As in the case of autonomous vehicles becoming more reality than a work of science fiction, so too is the inevitability that machines will also become more independent in the future, according to a study published in EPJ B.

Noting that machines can evolve over time to independently perform well-defined tasks—ranging from nanotechnology to biological applications—, study authors Agustín Bilen and Pablo Kaluza from Universidad Nacional de Cuyo, Mendoza, Argentina, explain that the autonomous systems don’t need an external tutor.

Additionally, the systems also do not report to a central unit created to alter what the system learns based on their performance. Instead, to increase independence, the researchers incorporated delayed dynamics and a feedback loop with the system's performance. The delayed dynamics provide historical information about the system, detailing past relationships between its structure and performance while the feedback loop presents information on the system's actual performance in terms of accomplishing a desired task.

Initially applying the approach to a neural network charged with classifying a number of patterns (yielding 66% robustness), researchers believe that the findings can be applied in instances where a piece of hardware can learn a task without external control or a central processing unit (for instance, analogue electronics).

To read more, click here.