Neuron-Level Architecture Growth: A Controlled Evaluation for EEG Time-Series Decoding
The results suggest that growth helps when its criterion can rank the candidate neurons, and that the rate of skipped neuron addition tells where a decoder can be grown small from scratch.