pith. sign in

arxiv: 2604.09681 · v1 · submitted 2026-04-03 · 💻 cs.NI · cs.CV· cs.DC

R2E-VID: Two-Stage Robust Routing via Temporal Gating for Elastic Edge-Cloud Video Inference

Zheming Yang , Lulu Zuo , Shun Lu , Yangyu Zhang , Zhicheng Li , Xiangyang Li , Yang You This is my paper

Pith reviewed 2026-05-13 18:25 UTC · model grok-4.3

classification 💻 cs.NI cs.CVcs.DC
keywords edge-cloud video inferencetemporal gatingrobust routingelastic resource allocationvideo analyticsmotion dynamicsdelay minimization
0
0 comments X

The pith

R2E-VID routes video inference tasks between edge and cloud nodes using temporal gating to cut costs by up to 60 percent.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces R2E-VID as a two-stage framework for adaptive routing of video analytics workloads in edge-cloud systems. The first stage applies temporal gating to evaluate motion dynamics and consistency in incoming video segments, deciding how to split processing between edge and cloud resources. The second stage performs robust optimization to refine those decisions under changing network and workload conditions. If the approach holds, it would allow video inference to maintain performance while using far fewer total resources than fixed cloud or static edge-cloud methods.

Core claim

R2E-VID establishes a two-stage robust routing framework via temporal gating for elastic edge-cloud video inference. The temporal gating stage models temporal consistency and motion dynamics of video streams to predict optimal routing patterns for each segment. The subsequent robust routing optimization module refines allocations through multi-model adaptation to jointly minimize inference delay and resource consumption under dynamic variations.

What carries the argument

Temporal gating mechanism that models temporal consistency and motion dynamics to predict optimal routing patterns for each video segment.

If this is right

  • Adaptive partitioning of inference workloads achieves fine-grained spatiotemporal elasticity between edge and cloud.
  • Robust optimization jointly minimizes inference delay and resource consumption under dynamic network and workload variations.
  • Overall cost reductions reach up to 60 percent compared to cloud-centric baselines.
  • Delay drops 35-45 percent and accuracy rises 2-7 percent relative to prior edge-cloud solutions.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same gating logic could be tested on streaming sensor data or audio feeds that share temporal structure.
  • Real deployments would need to measure how often gating predictions hold when bandwidth or compute availability shifts rapidly.
  • Future extensions might add forward prediction of upcoming segments to make routing decisions even more proactive.

Load-bearing premise

Temporal gating can reliably predict the optimal routing pattern for each video segment from motion dynamics and temporal consistency without adding significant overhead or error under real fluctuating conditions.

What would settle it

A test showing that temporal gating mispredicts routing decisions for a large fraction of segments under real fluctuating network conditions or high motion variability would show the core mechanism fails to deliver the claimed gains.

Figures

Figures reproduced from arXiv: 2604.09681 by Lulu Zuo, Shun Lu, Xiangyang Li, Yang You, Yangyu Zhang, Zheming Yang, Zhicheng Li.

