SourceForge.net Logo Neural Entity
 Support This Project
[Home]
[Up]
[Presentation]
[Definitions]
[Object Structure]
[Evaluation Procedure]
[Learning Procedure]
[Create Pattern]

 

2. Definitions

In a neural network, as we know it, the flow of information can be represented with a flood of information from the neurons of a set of points to another set. As the information propagate, it will at the end only affects the output nodes having no direct impact on another network. Even more, the network will be supposed to function as we believe it should - and to achieve this goal, we train the network intentionally to adapt its internal connection filter (weights of the links) until the solution provided by the output nodes satisfy the designed scheme. This learning procedure or initiation procedure is to alter the young interconnection definition in such a way that it matches the desired result.

A neuron Entity is a pool of neurons. Are they connected or not is not our concern as we assume they are not connected to serve our answer. The entity is the concept we will start with in order to define a neural process. the trivial definition of a neuron entity is a simple vector of neuron and you already have the entity - not quite practical form I guess.

For a specific processing, we will start to talk of a neural entity when we are working a solution. We do not concern of a predefined set of neurons, the amount can even be randomize between a defined boundary - neurons may be dynamically added or even removed from the entity; The only that matter is we have pointed out some input and  output neurons. We know that the input will be modified and we monitor the output neurons. In the case an output does not satisfy the reality, we will adapt and learn to balance correctly our pool of interaction.

A neural process could be seen as a neural network but I I prefer to see it as a function interface of the world: in fact our input neurons are or immediately connected to external sensors or collect the result of another neural process. It provides a none/have answer as the result of an interconnectivity - like the result of a mathematical operation of the conclusion of report. A neural process is linked to one and only one neural patterns but but execute and evaluate several profiles.

We also need to make a clear distinction between a neural pattern and the neural profile. Indeed, the pattern describe a set on neuron and their interconnection as the neural profile specifies the weights of the dendrites. Meaning we can easily apply a well known pattern  to a new situation, create the missing connection is necessary then adjust the weights to fit our situation. In a way, a neural pattern will be used by several profiles.

We will also be confronted with two different kind of modification of a neural process: when we learn and when we adapt. The learning procedure of a neural process has for aim to modify/alter the connectivity in order to satisfy and answer - we even may create a new connection to reach our goal. To adapt a neural process if more to adjust the weights of a neural process without altering the interconnection.

An error is the difference between a computed result and an expected one. Reducing the error will mainly be the tasks of the learning process.

The none/have values of a neuron is somewhat quite special in this environment. We will process information but the absence of information has a bigger meaning than the dual answer (this is why I do not call this as true/false). Indeed, we see the neuron as followed: it is an organism that propagates a signal (modifying the signal strength) - how can we then see the absence of a signal? Well in fact we should not concern ourselves with this as we will only be concerned of the signal. The absence of information could not be propagated and therefore we cannot propagate a none as we cannot differentiate in our concept from the rest state (doing nothing). In order to avoid this kind of ambiguity, we have to formulate our sentence differently: we will only use positive sentences (we make  the difference between 'not good and bad' between 'nothing and not anything' etc etc....).

For example what does mean the information 'not good'? It surely does not mean bad, maybe we do not know yet. Anyway what kind of information does it gives us? in fact it provides us with no information at all - only the absence of information associated to good and this does not help. The same information is processed in a neural entity only the presence of information but thi8s does not mean that the absence is not to be taken into account in a neural  process....

 

Should you have any comments or ideas, please let me know, you can always mail me at C.Hannosset