NetAdapt: Platform-Aware Neural Network Adaptation for Mobile Applications
This work addresses the need for efficient neural network deployment on mobile devices, offering an incremental improvement by automating adaptation with direct metric optimization.
The paper tackles the problem of adapting pre-trained deep neural networks to mobile platforms under resource constraints by proposing NetAdapt, which directly optimizes for latency and energy consumption using empirical measurements. It achieves up to 1.7x speedup in inference latency with equal or higher accuracy on MobileNets for ImageNet classification compared to state-of-the-art methods.
This work proposes an algorithm, called NetAdapt, that automatically adapts a pre-trained deep neural network to a mobile platform given a resource budget. While many existing algorithms simplify networks based on the number of MACs or weights, optimizing those indirect metrics may not necessarily reduce the direct metrics, such as latency and energy consumption. To solve this problem, NetAdapt incorporates direct metrics into its adaptation algorithm. These direct metrics are evaluated using empirical measurements, so that detailed knowledge of the platform and toolchain is not required. NetAdapt automatically and progressively simplifies a pre-trained network until the resource budget is met while maximizing the accuracy. Experiment results show that NetAdapt achieves better accuracy versus latency trade-offs on both mobile CPU and mobile GPU, compared with the state-of-the-art automated network simplification algorithms. For image classification on the ImageNet dataset, NetAdapt achieves up to a 1.7$\times$ speedup in measured inference latency with equal or higher accuracy on MobileNets (V1&V2).