Hebbian learning deduced from the stationarity principle leads to balanced chaos in fully adapting autonomously active networks

GROS, CLAUDIUS; TRAPP, PHILIP; ECHEVESTE, RODRIGO

Sardinia

Conferencia; 26th International Conference on Artificial Neural Networks; 2017

Neural information processing includes the extraction of information present in the statistics of afferent signals. For this, the afferent synaptic weights are continuously adapted, changing in turn the distribution of the postsynaptic neural activity y , which is in turn dependent on parameters θ of theprocessing neuron. The functional form of pθ(y) will hence continue to evolve as long as learning is ongoing, becoming stationary only when learning is completed. This stationarity principle can be captured by the Fisher information of the neural activity with respect to the afferent synaptic weights. The learning rules derived from the stationarity principle are self-limiting, performing a standard principal component analysis with a bias towards a negative excess Kurtosis.