Anders Modén

Just got published in Accu’s Overload paper. Read the article here. Really interesting that so many people want to discuss this new theory about deep genetic learning and training.

Imagine a set of instructions just like an assembler program. Lets assume in one variant that there are only 4 possible instructions. A C G T. The order and the combinations of these instructions can build any kind of complex sequence. And actually any kind of complex creature as our DNA is exactly what i am talking about. Only four instructions are the basic program for all living creature. Including man. Now in a digital lifetime lets assume that the same principle could be used to construct an artificial creature. A software complex enough to solve narrow tasks. And there is such a construction. I have called it CAL Cortex Assembler Language. It has about 30 instructions and it can be used to build any kind of neural network or state machine. The interesting part is that you can program it to solve tasks but also train it to handle tasks that we dont know How to program sequentially. And now to the most exciting statement. The code can also be evolved using the same principles that are the base of our excistence. An evolutionary build code by a genetic algorithm can build any kind of intelligent software. You have a problem. You dont know the exact solution but you know some expected goals. And that is enough to have an evolution of an artificial intelligence https://www.linkedin.com/feed/update/urn:li:activity:6351492545985548289

The Duality between Genetic Algorithms and Deep Learning I claim that it is possible to solve any deep learning problem better using a new technique I have discovered a method that combines back propagation and genetic evolution. It trains faster and more accurate. http://www.tooltech-software.com/CorTeX/Deep_Genetic_Training.pdf