pith. machine review for the scientific record. sign in

arxiv: 2605.06027 · v1 · submitted 2026-05-07 · 💻 cs.NI

Recognition: unknown

FluxShard: Motion-Aware Feature Cache Reuse for Collaborative Video Analytics in Mobile Edge Computing

Xiuxian Guan , Zongyuan Zhang , Zheng Lin , Zekai Sun , Tianyang Duan , Zihan Fang , Rui Wang , Heming Cui
show 3 more authors
Wei Ni Jun Luo Yuanwei Liu
Authors on Pith no claims yet

Pith reviewed 2026-05-08 04:58 UTC · model grok-4.3

classification 💻 cs.NI
keywords video analyticsfeature cachingmotion vectorsedge computingmobile edge computingcollaborative video analyticscache reusereceptive field alignment
0
0 comments X

The pith

FluxShard manages feature caches at per-region motion granularity using codec motion vectors to reduce redundant computation in edge video analytics.

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

Existing caching methods treat entire video frames as single units and lose most reusable features when different parts of the scene move at different speeds. FluxShard tracks motion at the level of individual codec blocks and re-indexes cached features according to each block's displacement, separating pure movement from actual content change. A new Receptive Field Alignment Principle then uses only the motion vector field to decide which positions truly require fresh computation because their receptive fields now contain inconsistent material. A dispatcher finally splits the remaining work between edge device and cloud. The result is sustained high reuse ratios across frames while meeting a given accuracy target.

Core claim

By re-indexing cached features along per-block motion vectors and applying the Receptive Field Alignment Principle to detect receptive-field misalignment, FluxShard separates spatial displacement from content change and sustains high cache reuse ratios across frames under non-uniform motion.

What carries the argument

Receptive Field Alignment Principle (RFAP): a rule that, from the input-level motion-vector field alone, identifies exactly which feature positions must be recomputed because their receptive fields now contain inconsistent spatial composition.

If this is right

  • Latency drops 32.6-83.8 percent compared with whole-scene baselines while accuracy stays within budget.
  • Energy consumption falls 14.9-64.0 percent under the same accuracy constraint.
  • MV-guided remapping keeps cache coherence across frames and maintains high reuse ratios over time.
  • Only a sparse residual workload remains, which a profiling-driven dispatcher routes efficiently between edge and cloud.

Where Pith is reading between the lines

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

  • Similar per-region remapping could be applied to other streaming sensor data where local motion or change varies spatially.
  • In networks with lower motion-vector overhead the recomputation savings would grow further.
  • The approach implicitly trades a small amount of per-frame remapping cost for large reductions in feature transmission and recomputation.

Load-bearing premise

The Receptive Field Alignment Principle can correctly identify from the motion vector field alone exactly which positions must be recomputed due to inconsistent spatial composition within receptive fields.

What would settle it

A video sequence in which block motion vectors indicate uniform regional shifts yet reusing the remapped features causes accuracy to fall below the target budget because of unaccounted content variation inside blocks.

Figures

Figures reproduced from arXiv: 2605.06027 by Heming Cui, Jun Luo, Rui Wang, Tianyang Duan, Wei Ni, Xiuxian Guan, Yuanwei Liu, Zekai Sun, Zheng Lin, Zihan Fang, Zongyuan Zhang.

Figure 1
Figure 1. Figure 1: Motivation for FluxShard. All panels use pose estimation view at source ↗
Figure 2
Figure 2. Figure 2: illustrates the pipeline. FluxShard introduces a dispatch layer (virtual layer 0) preceding the DNN. This layer applies the identity operator to the raw input, so its reuse criterion reduces to a direct per-position comparison between the current frame and the MV-aligned cached input. For both endpoints, it stores a cached input (Fˆ e 0 , Fˆ c 0 ) and an accumu￾lated MV field (mˆ e 0 , mˆ c 0 ) that tracks… view at source ↗
Figure 3
Figure 3. Figure 3: Testbed. Left: three mobile robots with embedded Jetson Xavier view at source ↗
Figure 4
Figure 4. Figure 4: End-to-end latency (solid bars, left axis) and edge energy (hatched view at source ↗
Figure 5
Figure 5. Figure 5: Mean per-frame latency breakdown under medium bandwidth. Pre view at source ↗
Figure 6
Figure 6. Figure 6: Inference latency vs. compute ratio for pose estimation under view at source ↗
Figure 7
Figure 7. Figure 7: End-to-end latency (solid bars, left axis) and edge energy (hatched view at source ↗
read the original abstract

