ChamNet: Towards Efficient Network Design Through Platform-Aware Model Adaptation

Dai, Xiaoliang; Jia, Yangqing; Vajda, Péter; Uyttendaele, Matt; Jha, Niraj K.; Zhang, Peizhao; Wu, Bichen; Yin, Hongxu; Sun, Fei; Wang, Yanghan; Dukhan, Marat; Hu, Yunqing; Wu, Yiming

doi:10.1109/cvpr.2019.01166

Cited by 264 publications

(212 citation statements)

References 38 publications

Supporting

Mentioning

210

Contrasting

Unclassified

Order By: Relevance

“…Determining the widths of feature maps in CNNs can be considered as a subset of NAS. Although various approaches have been proposed [9,23,8,45], shrink-andexpand [16,56] is a more suitable approach for object detectors because of its simplicity and scalability. MorphNet [16] shrinks and linearly expands networks.…”

Section: Neural Architecture Search (Nas)mentioning

confidence: 99%

See 1 more Smart Citation

Understanding the Effects of Pre-Training for Object Detectors via Eigenspectrum

Shinya¹,

Simo-Serra

Suzuki

2019

2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW)

View full text Add to dashboard Cite

ImageNet pre-training has been regarded as essential for training accurate object detectors for a long time. Recently, it has been shown that object detectors trained from randomly initialized weights can be on par with those finetuned from ImageNet pre-trained models. However, the effects of pre-training and the differences caused by pretraining are still not fully understood. In this paper, we analyze the eigenspectrum dynamics of the covariance matrix of each feature map in object detectors. Based on our analysis on ResNet-50, Faster R-CNN with FPN, and Mask R-CNN, we show that object detectors trained from Ima-geNet pre-trained models and those trained from scratch behave differently from each other even if both object detectors have similar accuracy. Furthermore, we propose a method for automatically determining the widths (the numbers of channels) of object detectors based on the eigenspectrum. We train Faster R-CNN with FPN from randomly initialized weights, and show that our method can reduce ∼27% of the parameters of ResNet-50 without increasing Multiply-Accumulate operations and losing accuracy. Our results indicate that we should develop more appropriate methods for transferring knowledge from image classification to object detection (or other tasks).

show abstract

Section: Neural Architecture Search (Nas)mentioning

confidence: 99%

“…We only consider MACs and the number of parameters as metrics of model efficiency. We should consider other metrics like memory footprint [64], memory access cost [49], and real latency on target platforms [86,76,81,8].…”

Section: Limitations and Weaknessmentioning

confidence: 99%

Understanding the Effects of Pre-Training for Object Detectors via Eigenspectrum

Shinya¹,

Simo-Serra

Suzuki

2019

2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW)

View full text Add to dashboard Cite

show abstract

“…Besides, automated compact architecture design also provides a promising solution [20], [21]. Dai et al develop efficient performance predictors to speed up the search process for efficient NNs [22]. Compared to Mo-bileNetV2 on the ImageNet dataset, the generated Cham-Nets achieve up to 8.5% absolute top-1 accuracy improvement while reducing inference latency substantially.…”

Section: Efficient Neural Networkmentioning

confidence: 99%

DiabDeep: Pervasive Diabetes Diagnosis Based on Wearable Medical Sensors and Efficient Neural Networks

Yin

Mukadam

Dai

et al. 2021

IEEE Trans. Emerg. Topics Comput.

