The Neuroblast difference

There are many artificial evolution projects and there are many neural network projects, and there are many neural network evolution projects. Here is why Neuroblast is so revolutionary.

Old school artificial neural networks

Artificial neural network (ANN) technology has not changed much in the past few decades. In fact, even the best of today's ANNs are very little more than correlation algorithms, able to determine the best match between different sets of data.

A typical ANN, shown below, consists of one or more layers, or rows of connected processing elements. Every layer is fully interconnected, and signal processing is feedforward: each neuron in a given layer receives input signals from every neuron in the layer behind it, and transmits its output (some mathematical operation performed on the input) to every neuron in the layer ahead of it. Some ANNs are more elaborate than this, and some have hundreds or even thousands of internal neurons, but this is the essential paradigm.

ANNs have a major limitation. They can operate only in two separate stages: first the programming (also called training) stage, during which the numerical strength of each of the connections is determined, and second the execution stage, during which the network is ready for use. As soon as programming is complete, learning ends, and the processing capability of the ANN is permanently frozen. An ANN in the execution stage will never again learn anything new.

If the network has been trained to recognize text characters, for example, the introduction of a new character into the alphabet would require that the ANN be wiped clean and retrained from scratch.

ANNs can also be trained using evolutionary algorithms. This means random mutations of the connection strengths and perhaps the number of neurons in the middle layers, followed by selection of the best-performing variants. ANNs trained in this manner suffer the same limitations as traditional ANNs. They can never learn anything from experience.

ANN technology has been quite successful in data classification problems, especially handwriting recognition. But the functional resemblance of these networks to biological brains is extremely superficial. Even the most sophisticated of ANNs do not exhibit anywhere near the complex dynamics of even the simplest of real brains. And they certainly show no promise of ever becoming as capable.

A new paradigm is clearly required.

Neuroblast's neural algorithms

An artificial evolution engine that matches the creative power of biological evolution must do much more than make fine-tuned adjustments to a set of predefined parameters. It must be capable of inventing entirely new kinds of parameters and structures. It must be capable of evolving types, arrangements, and dynamics of processing elements never before imagined by science or begotten by nature. It must be capable of evolving systems from as simple as ANNs to as complex as human cortices and much much more.

Neuroblast has created just such a system, and the brains that have emerged from its virtual world are unlike anything ever seen, and have capabilities to process, learn, and adapt in complex, dynamic, noisy, and unpredicatable environments that will revolutionize A.I.

And Neuroblast's brains are now ready for action.