Caching and reusing intermediate features across consecutive frames is a common technique to reduce redundant computation and transmission for edge-cloud video analytics in mobile edge computation. Existing methods manage the cache in a fixed or globally shifted coordinate system, treating it as an indivisible whole. Under the non-uniform motion patterns of mobile scenes, this whole-scene granularity invalidates large portions of the cache even when most content has merely shifted spatially, wasting computation and bandwidth. The root cause is a granularity mismatch: the cache is managed per scene, yet motion varies per region. In this paper, we present FluxShard, a motion-aware edge-cloud video analytics system that uses codec-level block motion vectors (MVs) to manage feature cache reuse and recomputation at the granularity of individual motion regions. By re-indexing cached features along per-block MVs, FluxShard separates spatial displacement from content changes, recovering reusable content that whole-scene methods would otherwise discard. To ensure correct reuse under heterogeneous motion, the Receptive Field Alignment Principle (RFAP) identifies, from the input-level MV field alone, the positions that must be recomputed due to inconsistent spatial composition within receptive fields. To maintain cache coherence across frames, MV-guided cache remapping warps the entire feature cache to the current coordinate system each frame, sustaining a high reuse ratio over time. A profiling-driven dispatcher routes the remaining sparse workload between edge and cloud for lower latency. Evaluation across multiple vision tasks, dynamic video benchmarks, and network conditions shows that FluxShard reduces latency by 32.6-83.8% and energy by 14.9-64.0% over all baselines under the prescribed accuracy budget.

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

3 major / 2 minor

Summary. The paper presents FluxShard, a motion-aware edge-cloud video analytics system that manages feature cache reuse at the granularity of individual motion regions using codec-level block motion vectors (MVs). It introduces the Receptive Field Alignment Principle (RFAP) to identify recomputation positions from the MV field to handle inconsistent receptive-field content under non-uniform motion, employs MV-guided cache remapping for coherence across frames, and uses a profiling-driven dispatcher to route sparse workloads. The central empirical claim is that FluxShard achieves 32.6-83.8% latency reduction and 14.9-64.0% energy reduction over baselines across multiple vision tasks, dynamic video benchmarks, and network conditions while remaining within a prescribed accuracy budget.

Significance. If the results hold and RFAP reliably preserves accuracy, the work could meaningfully advance collaborative video analytics in mobile edge computing by addressing the granularity mismatch between whole-scene cache management and regional motion patterns. The practical reliance on existing codec MVs without extra overhead is a strength, and the broad evaluation across tasks and conditions suggests potential for real deployments. The empirical systems focus provides concrete gains, but significance hinges on robust validation of accuracy and baselines.

