Fig. 12.
NeuralTree classifier: (a) probabilistic NeuralTree trained with energy-aware regularization and network pruning, and (b) the NeuralTree hardware implementation and system operation under the proposed singlepath channel-selective inference scheme. The NeuralTree is trained on neural activity from all 256 channels. Following the regularization and network pruning, each internal node contains an optimal set of up to 64 features, which can be associated with any input channels.