MCDNN: An Approximation-Based Execution Framework for Deep Stream Processing Under Resource Constraints

By Seungyeop Han, Haichen Shen, Matthai Philipose, Sharad K. Agarwal, Alec Wolman, Arvind Krishnamurthy. ACM SIGMOBILE International Conference on Mobile Systems, Applications, and Services, 2016

413 citations recorded by the source corpus.

Subfields: Computer Networks and Communications, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/a4ddcd04221d0713e3dc4b4515fdd5f0f8871112

How researchers see this paper

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

In Computer Networks and Communications

1. Xu Chen
2. Liekang Zeng
3. Michael J. Freedman
4. Phillip B. Gibbons
5. Zhi Zhou
6. Ganesh Ananthanarayanan
7. Onur Mutlu
8. Junshan Zhang

In Computer Vision and Pattern Recognition

1. Xu Chen
2. Jiasi Chen
3. Xukan Ran
4. Mi Zhang
5. Biyi Fang
6. Xiao Zeng
7. Liekang Zeng
8. Zhi Zhou

Citations in the corpus

• NestDNN: Resource-Aware Multi-Tenant On-Device Deep Learning for Continuous Mobile Vision (2018, methodology): “To enable on-device deep learning, one of the common techniques used by application developers is compressing the deep learning model to reduce its resource demand at a modest loss in accuracy as trade-off [12, 39].”
• CoEdge: Cooperative DNN Inference With Adaptive Workload Partitioning Over Heterogeneous Edge Devices (2020, background): “MCDNN [41] fully offloads DNN computation.”
• Real-Time Video Analytics: The Killer App for Edge Computing (2017, methodology): “Execution of DNNs on GPUs is made dramatically more efficient by running a DNN execution service on each machine that handles all DNN requests on the GPU (described below in “Efficient DNNs execution on GPUs”).”
• DeepDecision: A Mobile Deep Learning Framework for Edge Video Analytics (2018, methodology): “The strawman is based on [8] and uses a fixed resolution of 320×240 and a video bitrate of 500 kbps, and picks the model (local or remote) with the best accuracy.”
• Deep Learning With Edge Computing: A Review (2019, methodology): “1) Offloading: Recent approaches, such as DeepDecision [20], [111] and MCDNN [104], take an optimization-based offloading approach with constraints such as network latency and bandwidth, device energy, and monetary cost.”
• Live Video Analytics at Scale with Approximation and Delay-Tolerance (2017, background): “As DNN models get compressed [45,46], reducing their resource demand at the cost of quality, the compression factor presents another knob.”
• DeepMon: Mobile GPU-based Deep Learning Framework for Continuous Vision Applications (2017, methodology): “One way to overcome this limitation is to use cloud resources to run the required DNNs [25].”
• Chameleon: scalable adaptation of video analytics (2018, background): “While prior video analytics systems [16, 32, 33] profile the processing pipeline to minimize cost, they only do so once, at the beginning of the video.”
• Efficient Incremental Learning for Mobile Object Detection (2019, background): “Researchers have focused on reducing the computation cost on the mobile devices by compressing pre-trained deep learning models [10, 9, 20, 3], designing lightweight neural network architectures [16, 13], or hardware accelerations [15, 37] — these papers have not addressed the incremental learning…”
• Focus: Querying Large Video Datasets with Low Latency and Low Cost (2018, methodology): “The second technique is CNN specialization [43], where the CNNs are trained on a subset of a dataset specific to a particular context (such as a video stream).”