major comments (3)
  1. [Evaluation] Evaluation section: The abstract reports concrete latency (32.6-83.8%) and energy (14.9-64.0%) reductions under an accuracy budget, but provides no details on baseline implementations, exact accuracy budgets, error bars, or how accuracy was verified to remain within budget. This is load-bearing for assessing whether the gains are achieved without violating the accuracy constraint.
  2. [RFAP] Receptive Field Alignment Principle (RFAP) description: RFAP is the load-bearing mechanism claimed to correctly identify, from the input MV field alone, exactly which positions must be recomputed due to inconsistent spatial composition within receptive fields. No ablation studies, formal derivation, or empirical validation (e.g., comparison to explicit feature differencing) is provided to confirm it neither misses positions (risking accuracy loss) nor over-flags them (reducing the reported reuse gains).
  3. [Cache Remapping and Dispatcher] Cache remapping and dispatcher: The MV-guided remapping is stated to sustain high reuse ratios over time, yet no quantitative cache hit-rate analysis, sensitivity to codec MV errors, or breakdown of dispatcher decisions across network conditions is given, leaving the long-term coherence claim unsupported.
minor comments (2)
  1. [Abstract] Abstract: The phrase 'post-hoc accuracy constraints' is mentioned without elaboration on enforcement or measurement, which could be clarified for readers.
  2. [Introduction] Notation: The distinction between 'motion regions' and 'receptive fields' could be defined more explicitly early in the paper to avoid ambiguity in the RFAP description.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback and the recommendation for major revision. The comments highlight important areas where additional details and validation will strengthen the empirical claims and mechanisms in FluxShard. We address each major comment below and commit to incorporating the requested clarifications and analyses in the revised manuscript.

read point-by-point responses

  1. Referee: [Evaluation] Evaluation section: The abstract reports concrete latency (32.6-83.8%) and energy (14.9-64.0%) reductions under an accuracy budget, but provides no details on baseline implementations, exact accuracy budgets, error bars, or how accuracy was verified to remain within budget. This is load-bearing for assessing whether the gains are achieved without violating the accuracy constraint.

    Authors: We agree that the evaluation section requires more explicit details to support the reported gains. In the revised manuscript, we will expand this section to describe all baseline implementations in full, specify the exact accuracy budgets/thresholds used per task, include error bars from repeated runs with statistical significance, and detail the verification methodology (including how accuracy was measured against ground truth and ensured to stay within budget). These additions will make the latency and energy reductions fully verifiable. revision: yes

  2. Referee: [RFAP] Receptive Field Alignment Principle (RFAP) description: RFAP is the load-bearing mechanism claimed to correctly identify, from the input MV field alone, exactly which positions must be recomputed due to inconsistent spatial composition within receptive fields. No ablation studies, formal derivation, or empirical validation (e.g., comparison to explicit feature differencing) is provided to confirm it neither misses positions (risking accuracy loss) nor over-flags them (reducing the reported reuse gains).

    Authors: The referee correctly identifies the need for stronger substantiation of RFAP. We will add to the revised paper: a formal derivation of the principle based on receptive field geometry and motion vector fields, ablation studies comparing RFAP decisions against explicit per-position feature differencing, and quantitative results (e.g., accuracy impact and reuse ratio trade-offs) showing that RFAP avoids both under- and over-recomputation. This will empirically validate its reliability as the core mechanism. revision: yes

  3. Referee: [Cache Remapping and Dispatcher] Cache remapping and dispatcher: The MV-guided remapping is stated to sustain high reuse ratios over time, yet no quantitative cache hit-rate analysis, sensitivity to codec MV errors, or breakdown of dispatcher decisions across network conditions is given, leaving the long-term coherence claim unsupported.

    Authors: We acknowledge that quantitative evidence for sustained cache coherence and dispatcher behavior is needed. In the revision, we will include time-series cache hit-rate measurements across video sequences, sensitivity analysis to injected codec MV errors (measuring effects on hit rates and accuracy), and per-condition breakdowns of dispatcher routing decisions (e.g., edge vs. cloud allocation under varying bandwidth). These will directly support the long-term reuse and performance claims. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical systems design with no self-referential derivations or fitted predictions

full rationale

The paper presents FluxShard as an engineering system that applies codec motion vectors to manage per-region feature cache reuse, with RFAP introduced as a heuristic rule for detecting receptive-field inconsistency. No equations, derivations, or first-principles claims are shown that reduce to fitted parameters or self-definitions; performance numbers (latency/energy reductions) are obtained from direct experimental comparison against baselines under an accuracy budget, not from any internal prediction that is forced by construction. The central mechanism (MV-guided remapping plus RFAP mask) is evaluated externally rather than justified by self-citation chains or ansatz smuggling. This is a standard non-circular empirical contribution.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Only the abstract is available, so the ledger is necessarily incomplete. The system appears to rest on domain assumptions about motion vectors and receptive-field consistency rather than new invented entities or many fitted parameters.

