong>in</strong>directly through an <strong>in</strong>termediate array ofpo<strong>in</strong>ters. The reason for this <strong>in</strong>termediate structure is aga<strong>in</strong> related to our desireto optimize the data structures for efficient computation of the net 4 value foreach node. S<strong>in</strong>ce each node on the layer will produce exactly one output, theoutputs for all the nodes on any layer can be stored <strong>in</strong> a s<strong>in</strong>gle array. However,each node will also have many <strong>in</strong>put connections, each with weights unique tothat node. We must therefore construct our data structures to allow <strong>in</strong>put-weightarrays to be identified uniquely with specific nodes on the layer. The <strong>in</strong>termediateweight-po<strong>in</strong>ter array satisfies the need to associate <strong>in</strong>put weights withthe appropriate node (via the position of the po<strong>in</strong>ter <strong>in</strong> the <strong>in</strong>termediate array),

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!