What Is The Main Goal Of A Natural Neural Network?

Looking at how brain cells are trying to form connections with each other, one may ask if solving some task required for survival of any natural neural network (NNN) is really what that very NNN is trying to do. What if all (and only) goals of an NNN is to form connections? Emerged ability to tell cats and dogs apart, or memorize multiplication table, or whatever human beings do, can be just a side effect, because those who didn’t know that 2 * 2 = 4 didn’t live long enough to produce offsprings?

In approximately a month Qubic is changing its mining algorithm. The current one generates artificial neural networks (ANNs), having static structure, trying to solve some artificial task. The new algorithm aims to generate ANNs which change their structure while trying to solve a task. What do we expect from this? First of all, we’ll see if “fine-tuning” of such ANNs will be improved with the existing approach mentioned in this article. Second, we’ll get some insight about the hypothesis of NNN not worrying about anything except forming connections. Third, we’ll make a step closer to hypercomputations conducted by self-transforming neural networks, which is probably the key to self-awareness.

The reference implementation of the new mining algorithm is still being developed, we expect to have it ready in a couple of weeks, so others will have some time to write highly optimized versions.