axioms (2)
  • domain assumption Codec-level block motion vectors accurately capture the spatial displacement of visual content between frames for the purpose of feature cache remapping.
    Invoked when the paper states that re-indexing cached features along per-block MVs separates displacement from content change.
  • ad hoc to paper The Receptive Field Alignment Principle correctly flags recomputation positions solely from the MV field when receptive-field content becomes inconsistent.
    Central to ensuring correct reuse under heterogeneous motion; introduced as a new principle in the paper.

pith-pipeline@v0.9.0 · 5636 in / 1477 out tokens · 21968 ms · 2026-05-08T04:58:25.651400+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

59 extracted references · 12 canonical work pages · 3 internal anchors

  1. [1]

    Deepvision: Enhanced drone detection and recognition in visible im- agery through deep learning networks,

    H. J. Al Dawasari, M. Bilal, M. Moinuddin, K. Arshad, and K. Assaleh, “Deepvision: Enhanced drone detection and recognition in visible im- agery through deep learning networks,”Sensors, vol. 23, 2023

    2023

  2. [2]

    Person recognition in aerial surveillance: A decade survey,

    K. Nguyen, F. Liu, C. Fookes, S. Sridharan, X. Liu, and A. Ross, “Person recognition in aerial surveillance: A decade survey,”IEEE Transactions on Biometrics, Behavior, and Identity Science, 2025

    2025

  3. [3]

    Helpful DoggyBot: Open-world object fetching using legged robots and vision-language models,

    Q. Wu, Z. Fu, X. Cheng, X. Wang, and C. Finn, “Helpful DoggyBot: Open-world object fetching using legged robots and vision-language models,”arXiv:2410.00231, 2024

    work page arXiv 2024

  4. [4]

    Fetchbench: A simulation benchmark for robot fetching,

    B. Han, M. Parakh, D. Geng, J. A. Defay, L. Gan, and J. Deng, “Fetchbench: A simulation benchmark for robot fetching,” inProc. CoRL, 2024, pp. 3053–3071

    2024

  5. [5]

    VDO-SLAM: A visual dynamic object-aware SLAM system,

    J. Zhang, M. Henein, R. Mahony, and V . Ila, “VDO-SLAM: A visual dynamic object-aware SLAM system,”arXiv:2005.11052, 2021

    work page arXiv 2005

  6. [6]

    Robust scene aware multi-object tracking for surveillance videos,

    F. Jalali, M. Khademi, A. E. Moghadam, and H. S. Yazdi, “Robust scene aware multi-object tracking for surveillance videos,”Neurocomputing, vol. 638, p. 130114, 2025

    2025

  7. [7]

    Efficient Parallel Split Learning over Resource-Constrained Wireless Edge Networks,

    Z. Lin, G. Zhu, Y . Deng, X. Chen, Y . Gao, K. Huang, and Y . Fang, “Efficient Parallel Split Learning over Resource-Constrained Wireless Edge Networks,”IEEE Trans. Mobile Comput., vol. 23, no. 10, pp. 9224–9239, 2024

    2024

  8. [8]

    IC3M: In-Car Multimodal Multi-Object Monitoring for Abnormal Status of Both Driver and Passengers,

    Z. Fang, Z. Lin, S. Hu, H. Cao, Y . Deng, X. Chen, and Y . Fang, “IC3M: In-Car Multimodal Multi-Object Monitoring for Abnormal Status of Both Driver and Passengers,”arXiv preprint arXiv:2410.02592, 2024

    work page arXiv 2024

  9. [9]

