Self-reinforced Meta Learning for Belief Generation

Contrary to common perception, learning does not stop once knowledge has been transferred to an agent. Intelligent behaviour observed in humans and animals strongly suggests that after learning, we self-organise our experiences and knowledge, so that they can be more efficiently reused; a process that is unsupervised and employs reasoning based on the acquired knowledge. Our proposed algorithm emulates meta-learning in-silico: creating beliefs from previously acquired knowledge representations, which in turn become subject to learning, and are further self-reinforced. The proposition of meta-learning, in the form of an algorithm that can learn how to create beliefs on its own accord, raises an interesting question: can artificial intelligence arrive to similar beliefs, rules or ideas, as the ones we humans come to? The described work briefly analyses existing theories and research, and formalises a practical implementation of a meta-learning algorithm.