Abstract
The free-energy principle, a candidate unified theory of learning and memory, predicts that the neurons, synapses, and neuro-modulators work so as to minimize free energy. However, its neural and synaptic bases remain largely unsupported electrophysiologically. Here, we proposed a novel theory linking between the information-theoretical principle and biologically plausible spike-timing dependent plasticity (STDP) modulated by neuro-modulators, called mSTDP. We derived mSTDP rule from the free-energy function, and numerically and analytically showed that dopamine (DA) and noradrenaline (NA) modulate the accuracy of principal component analysis performed by STDP. Specifically, synaptic connection strengths were established so that the neural output represents the first principal component (PC1) in the absence of DA/NA. The DA up-regulation affects all synaptic connections of the neuron attaining a similar amplitude, while NA makes only a connection dominant like a winner-takes-all manner. These results are consistent with knowledge of electrophysiological studies, which supports the validity of the free-energy principle and has potential to be applied for computational psychiatry.
