NetAdapt: Platform-Aware Neural Network Adaptation for Mobile Applications

By Tien-Ju Yang, Andrew G. Howard, Xiao Zhang, Alec Go, Vivienne Sze, Hartwig Adam. 2018

575 citations recorded by the source corpus.

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).

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Radiology, Nuclear Medicine and Imaging, Efficiency & Compression

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Radiology, Nuclear Medicine and Imaging

In Efficiency & Compression

Citations in the corpus

EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks (2019, background): “Model compression (Han et al., 2016; He et al., 2018; Yang et al., 2018) is a common way to reduce model size by trading accuracy for efficiency.”
MnasNet: Platform-Aware Neural Architecture Search for Mobile (2018, methodology): “Some commonly-used approaches include 1) quantizing the weights and/or activations of a baseline CNN model into lower-bit representations [8, 16], or 2) pruning less important filters according to FLOPs [6, 10], or to platform-aware metrics such as latency introduced in [32].”
Group Knowledge Transfer: Collaborative Training of Large CNNs on the Edge (2020, methodology): “Our work also relates to efficient deep learning on edge devices, such as model compression [43, 44, 45], manually designed architectures (MobileNets [46], ShuffeNets [47], SqueezeNets [48]), or even efficient neural architecture search (EfficientNets [4], FBNet [49]).”
FBNetV3: Joint Architecture-Recipe Search using Neural Acquisition Function (2020, background): “However, pruning either considers only one architecture [10] or can only sequentially search smaller and smaller architectures [45].”
FBNet: Hardware-Aware Efficient ConvNet Design via Differentiable Neural Architecture Search (2018, background): “Besides searching for new architectures, works such as [27, 5] focus on adapting existing models to improve efficiency.”
Layer-compensated Pruning for Resource-constrained Convolutional Neural Networks (2018, methodology): “…Layer Scheduling Prior art (Luo et al., 2017; Li et al., 2017; Yoon et al., 2018; He et al., 2018a;b; 2017; Jiang et al., 2018; Lin et al., 2018; Yang et al., 2018b) adopts the aforementioned assumption and approaches equation (6) by introducing the layer scheduling variable into the picture:…”
AMC: AutoML for Model Compression and Acceleration on Mobile Devices (2018, methodology): “We also compare our learning based policy with a heuristic-based policy [52], and AMC better trades off accuracy and latency.”
Piggyback GAN: Efficient Lifelong Learning for Image Conditioned Generation (2021, background): “In recent years, there has been increasing interest in better aligning the resource consumption of neural networks with the computational constraints of the platforms on which they will be deployed [7,36,38].”
Single Path One-Shot Neural Architecture Search with Uniform Sampling (2019, background): “Previous works usually deal with the problem by channel pruning [8, 9] or model adaption [30, 6].”
Searching for MobileNetV3 (2019, methodology): “In [48] the metric was to minimize the accuracy change.”
ChamNet: Towards Efficient Network Design Through Platform-Aware Model Adaptation (2018, methodology): “For example, NetAdapt [33] utilizes a hardware-aware filter pruning algorithm and achieves up to 1.”
A Survey of Quantization Methods for Efficient Neural Network Inference (2021, neutral): “It is important to consider this fact for achieving optimal beneﬁts through hardware-aware quantization [86, 90, 238, 242, 246, 248, 257, 259].”
Group Knowledge Transfer: Federated Learning of Large CNNs at the Edge (2020, methodology): “Our work also relates to efficient deep learning on edge devices, such as model compression [47, 48, 49], manually designed architectures (MobileNets [50], ShuffeNets [51], SqueezeNets [52]), or even efficient neural architecture search (EfficientNets [4], FBNet [53]).”
InstaNAS: Instance-aware Neural Architecture Search (2018, background): “InstaNAS is critical in many of the recently proposed settings such as multi-objective NAS [5, 9, 26, 34], which optimizes not only task-dependent metrics such as accuracy but also those metrics that are architecture-dependent such as latency.”
AOWS: Adaptive and Optimal Network Width Search With Latency Constraints (2020, background): “M ( N ) can represent latency, power consumption constraints, or combinations of these objectives [29, 30].”
Energy-Constrained Compression for Deep Neural Networks via Weighted Sparse Projection and Layer Input Masking (2018, methodology): “For SSL and NetAdapt, we apply 20 additional epochs to achieve comparable results.”
Lite Transformer with Long-Short Range Attention (2020, methodology): “Recently, AutoML has also been used to automate the model compression and acceleration (He et al., 2018; Yang et al., 2018; Wang et al., 2019; Liu et al., 2019a).”