Abstract: End-edge collaborative inference improves computational efficiency by dividing a deep neural network into two parts, executed across the end device and the edge node in parallel. However, ...