Tailored ensembles of neural networks optimize sensitivity to stimulus statistics
Johannes Zierenberg, Jens Wilting, Viola Priesemann, Anna Levina · Physical Review Research · 2020
The authors show that a properly tuned ensemble of recurrent networks can discriminate stimuli of arbitrary intensity distributions. This offers a solution for a long-standing dynamic range problem. The results are relevant both for the understanding of the sensitivity of biological networks and for the optimization of artificial networks, for example, in a reservoir computing setting.