Biomedical Engineering Reference
In-Depth Information
behavior as would be expected by chance) (
Fig. 7.8A
). To per-
form statistical pattern recognition, one should:
1. Construct perievent matrices for each neuron.
2. Prepocess the perievent matrices (smoothing and decimation).
3. Reduce the dimensions of the data (via PCA or wavelet-based
methods). This step is important for optimizing classifier per-
formance; irrelevant dimensions will both slow and decrease
classifier performance.
Fig. 7.8. Predictive interactions. (A) Statistical pattern recognition uses trial-by-trial spike train to generate predictions of
behavior on a trial-by-trial basis. These predictions can be compared with predictions from another neuron. (B) Peristim-
ulus rasters of neurons and their predictions of reaction time; black corresponds to trials with short RTs, red corresponds
to trials with slow RTs. In between the peristimulus rasters, predictions are plotted: black means that a fast RT was pre-
dicted from spike trains; red means that a slow RT was predicted. Central column: Green indicates that the two neurons
agreed in their predictive information; white indicates that the neurons disagreed. These neurons tended to agree more
than could be expected by chance, and thus had an interaction in their predictive information. (C) Predictive interactions
across the ensemble.
