On Heuristic Randomization and Reuse as an Enabler of Domain Transference

The solution of NP-hard problems requires the use of one or more explicit or implicit heuristics as a practical measure. Quantum computers promise to make this practical for O (2ⁿ) problems or less, but have yet to deliver a solution to a single NP-hard problem. The question addressed by this paper is whether domain transference and reuse of problem-solving knowledge can be mediated through the reuse of heuristics, and, if so, the extent to which such transference may occur in the solution of NP-hard problems. Neural networks have zero domain transference on account of their inability to represent modus ponens. Similarly, CBR, deep learning, EP, GAs, SVMs, the predicate calculus, learning via conventional expert systems, and all other machine learning technologies are unable to theoretically or practically mediate domain transference because they don´t respect randomization as the core underpinning technology. The paper offers a constructive proof of the unbounded density of knowledge in support of the Semantic Randomization Theorem (SRT). It details this result and its potential impact on the machine learning community.