Self Cite

View full text Add to dashboard Cite

Diabetes impacts the quality of life of millions of people around the globe. However, diabetes diagnosis is still an arduous process, given that this disease develops and gets treated outside the clinic. The emergence of wearable medical sensors (WMSs) and machine learning points to a potential way forward to address this challenge. WMSs enable a continuous, yet user-transparent, mechanism to collect and analyze physiological signals. However, disease diagnosis based on WMS data and its effective deployment on resource-constrained edge devices remain challenging due to inefficient feature extraction and vast computation cost. To address these problems, we propose a framework called DiabDeep that combines efficient neural networks (called DiabNNs) with off-the-shelf WMSs for pervasive diabetes diagnosis. DiabDeep bypasses the feature extraction stage and acts directly on WMS data. It enables both an (i) accurate inference on the server, e.g., a desktop, and (ii) efficient inference on an edge device, e.g., a smartphone, to obtain a balance between accuracy and efficiency based on varying resource budgets and design goals. On the resource-rich server, we stack sparsely connected layers to deliver high accuracy. On the resource-scarce edge device, we use a hidden-layer long short-term memory based recurrent layer to substantially cut down on computation and storage costs while incurring only a minor accuracy loss. At the core of our system lies a grow-and-prune training flow: it leverages gradient-based growth and magnitude-based pruning algorithms to enable DiabNNs to learn both weights and connections, while improving accuracy and efficiency. We demonstrate the effectiveness of DiabDeep through a detailed analysis of data collected from 52 participants. For server (edge) side inference, we achieve a 96.3% (95.3%) accuracy in classifying diabetics against healthy individuals, and a 95.7% (94.6%) accuracy in distinguishing among type-1 diabetic, type-2 diabetic, and healthy individuals. Against conventional baselines, such as support vector machines with linear and radial basis function kernels, k-nearest neighbor, random forest, and linear ridge classifiers, DiabNNs achieve higher accuracy, while reducing the model size (floating-point operations) by up to 454.5× (8.9×). Therefore, the system can be viewed as pervasive and efficient, yet very accurate.

show abstract

“…Similarly, we profile the models on the same Pixel 1 device. For prior work that does not optimize for Pixel 1, we retrain and profile their model closest to the MnasNet baseline (e.g., the FBNet-B and ChamNet-B networks [15], [16], since the authors use these ConvNets to compare against the MnasNet model). Finally, we directly report the number of epochs reported per method, hence canceling out the effect of different hardware systems (GPU vs. TPU hours).…”

Section: State-of-the-art Runtime-constrained Imagenet Classificationmentioning

confidence: 99%

Single-Path Mobile AutoML: Efficient ConvNet Design and NAS Hyperparameter Optimization

Stamoulis

Ding

Wang

et al. 2020

IEEE J. Sel. Top. Signal Process.

View full text Add to dashboard Cite

Can we reduce the search cost of Neural Architecture Search (NAS) from days down to only few hours? NAS methods automate the design of Convolutional Networks (ConvNets) under hardware constraints and they have emerged as key components of AutoML frameworks. However, the NAS problem remains challenging due to the combinatorially large design space and the significant search time (at least 200 GPUhours). In this work, we alleviate the NAS search cost down to less than 3 hours, while achieving state-of-the-art image classification results under mobile latency constraints. We propose a novel differentiable NAS formulation, namely Single-Path NAS, that uses one single-path over-parameterized ConvNet to encode all architectural decisions based on shared convolutional kernel parameters, hence drastically decreasing the search overhead. Single-Path NAS achieves state-of-the-art top-1 ImageNet accuracy (75.62%), hence outperforming existing mobile NAS methods in similar latency settings (∼ 80ms). In particular, we enhance the accuracy-runtime trade-off in differentiable NAS by treating the Squeeze-and-Excitation path as a fully searchable operation with our novel single-path encoding. Our method has an overall cost of only 8 epochs (24 TPU-hours), which is up to 5,000× faster compared to prior work. Moreover, we study how different NAS formulation choices affect the performance of the designed ConvNets. Furthermore, we exploit the efficiency of our method to answer an interesting question: instead of empirically tuning the hyperparameters of the NAS solver (as in prior work), can we automatically find the hyperparameter values that yield the desired accuracy-runtime trade-off (e.g., target runtime for different platforms)? We view our extensive experimental results as a valuable exploration for NAS-based cloud AutoML services, and we open-source our entire codebase at: https://github.com/dstamoulis/single-path-nas.

show abstract

ChamNet: Towards Efficient Network Design Through Platform-Aware Model Adaptation

Cited by 264 publications

References 38 publications

Understanding the Effects of Pre-Training for Object Detectors via Eigenspectrum

Understanding the Effects of Pre-Training for Object Detectors via Eigenspectrum

DiabDeep: Pervasive Diabetes Diagnosis Based on Wearable Medical Sensors and Efficient Neural Networks

Single-Path Mobile AutoML: Efficient ConvNet Design and NAS Hyperparameter Optimization

Contact Info

Product

Resources

About