Figure 1
Figure 1. Figure 1: The illustration of edge-cloud collaborative archi￾tecture for video inference. 48]. Conversely, smaller models offer reduced inference de￾lay and energy consumption but at the cost of lower accu￾racy [8, 32]. To mitigate costs, deploying models of varying sizes across servers can cater to diverse inference tasks ef￾fectively [42]. In practical scenarios, task requirements fre￾quently fluctuate with change… view at source ↗
Figure 3
Figure 3. Figure 3: The workflow of the proposed R2E-VID frame￾work. assignment (edge or cloud). In the second stage, the frame￾work performs multi-model elastic inference, dynamically selecting the most appropriate model version based on the initial configuration and real-time resource conditions. This ensures that the inference process remains both cost-efficient and accurate under varying workloads. 3.1 Two-Stage Robust Op… view at source ↗
Figure 4
Figure 4. Figure 4: The illustration of adaptive edge-cloud collabora￾tive configuration via temporal gating. the optimal solution of subproblem 1 is ( 𝑢 ∗ i , 𝜋∗ i ) for a given 𝑦 ∗ 𝑖 . According to the duality theorem, the following cutting planes can be constructed: 𝜂 ≥ ( h − Q𝑦 − Lu∗ 𝑖 )T 𝜋 ∗ 𝑖 . The 𝜂 = max𝑢∈U min𝑣∈𝐹 (𝑦,𝑢) b 𝑇 𝑣 is a one-dimensional scalar, and then the cut-plane constraint is added to the first stage of… view at source ↗
Figure 5
Figure 5. Figure 5: The accuracy-cost tradeoff analysis under different datasets [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: The comparative results of different methods un￾der the COCO dataset. 2 4 6 8 10 5 10 15 20 25 Delay (s) Number of tasks A2 JCAB RDAP Sniper R2E-VID ×10 (a) Delay 2 4 6 8 10 80 160 240 320 400 Energy Consumption (J) Number of tasks A2 JCAB RDAP Sniper R2E-VID ×10 (b) Energy Consumption [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The comparative results of different methods un￾der the UA-DETRAC dataset. 2 4 6 8 10 5 10 15 20 25 30 Delay (s) Number of tasks A2 JCAB RDAP Sniper R2E-VID ×10 (a) Delay 2 4 6 8 10 80 160 240 320 400 Energy Consumption (J) Number of tasks A2 JCAB RDAP Sniper R2E-VID ×10 (b) Energy Consumption [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: The comparative results of different methods un￾der the ADE20K dataset. 4.3.2 Robustness to Changes in the Number of Tasks. To assess overall performance, we evaluate all methods un￾der varying task volumes, with results averaged over sta￾ble and fluctuating accuracy requirements. As shown in Fig￾ure 6, [PITH_FULL_IMAGE:figures/full_fig_p007_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: The average cost comparison of different methods under dynamic bandwidths. 4.3.3 Robustness to Dynamic Network. To further as￾sess robustness under real-world conditions, we evaluate all methods under dynamic network environments, where the bandwidth fluctuates within 0, 10%, 20%, 30%. All other set￾tings remain unchanged. The results in [PITH_FULL_IMAGE:figures/full_fig_p008_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: The ablation studies of two-stage robust opti￾mization. yet tightly coordinated stages. The first stage employs a tem￾poral gating–based adaptive configuration module that cap￾tures motion dynamics and temporal consistency in the video stream, allowing the system to adjust resolution, frame rate, and edge–cloud partitioning in real time. The second stage further refines the decision space through robust m… view at source ↗
read the original abstract

With the rapid growth of large-scale video analytics applications, edge-cloud collaborative systems have become the dominant paradigm for real-time inference. However, existing approaches often fail to dynamically adapt to heterogeneous video content and fluctuating resource conditions, resulting in suboptimal routing efficiency and high computational costs. In this paper, we propose R2E-VID, a two-stage robust routing framework via temporal gating for elastic edge-cloud video inference. In the first stage, R2E-VID introduces a temporal gating mechanism that models the temporal consistency and motion dynamics of incoming video streams to predict the optimal routing pattern for each segment. This enables adaptive partitioning of inference workloads between edge and cloud nodes, achieving fine-grained spatiotemporal elasticity. In the second stage, a robust routing optimization module refines the allocation through multi-model adaptation, jointly minimizing inference delay and resource consumption under dynamic network and workload variations. Extensive experiments on public datasets demonstrate that R2E-VID achieves up to 60% reduction in overall cost compared to cloud-centric baselines, and delivers 35-45% lower delay while improving inference accuracy by 2-7% over state-of-the-art edge-cloud solutions.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes R2E-VID, a two-stage framework for robust routing in elastic edge-cloud video inference. Stage 1 uses a temporal gating mechanism to predict per-segment routing patterns from motion dynamics and temporal consistency, enabling adaptive edge-cloud workload partitioning. Stage 2 applies a robust multi-model optimizer to jointly minimize inference delay and resource cost under network and workload variations. Experiments on public datasets are reported to yield up to 60% cost reduction versus cloud-centric baselines, 35-45% lower delay, and 2-7% higher accuracy versus prior edge-cloud solutions.

Significance. If the performance numbers are reproducible, the work would offer a practical advance in adaptive video analytics systems by combining lightweight temporal prediction with robust optimization, potentially improving cost and latency in heterogeneous edge-cloud deployments.

