Figure 4.

A modified Euclidean distance, where each dimension is weighted, allows accurate discrimination with similar -or better- performance than SOM neural nets. In the “WED” (Weighted Euclidean Distance) analysis, each dimension in Euclidean space is weighted based on the Kullback–Leibler divergence of the response distribution in that dimension. Each dimension (neuron/time bin) can be weighted independently (‘independent W’), or a single weight can be set for a given neuron across time bins (‘fixed W’). Although using independent weights maximizes the information extracted about the difference in stimuli, using a fixed weight emulates a biologically more realistic decoding network. The best method varies across systems: (A) Electrosensory; (B) Olfactory; (C) LIF model. Curves show averages (± s.d.) across all pairs of stimuli.