The application

INNE provides tools to design and simulate different neural network architectures. A net editor allows to define the network layout by displaying interactively nodes and connections, with any architecture. Thereafter there is the possibility to associate to it a computational model and a learning model. The neural models are implemented so far:

Boltzmann machines and Hopfield networks, combinatoric optimization models, example of networks that do not change dynamically their structure;
Back-error propagation networks, a supervised model, where learning takes place under supervision, that is confronting the output with the correct answer;
Hebbian networks, an unsupervised model, where learning takes place with no feed-back from the environment and in presence of redundancy in input data;
Simple competitive networks, an unsupervised competitive model, where single neurons compete for being the one to fire,
Kohonen feature maps, an unsupervised competitive model, where single neurons compete for being the one to fire, as in the previous model, that moreover performs a feature mapping.

At the time in which the association between the network and the model is established, the network is checked throughout to assure that there are no inconsistencies with the model requirements; otherwise a warning will advise users that the layout has to be changed in order to be able to accomplish the given model. If the net requires a learning algorithm, a training phase is necessary to establish appropriate connection weights. The training phase depends on the model associated to the network and is performed by submitting several examples to the net in a given format. In the examples, input samples together with their expected output are fed into the net; from them the network learns how to adapt in order to respond correctly even to general inputs, not previously submitted.

Back to index