major comments (2)
  1. [§4] §4 (Experiments): the headline claims of 60% cost reduction and 35-45% delay improvement rest on the temporal gating stage producing near-optimal initial partitions; however, the section provides no quantitative metrics (e.g., gating prediction error rate, false-positive routing fraction, or sensitivity to workload fluctuation) that would allow verification that mispredictions remain below the threshold at which the second-stage optimizer can still recover the reported gains.
  2. [§3.2] §3.2 (Temporal Gating Mechanism): the description of how motion dynamics and temporal consistency are encoded into routing decisions lacks any formal bound or empirical characterization of decision overhead and error under the fluctuating network conditions stated as the target regime; without this, the claim that the two-stage design achieves fine-grained spatiotemporal elasticity cannot be assessed.
minor comments (2)
  1. [Abstract and §4] The abstract and §4 refer to 'public datasets' and 'state-of-the-art edge-cloud solutions' without naming the specific datasets, video resolutions, or exact baseline implementations, which hinders reproducibility.
  2. [§3] Notation for the gating function and the robust optimizer objective is introduced without a consolidated table of symbols, making cross-references between §3.1 and §3.2 harder to follow.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments and the recommendation for major revision. We address each major comment below, agreeing to enhance the manuscript with additional quantitative analysis as requested.

read point-by-point responses

  1. Referee: [§4] §4 (Experiments): the headline claims of 60% cost reduction and 35-45% delay improvement rest on the temporal gating stage producing near-optimal initial partitions; however, the section provides no quantitative metrics (e.g., gating prediction error rate, false-positive routing fraction, or sensitivity to workload fluctuation) that would allow verification that mispredictions remain below the threshold at which the second-stage optimizer can still recover the reported gains.

    Authors: We agree that providing quantitative metrics on the temporal gating stage would strengthen the verification of our performance claims. In the revised version, we will add to §4 the gating prediction error rate, false-positive routing fraction, and sensitivity analysis to workload fluctuations. This will demonstrate that the misprediction levels allow the second-stage optimizer to recover the reported gains in cost, delay, and accuracy. revision: yes

  2. Referee: [§3.2] §3.2 (Temporal Gating Mechanism): the description of how motion dynamics and temporal consistency are encoded into routing decisions lacks any formal bound or empirical characterization of decision overhead and error under the fluctuating network conditions stated as the target regime; without this, the claim that the two-stage design achieves fine-grained spatiotemporal elasticity cannot be assessed.

    Authors: We acknowledge the need for a more rigorous characterization of the temporal gating mechanism. In the revision, we will expand §3.2 to include empirical measurements of decision overhead and error rates under fluctuating network conditions, as well as any formal bounds that can be derived from the model's design. This will better support the claim of fine-grained spatiotemporal elasticity. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper describes a two-stage framework (temporal gating for motion-based routing prediction followed by robust optimization) but supplies no equations, fitted parameters, self-citations, or derivations in the abstract or visible text. Performance numbers are presented as experimental outcomes on public datasets rather than reductions to inputs by construction. No self-definitional, fitted-input-as-prediction, or uniqueness-via-self-citation patterns are detectable, so the central claims remain independent of the described mechanisms.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no free parameters, axioms, or invented entities are specified in the text. Full manuscript details on any modeling assumptions are unavailable.

pith-pipeline@v0.9.0 · 5528 in / 1160 out tokens · 39965 ms · 2026-05-13T18:25:05.269218+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

50 extracted references · 50 canonical work pages

  1. [1]

    Tong Bai, Haoran Zhao, Lei Huang, Zhipeng Wang, Dong In Kim, and Arumugam Nallanathan. 2026. A Decade of Video Analytics at Edge: Training, Deployment, Orchestration, and Platforms. IEEE Communi- cations Surveys & Tutorials 28 (2026), 2127–2162

    work page 2026

  2. [2]

    Dimitris Bertsimas, Eugene Litvinov, Xu Andy Sun, Jinye Zhao, and Tongxin Zheng. 2012. Adaptive robust optimization for the security constrained unit commitment problem. IEEE Transactions on Power Systems 28, 1 (2012), 52–63

    work page 2012

  3. [3]

    Bedrettin Cetinkaya, Sinan Kalkan, and Emre Akbas. 2024. Ranked: Addressing imbalance and uncertainty in edge detection using ranking-based losses. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . 3239–3249

    work page 2024

  4. [4]

    Jiasi Chen and Xukan Ran. 2019. Deep learning with edge computing: A review. Proc. IEEE 107, 8 (2019), 1655–1674

    work page 2019

  5. [5]

    Marc Goerigk, Stefan Lendl, and Lasse Wulf. 2022. Two-stage robust optimization problems with two-stage uncertainty. European Journal of Operational Research 302, 1 (2022), 62–78

    work page 2022

  6. [6]