    SatSense: Multi-Satellite Collaborative Framework for Spectrum Sensing,

    H. Yuan, Z. Chen, Z. Lin, J. Peng, Z. Fang, Y . Zhong, Z. Song, and Y . Gao, “SatSense: Multi-Satellite Collaborative Framework for Spectrum Sensing,”IEEE Trans. Cogn. Commun. Netw., 2025

    2025

  10. [10]

    Gapsl: A gradient-aligned parallel split learning on heterogeneous data,

    Z. Lin, O. Aouedi, W. Ni, S. Chatzinotas, and X. Chen, “Gapsl: A gradient-aligned parallel split learning on heterogeneous data,”arXiv preprint arXiv:2603.18540, 2026

    work page arXiv 2026

  11. [11]

    Conflict-aware client selection for multi-server federated learning,

    M. Hong, Z. Lin, Z. Lin, L. Li, M. Yang, X. Du, Z. Fang, Z. Kang, D. Luan, and S. Zhu, “Conflict-aware client selection for multi-server federated learning,”arXiv preprint arXiv:2602.02458, 2026

    work page arXiv 2026

  12. [12]

    SUMS: Sniffing Unknown Multiband Signals under Low Sampling Rates,

    J. Peng, Z. Chen, Z. Lin, H. Yuan, Z. Fang, L. Bao, Z. Song, Y . Li, J. Ren, and Y . Gao, “SUMS: Sniffing Unknown Multiband Signals under Low Sampling Rates,”IEEE Trans. Mobile Comput., 2024

    2024

  13. [13]

    Satfed: A resource-efficient leo satellite-assisted heterogeneous federated learning framework,

    Y . Zhang, Z. Lin, Z. Chen, Z. Fang, W. Zhu, X. Chen, J. Zhao, and Y . Gao, “Satfed: A resource-efficient leo satellite-assisted heterogeneous federated learning framework,”Engineering, 2024

    2024

  14. [14]

    Fedsn: A federated learning framework over heterogeneous leo satellite networks,

    Z. Lin, Z. Chen, Z. Fang, X. Chen, X. Wang, and Y . Gao, “Fedsn: A federated learning framework over heterogeneous leo satellite networks,” IEEE Transactions on Mobile Computing, vol. 24, no. 3, pp. 1293–1307, 2024

    2024

  15. [15]

    Spinn: synergistic progressive inference of neural networks over device and cloud,

    S. Laskaridis, S. I. Venieris, M. Almeida, I. Leontiadis, and N. D. Lane, “Spinn: synergistic progressive inference of neural networks over device and cloud,” inProc. ACM MobiCom, 2020, pp. 1–15

    2020

  16. [16]

    Accelerating end- cloud collaborative inference via near bubble-free pipeline optimization,

    L. Gao, J. Liu, H. Xu, S. Xu, Q. Ma, and L. Huang, “Accelerating end- cloud collaborative inference via near bubble-free pipeline optimization,” inProc. IEEE INFOCOM, 2025, pp. 1–10. IEEE TRANSACTIONS ON MOBILE COMPUTING 14

    2025

  17. [17]

    Cbinfer: Change-based inference for convolutional neural networks on video data,

    L. Cavigelli, P. Degen, and L. Benini, “Cbinfer: Change-based inference for convolutional neural networks on video data,” inProc. ICDSC, 2017, p. 1–8

    2017

  18. [18]

    Diffy: a d ´ej`a vu-free differ- ential deep neural network accelerator,

    M. Mahmoud, K. Siu, and A. Moshovos, “Diffy: a d ´ej`a vu-free differ- ential deep neural network accelerator,” inProc. IEEE/ACM MICRO, 2018, pp. 134–147

    2018

  19. [19]

    Deltacnn: End-to-end cnn inference of sparse frame differences in videos,

    M. Parger, C. Tang, C. D. Twigg, C. Keskin, R. Wang, and M. Stein- berger, “Deltacnn: End-to-end cnn inference of sparse frame differences in videos,” inProc. IEEE/CVF CVPR, 2022, pp. 12 497–12 506

    2022

  20. [20]

