1 paper · 1 filter
Pouya Shaeri, Ariane Middel
Modern neural networks often activate all neurons for every input, leading to unnecessary computation and inefficiency. We introduce Matrix-Interpolated Dropout Layer (MID-L), a no…