Multiresolution Representations and Data Mining of Neural Spikes for Brain-Machine Interfaces

In brain-machine interface (BMI) applications, neural firing activities have been represented by spike counts with a fixed-width time bin. Adaptive models have been designed to utilize these bin counts for mapping the associated behavior which is typically 2D or 3D hand movement. However, the representation of the firing activities can be enriched by binning neural spikes with multiple time scales based on multiresolution analysis. This multiresolution representation of neural activities can provide more accurate prediction of the hand movement parameters. Data mining techniques must be applied to models using multiresolution representation in order to avoid over-fitting. In this paper, we demonstrate that the multiresolution representation improves the performance of the linear model for BMIs compared to the model with the fixed-width time bin