1 paper
Kohga Tanaka, Hiroaki Nishi
Deploying deep neural networks on memory-constrained edge accelerators is bottlenecked by per-inference off-chip weight transfer rather than computation: the dense network cannot b…