NeuroStrategy FAQ

12. What is a mesh?

An artificial neural mesh (ANM) is a further development and combination of elements of artificial neural networks (ANN). In conjunction with the resources of a computer, these software components form a structure that is similar to the brains of higher organisms.

Like the brain, a neural network also consists of neurons or nodes that are interconnected over synapses or links. In the brain, stimuli are transmitted via electrical charges. The stimuli can be delayed, amplified or weakened by the synapses before reaching the targeted neuron. In this neuron, the transmitted energy will lead to a reaction (activation). The neuron, for example, can block the stimulus or transmit it via further synapses to other nodes.

In an artificial neural network, external stimuli are likewise transmitted through a mesh of links and neurons in order to prompt a certain reaction in the system. In this case, however, the transmission of the stimulus occurs in a mathematically determinate way. The type and strength of the signal must be encoded in a number. This number is transferred to the target neuron by being multiplied by a so-called weight. Through this multiplication, the stimulus can be amplified or weakened arbitrarily. Many stimuli from different sources may arrive at a node at the same time. In their entirety, they make up the activation of the neuron, that is, a certain quantity of activation energy. This can now in turn be transmitted, delayed, amplified, weakened or blocked by synapses.

The transmission of stimuli through an artificial neural network occurs in a structured way and eventually leads to an activation in one or several output neurons. Depending on the application, the activation in a particular output neuron represents the encoded reaction of the network to the external stimulus. The user can evaluate this reaction and provide feedback to the network regarding the quality of the reaction. According to certain algorithms, the weights within the network are then modified. The network will now react differently than before to the same stimulus. In this way, neural networks are capable of learning.

There are very many different types of neural networks for the most diverse fields of application. Some of these networks require a teacher (supervised learning). Others learn on their own, since their structures provide them with options for determining the quality of the reactions by themselves.

The meshes developed by Ivorix integrate a variety of these types of networks into universal software components that are able to learn both as self-organizing mechanisms and through guided training. In addition, the meshes are equipped with functions for automatic checking that are designed to promote the formation of intelligent structures.

The meshes are set up in such a way that they can be encoded in blueprints similar to the genome of a living being. They also have communications interfaces that allow them to cooperate with other meshes, in order to solve a problem that exceeds their individual capacity, for example. Due to these characteristics, the meshes can be trained and tested in populations. They are able to propagate by generating a new mesh genome from the genomes of the parents. This genome will feature character traits of both parents and can be decoded to become a useful child mesh.

Like all neural networks, meshes are initially "dumb". Their user - or the developer - must provide a mathematical formulation of the problem to be solved, establish the encoding of the stimuli and reactions and create an evolutionary environment, within which a population of meshes can develop that will solve precisely this and only this problem. Our goal at Ivorix is the development of applications that will relieve the user completely of these complex tasks. It is clear from the above descriptions that the meshes unfortunately do not speak our language at all. The environments within which they thrive bear little or no resemblance to our own environment. The applications developed by Ivorix are to function as interpreters such that the user can formulate the problem in his own language and that the software responds in the same intelligible idiom.

Thus, our artificial neural meshes are a universally applicable basic technology. The formation of strategies for trading with financial instruments is only one of countless possible applications.