    Kevin Hsieh, Ganesh Ananthanarayanan, Peter Bodik, Shivaram Venkataraman, Paramvir Bahl, Matthai Philipose, Phillip B Gibbons, and Onur Mutlu. 2018. Focus: Querying large video datasets with low latency and low cost. In 13th USENIX Symposium on Operating Systems Design and Implementation (OSDI) . 269–286

    work page 2018

  7. [7]

    Yunqing Hu, Zheming Yang, Chang Zhao, Qi Guo, Meng Gao, Pengcheng Li, and Wen Ji. 2026. AIVD: Adaptive Edge-Cloud Col- laboration for Accurate and Efficient Industrial Visual Detection. arXiv:2601.04734 [cs.CV] https://arxiv.org/abs/2601.04734

    work page arXiv 2026

  8. [8]

    Yunqing Hu, Zheming Yang, Chang Zhao, and Wen Ji. 2025. Adap- tive Guidance Semantically Enhanced via Multimodal LLM for Edge- Cloud Object Detection. arXiv: 2509.19875 [cs.CV] https://arxiv.org/ abs/2509.19875

    work page arXiv 2025

  9. [9]

    Wen Ji, Bing Liang, Yuqin Wang, Rui Qiu, and Zheming Yang. 2020. Crowd V-IoE: Visual internet of everything architecture in AI-driven fog computing. IEEE Wireless Communications 27, 2 (2020), 51–57

    work page 2020

  10. [10]

    Junchen Jiang, Ganesh Ananthanarayanan, Peter Bodik, Siddhartha Sen, and Ion Stoica. 2018. Chameleon: Scalable adaptation of video analytics. In ACM Special Interest Group on Data Communication (SIG- COMM). 253–266

    work page 2018

  11. [11]

    Jingyan Jiang, Ziyue Luo, Chenghao Hu, Zhaoliang He, Zhi Wang, Shutao Xia, and Chuan Wu. 2021. Joint model and data adaptation for cloud inference serving. In 2021 IEEE Real-Time Systems Symposium (RTSS). IEEE, 279–289

    work page 2021

  12. [12]

    Yiping Kang, Johann Hauswald, Cao Gao, Austin Rovinski, Trevor Mudge, Jason Mars, and Lingjia Tang. 2017. Neurosurgeon: Collabora- tive intelligence between the cloud and mobile edge. ACM SIGARCH Computer Architecture News 45, 1 (2017), 615–629

    work page 2017

  13. [13]

    Seah Kim, Hasan Genc, Vadim Vadimovich Nikiforov, Krste Asanović, Borivoje Nikolić, and Yakun Sophia Shao. 2023. MoCA: Memory- centric, adaptive execution for multi-tenant deep neural networks. In 2023 IEEE International Symposium on High-Performance Computer Ar- chitecture (HPCA). 828–841

    work page 2023

  14. [14]

    Pavel Koupil, Sebastián Hricko, and Irena Holubová. 2022. MM-infer: A tool for inference of multi-model schemas. In EDBT, Vol. 22. 1–4

    work page 2022

  15. [15]

    Duan Li and XL Sun. 2006. Towards strong duality in integer pro- gramming. Journal of Global Optimization 35, 2 (2006), 255–282

    work page 2006

  16. [16]

    En Li, Liekang Zeng, Zhi Zhou, and Xu Chen. 2019. Edge AI: On- demand accelerating deep neural network inference via edge com- puting. IEEE Transactions on Wireless Communications 19, 1 (2019), 447–457

    work page 2019

  17. [17]

    Guo Li, Jiandian Zeng, Zihao Peng, Yuzhu Liang, Xi Zheng, and Tian Wang. 2025. E2EC: Edge-to-Edge Collaboration for Efficient Real- Time Video Surveillance Inference. IEEE Transactions on Mobile Com- puting 24, 9 (2025), 9126–9140

    work page 2025

  18. [18]

    Jingzong Li, Yik Hong Cai, Libin Liu, Yu Mao, Chun Jason Xue, and Hong Xu. 2023. Moby: Empowering 2D models for efficient point cloud analytics on the edge. In Proceedings of the 31st ACM Interna- tional Conference on Multimedia (MM) . 9012–9021

    work page 2023

  19. [19]

    Min Li, Yu Li, Ye Tian, Li Jiang, and Qiang Xu. 2021. AppealNet: An ef- ficient and highly-accurate edge/cloud collaborative architecture for DNN inference. In ACM/IEEE Design Automation Conference (DAC) . 409–414

    work page 2021

  20. [20]