    MotionDeltaCNN: Sparse CNN inference of frame differences in moving camera videos with spherical buffers and padded convolutions,

    M. Parger, C. Tang, T. Neff, C. D. Twigg, C. Keskin, R. Wang, and M. Steinberger, “MotionDeltaCNN: Sparse CNN inference of frame differences in moving camera videos with spherical buffers and padded convolutions,” inProc. IEEE/CVF ICCV, 2023, pp. 17 292–17 301

    2023

  21. [21]

    YOLOv11: An Overview of the Key Architectural Enhancements

    R. Khanam and M. Hussain, “Yolov11: An overview of the key architectural enhancements,”arXiv:2410.17725, 2024

    work page internal anchor Pith review arXiv 2024

  22. [22]

    A variegated look at 5g in the wild: performance, power, and QoE implications,

    A. Narayanan, X. Zhang, R. Zhu, A. Hassan, S. Jin, X. Zhu, X. Zhang, D. Rybkin, Z. Yang, Z. M. Mao, F. Qian, and Z.-L. Zhang, “A variegated look at 5g in the wild: performance, power, and QoE implications,” in Proc. ACM SIGCOMM, 2021, pp. 610–625

    2021

  23. [23]

    Deep generative adversarial compression artifact removal,

    L. Galteri, L. Seidenari, M. Bertini, and A. D. Bimbo, “Deep generative adversarial compression artifact removal,” inProc. IEEE ICCV, 2017, pp. 4836–4845, ISSN: 2380-7504

    2017

  24. [24]

    Video coding for machines: A paradigm of collaborative compression and intelligent analytics,

    L. Duan, J. Liu, W. Yang, T. Huang, and W. Gao, “Video coding for machines: A paradigm of collaborative compression and intelligent analytics,”IEEE Transactions on Image Processing, vol. 29, pp. 8680– 8695, 2020

    2020

  25. [25]

    Scalable image coding for humans and machines,

    H. Choi and I. V . Baji ´c, “Scalable image coding for humans and machines,”IEEE Transactions on Image Processing, vol. 31, pp. 2739– 2754, 2022

    2022

  26. [26]

    The 2017 DAVIS Challenge on Video Object Segmentation

    J. Pont-Tuset, F. Perazzi, S. Caelles, P. Arbel ´aez, A. Sorkine-Hornung, and L. Van Gool, “The 2017 davis challenge on video object segmen- tation,”arXiv:1704.00675, 2017

    work page internal anchor Pith review arXiv 2017

  27. [27]

    Recovering accurate 3d human pose in the wild using imus and a moving camera,

    T. von Marcard, R. Henschel, M. Black, B. Rosenhahn, and G. Pons- Moll, “Recovering accurate 3d human pose in the wild using imus and a moving camera,” inProc. ECCV, sep 2018

    2018

  28. [28]

    Keepedge: A knowledge distillation empowered edge intelligence framework for visual assisted positioning in UA V delivery,

    H. Luo, T. Chen, X. Li, S. Li, C. Zhang, G. Zhao, and X. Liu, “Keepedge: A knowledge distillation empowered edge intelligence framework for visual assisted positioning in UA V delivery,”IEEE Trans. Mob. Comput., vol. 22, pp. 4729–4741, 2023

    2023

  29. [29]

    Reshaping edge-assisted visual SLAM by embracing on-chip intelligence,

    D. Li, Y . Zhao, J. Xu, S. Zhang, L. Shangguan, Q. Ma, X. Ding, and Z. Yang, “Reshaping edge-assisted visual SLAM by embracing on-chip intelligence,”IEEE Trans. Mob. Comput., vol. 23, pp. 12 983–12 997, 2024

    2024

  30. [30]

    Ifresher: Information freshening for mobile augmented reality with multi-agent reinforcement learning in edge computing,

    S. Cheng, Z. Wang, F. Feng, Y . Zhang, T. Bi, and T. Jiang, “Ifresher: Information freshening for mobile augmented reality with multi-agent reinforcement learning in edge computing,”IEEE Trans. Mob. Comput., vol. 24, pp. 11 703–11 716, 2025

