Cost minimisation and Reward maximisation. A neuromodulating minimal disturbance system using anti-hebbian spike timing-dependent plasticity

摘要

In the brain, different levels of neuro-active substances modulate the behaviour of neurons that have receptors for them, such as sensitivity-to-input, Koch (1999). An artificial neural network is described that learns which actions have the immediate effect of minimising cost and maximising reward for an agent. Two versions of the network are compared, one that uses neuromodulation and one that does not. It is shown that although neuromodulation can decrease performance it agitates the agent and stops it from over-fitting the environment.