    Rui Li, Zhi Zhou, Xu Chen, and Qing Ling. 2019. Resource price-aware offloading for edge-cloud collaboration: A two-timescale online con- trol approach. IEEE Transactions on Cloud Computing 10, 1 (2019), 648–661

    work page 2019

  21. [21]

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Per- ona, Deva Ramanan, Piotr Dollár, and C Lawrence Zitnick. 2014. Mi- crosoft coco: Common objects in context. In European Conference on Computer Vision (ECCV). 740–755

    work page 2014

  22. [22]

    Jing Liu, Yao Du, Kun Yang, Jiaqi Wu, Yan Wang, Xiping Hu, Zehua Wang, Yang Liu, Peng Sun, Azzedine Boukerche, and Victor C. M. Le- ung. 2026. Edge-Cloud Collaborative Computing on Distributed In- telligence and Model Optimization: A Survey. IEEE Communications Surveys & Tutorials 28 (2026), 5049–5080

    work page 2026

  23. [23]

    Shengzhong Liu, Tianshi Wang, Jinyang Li, Dachun Sun, Mani Sri- vastava, and Tarek Abdelzaher. 2022. Adamask: Enabling machine- centric video streaming with adaptive frame masking for dnn infer- ence offloading. In Proceedings of the 30th ACM International Confer- ence on Multimedia (MM) . 3035–3044

    work page 2022

  24. [24]

    Weihong Liu, Jiawei Geng, Zongwei Zhu, Jing Cao, and Zirui Lian

    work page

  25. [25]

    In ACM/IEEE Design Automation Conference (DAC)

    Sniper: Cloud-edge collaborative inference scheduling with neural network similarity modeling. In ACM/IEEE Design Automation Conference (DAC). 505–510

    work page

  26. [26]

    Burhan A Mudassar, Jong Hwan Ko, and Saibal Mukhopadhyay. 2018. Edge-cloud collaborative processing for intelligent internet of things: A case study on smart surveillance. In ACM/IEEE Design Automation Conference (DAC). 1–6. Conference acronym ’XX, June 03–05, 2026, Anonymous Authors

    work page 2018

  27. [27]

    Ragheb Rahmaniani, Shabbir Ahmed, Teodor Gabriel Crainic, Michel Gendreau, and Walter Rei. 2020. The Benders dual decomposition method. Operations Research 68, 3 (2020), 878–895

    work page 2020

  28. [28]

    Jiawei Shao and Jun Zhang. 2020. Communication-computation trade- off in resource-constrained edge inference. IEEE Communications Magazine 58, 12 (2020), 20–26

    work page 2020

  29. [29]

    Weisong Shi, Jie Cao, Quan Zhang, Youhuizi Li, and Lanyu Xu. 2016. Edge computing: Vision and challenges.IEEE Internet of Things journal 3, 5 (2016), 637–646

    work page 2016

  30. [30]

    Mingfeng Su, Guojun Wang, Kim-Kwang Raymond Choo, et al. 2022. Prediction-based resource deployment and task scheduling in edge- cloud collaborative computing. Wireless Communications and Mobile Computing 2022 (2022)

    work page 2022

  31. [31]

    Samer Takriti and Shabbir Ahmed. 2004. On robust optimization of two-stage systems. Mathematical Programming 99, 1 (2004), 109–126

    work page 2004

  32. [32]

    Lior Talker, Aviad Cohen, Erez Yosef, Alexandra Dana, and Michael Dinerstein. 2024. Mind the edge: Refining depth edges in sparsely-supervised monocular depth estimation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 10606–10616

    work page 2024

  33. [33]

    Yuhao Tian and Zheming Yang. 2025. SAEC: Scene-Aware Enhanced Edge-Cloud Collaborative Industrial Vision Inspection with Multi- modal LLM. arXiv: 2509.17136 [cs.CV] https://arxiv.org/abs/2509. 17136

    work page arXiv 2025

  34. [34]

    Can Wang, Sheng Zhang, Yu Chen, Zhuzhong Qian, Jie Wu, and Mingjun Xiao. 2020. Joint configuration adaptation and bandwidth al- location for edge-based real-time video analytics. In IEEE Conference on Computer Communications (INFOCOM) . 257–266

    work page 2020

  35. [35]

    Liang Wang, Kai Lu, Nan Zhang, Xiaoyang Qu, Jianzong Wang, Jiguang Wan, Guokuan Li, and Jing Xiao. 2023. Shoggoth: Towards ef- ficient edge-cloud collaborative real-time video inference via adaptive online learning. In ACM/IEEE Design Automation Conference (DAC) . 1–6

    work page 2023

  36. [36]

