Quantum random, self-modifiable computer

Abstract

We describe a computing machine, called a quantum random, self-modifiable computer, that uses self-modification and randomness to enhance the computating power. Sometimes it is called an ex-machine, derived from the latin extra machinam because its can evolve as it computes so that its complexity increases without an upper bound. In an embodiment, an ex-machine program can compute languages that a Turing or standard machine cannot compute. In an embodiment, the ex-machine has three types of instructions: standard instructions, meta instructions and random instructions. In an embodiment, the meta instruction self-modify the machine as it is executing so that new instructions are added. In an embodiment, the standard instructions are expressed in the C programming language or a hardware description language such as VHDL. Random instructions take random measurements from a random source. In an embodiment, the random source produces quantum events which are measured during the machine's execution. In an embodiment, an ex-machine receives a computer program as input, containing only standard instructions. In an embodiment, the ex-machine combines its random instructions and its meta instructions to self-modify the ex-machine instructions, so that it can evolve to compute (i.e., verify) the correctness of the computer program that it received as input. In an embodiment, an ex-machine uses its meta instructions and random instructions to improve its machine learning procedures as the ex-machine is computing. In an embodiment, machine computation that adds randomness and self-modification to the standard digital computer instructions has more computing capability than a standard digital computer. This capability enables more advanced machine learning procedures where in some embodiments meta instructions and random instructions improve the machine learning procedure, as it is executing. In an embodiment, differential forms, the curvature tensor, and curvature of saddle points are used to help self-modify and improve an initial, standard gradient descent method.