Sparse Activation

A mechanism that activates only a subset of parameters or sub-networks during inference to reduce computation