    Shibo Wang, Shusen Yang, and Cong Zhao. 2020. SurveilEdge: Real- time video query based on collaborative cloud-edge deep learning. In IEEE Conference on Computer Communications (INFOCOM) . 2519– 2528

    work page 2020

  37. [37]

    Yingchao Wang, Chen Yang, Shulin Lan, Liehuang Zhu, and Yan Zhang. 2024. End-Edge-Cloud Collaborative Computing for Deep Learning: A Comprehensive Survey. IEEE Communications Surveys & Tutorials 26, 4 (2024), 2647–2683

    work page 2024

  38. [38]

    Longyin Wen, Dawei Du, Zhaowei Cai, Zhen Lei, Ming-Ching Chang, Honggang Qi, Jongwoo Lim, Ming-Hsuan Yang, and Siwei Lyu. 2020. UA-DETRAC: A new benchmark and protocol for multi-object detec- tion and tracking. Computer Vision and Image Understanding 193 (2020), 102907

    work page 2020

  39. [39]

    Xiaowei Xu, Yukun Ding, Sharon Xiaobo Hu, Michael Niemier, Jason Cong, Yu Hu, and Yiyu Shi. 2018. Scaling for edge inference of deep neural networks. Nature Electronics 1, 4 (2018), 216–222

    work page 2018

  40. [40]

    Zheming Yang, Dieli Hu, Qi Guo, Lulu Zuo, and Wen Ji. 2023. Vi- sual E2C: AI-driven visual end-edge-cloud architecture for 6G in low- carbon smart cities. IEEE Wireless Communications 30, 3 (2023), 204– 210

    work page 2023

  41. [41]

    Zheming Yang, Wen Ji, Qi Guo, and Zhi Wang. 2023. JA VP: Joint- aware video processing with edge-cloud collaboration for DNN in- ference. In Proceedings of the 31st ACM International Conference on Multimedia (MM). 9152–9160

    work page 2023

  42. [42]

    Zheming Yang, Bing Liang, and Wen Ji. 2021. An intelligent end– edge–cloud architecture for visual IoT-assisted healthcare systems. IEEE Internet of Things Journal 8, 23 (2021), 16779–16786

    work page 2021

  43. [43]

    Mu Yuan, Lan Zhang, and Xiang-Yang Li. 2022. Mlink: Linking black- box models for collaborative multi-model inference. In Proceedings of the AAAI Conference on Artificial Intelligence . 9475–9483

    work page 2022

  44. [44]

    Bo Zeng and Long Zhao. 2013. Solving two-stage robust optimization problems using a column-and-constraint generation method. Opera- tions Research Letters 41, 5 (2013), 457–461

    work page 2013

  45. [45]

    Ben Zhang, Xin Jin, Sylvia Ratnasamy, John Wawrzynek, and Ed- ward A Lee. 2018. Awstream: Adaptive wide-area streaming analytics. In ACM Special Interest Group on Data Communication (SIGCOMM) . 236–252

    work page 2018

  46. [46]

    Haoyu Zhang, Ganesh Ananthanarayanan, Peter Bodik, Matthai Phili- pose, Paramvir Bahl, and Michael J Freedman. 2017. Live video analyt- ics at scale with approximation and delay-tolerance. In 14th USENIX Symposium on Networked Systems Design and Implementation (NSDI) . 377–392

    work page 2017

  47. [47]

    Bolei Zhou, Hang Zhao, Xavier Puig, Sanja Fidler, Adela Barriuso, and Antonio Torralba. 2017. Scene Parsing through ADE20K Dataset. In 2017 IEEE Conference on Computer Vision and Pattern Recognition . 5122–5130

    work page 2017

  48. [48]

    Zhi Zhou, Xu Chen, En Li, Liekang Zeng, Ke Luo, and Junshan Zhang

    work page

  49. [49]

    Edge intelligence: Paving the last mile of artificial intelligence with edge computing. Proc. IEEE 107, 8 (2019), 1738–1762

    work page 2019

  50. [50]

    Lulu Zuo, Qingfang Zheng, Zheming Yang, and Wen Ji. 2025. AODMS: Adaptive Online Edge-Cloud Collaborative Inference with Dynamic Model Switching and Resource Allocation. In 31th IEEE International Conference on Parallel and Distributed Systems . 1–8

    work page 2025