A Linear NAS Service of ConvNets for Fast Deployment in the Edge of 5G Networks
A Linear NAS Service of ConvNets for Fast Deployment in the Edge of 5G Networks
复制标题
用于在 5G 网络边缘快速部署的 ConvNet 线性 NAS 服务
DOI：
10.1109/mnet.011.1900336
复制
发表时间：
2021-03
期刊：
影响因子：
9.3
通讯作者：
Meikang Qiu
中科院分区：
文献类型：
--
作者：
Jihe Wang;Jiaxiang Zhao;Jianfeng An;Danghui Wang;Meikang Qiu
5G网络为在边缘设备上部署神经网络模型带来了极大的便利。然而，这种灵活性挑战了当前一代神经网络架构在结构搜索过程中很少将目标平台作为边界的问题。部署困难的根源在于，在SW/HW联合调优过程中，目标设备的隐藏资源提供导致网络模型与系统配置不匹配。本文提出了一种5G环境下可扩展的神经网络搜索服务，支持网络尺度的连续旋钮，使信道组可以相互重叠，共享连续覆盖的特征。实验证明，所提出的网络扩展维度在特定平台上提供了良好的模型性能可预测性，可进一步用于简化常规的网络架构搜索过程。然后，我们设计了一个涉及云和边缘的软件/硬件协同设计工作流，以充分利用目标平台上的计算资源，同时保持网络规模尽可能小，以节省资源供应。实验结果表明，使用我们的可扩展搜索服务，网络模型的关键指标对所提出的超参数具有连续、单调和线性的函数。在树莓派板上的部署表明，所提出的方法准确地控制了模型的精度和尺寸；同时，相应的搜索工作流成功地找到了合适的网络规模，减少了65%的NAS例程。
The 5G network brings about significant convenience in deploying neural network models to edge devices. However, the flexibility challenges the current generation of neural network architectures that seldom involve the target platform as the bounds during the structure searching. The difficult deployment is rooted in the hidden resource provision of the target devices during the SW/HW joint tuning, leading to mismatching between network models and system configuration. This work proposes a scalable neural network search service in the 5G environment to support a continuous knob of the network scales, by which the channel groups can overlap with each other to share the features with continuous coverage. It is proved that the proposed dimension of network scaling provides good predictability of the model performance on specific platforms, which can be further utilized to simplify the regular network architectural search procedures. Then we design a SW/HW co-design workflow that involves both the cloud and edge to fully utilize computing resources on target platforms, meanwhile keeping the network size as small as possible to save the provision of resources. The experimental results show that, with our scalable search service, the key metrics of the network model enjoy a continuous, monotonic, and linear function to the proposed hyper-parameter. The deployment to the Raspberry Pi board shows that the proposed method accurately controls both precision and size of the models; meanwhile, the corresponding search workflow successfully finds the proper network scales with 65 percent reduction of the NAS routines.
登录
查看更多内容
DOI：
10.1109/iccv48922.2021.01202
发表时间：
2019-07
期刊：
2021 IEEE/CVF International Conference on Computer Vision (ICCV)
影响因子：
--
作者：
Xiangxiang Chu;Bo Zhang;Ruijun Xu;Jixiang Li
通讯作者：
Xiangxiang Chu;Bo Zhang;Ruijun Xu;Jixiang Li
DOI：
--
发表时间：
2018-02
期刊：
ArXiv
影响因子：
--
作者：
Hieu Pham;M. Guan;Barret Zoph;Quoc V. Le;J. Dean
通讯作者：
Hieu Pham;M. Guan;Barret Zoph;Quoc V. Le;J. Dean
DOI：
--
发表时间：
2015-10
期刊：
arXiv: Computer Vision and Pattern Recognition
影响因子：
--
作者：
Song Han;Huizi Mao;W. Dally
通讯作者：
Song Han;Huizi Mao;W. Dally
影响因子：
3.9
作者：
Keke Gai;Meikang Qiu;Hui Zhao;Xiaotong Sun
通讯作者：
Keke Gai;Meikang Qiu;Hui Zhao;Xiaotong Sun
DOI：
--
发表时间：
2018-06
期刊：
ArXiv
影响因子：
--
作者：
Hanxiao Liu;K. Simonyan;Yiming Yang
通讯作者：
Hanxiao Liu;K. Simonyan;Yiming Yang