    2025

  31. [31]

    Enabling real-time video detection with adaptive and distributed scheduling in mobile edge computing,

    Z. Liu, Y . Wang, Y . Zhao, C. Qiu, C. Zhang, X. Wang, and M. Dong, “Enabling real-time video detection with adaptive and distributed scheduling in mobile edge computing,”IEEE Trans. Mob. Comput., vol. 24, pp. 12 784–12 801, 2025

    2025

  32. [32]

    Fasttuner: Fast resolution and model tuning for multi-object tracking in edge video analytics,

    R. Xu, K. Nalaie, and R. Zheng, “Fasttuner: Fast resolution and model tuning for multi-object tracking in edge video analytics,”IEEE Trans. Mob. Comput., vol. 24, pp. 4747–4761, 2025

    2025

  33. [33]

    Mystique: User-level adaptation for real-time video analytics in edge networks via meta-rl,

    X. Shi, S. Zhang, M. Liu, L. Meng, L. Wei, Y . Gu, K. Liu, H. Cheng, Y . Song, L. Tang, A. Zhu, N. Chen, and Z. Qian, “Mystique: User-level adaptation for real-time video analytics in edge networks via meta-rl,” IEEE Trans. Mob. Comput., vol. 24, pp. 3615–3632, 2025

    2025

  34. [34]

    Recurrent residual module for fast inference in videos,

    B. Pan, W. Lin, X. Fang, C. Huang, B. Zhou, and C. Lu, “Recurrent residual module for fast inference in videos,” inProc. IEEE/CVF CVPR, 2018, pp. 1536–1545, ISSN: 2575-7075

    2018

  35. [35]

    Skip- convolutions for efficient video processing,

    A. Habibian, D. Abati, T. S. Cohen, and B. Ehteshami Bejnordi, “Skip- convolutions for efficient video processing,” inProc. IEEE/CVF CVPR, 2021, pp. 2694–2703, ISSN: 2575-7075

    2021

  36. [36]

    Glimpse: Continuous, real-time object recognition on mobile devices,

    T. Y .-H. Chen, L. Ravindranath, S. Deng, P. Bahl, and H. Balakrishnan, “Glimpse: Continuous, real-time object recognition on mobile devices,” inProc. ACM SenSys, 2015, pp. 155–168

    2015

  37. [37]

    Reducto: On-camera filtering for resource-efficient real-time video analytics,

    Y . Li, A. Padmanabhan, P. Zhao, Y . Wang, G. H. Xu, and R. Netravali, “Reducto: On-camera filtering for resource-efficient real-time video analytics,” inProc. ACM SIGCOMM, 2020, pp. 359–376

    2020

  38. [38]

    Chameleon: scalable adaptation of video analytics,

    J. Jiang, G. Ananthanarayanan, P. Bodik, S. Sen, and I. Stoica, “Chameleon: scalable adaptation of video analytics,” inProc. ACM SIGCOMM, 2018, pp. 253–266

    2018

  39. [39]

    Accmpeg: Optimizing video encoding for accurate video analytics,

    K. Du, Q. Zhang, A. Arapin, H. Wang, Z. Xia, and J. Jiang, “Accmpeg: Optimizing video encoding for accurate video analytics,” inProc. MLSys, 2022

    2022

  40. [40]

    Quantization and training of neural networks for efficient integer-arithmetic-only inference,

    B. Jacob, S. Kligys, B. Chen, M. Zhu, M. Tang, A. Howard, H. Adam, and D. Kalenichenko, “Quantization and training of neural networks for efficient integer-arithmetic-only inference,” inProc. IEEE/CVF CVPR, 2018, pp. 2704–2713

    2018

  41. [41]

    Learning both weights and connections for efficient neural networks,

    S. Han, J. Pool, J. Tran, and W. J. Dally, “Learning both weights and connections for efficient neural networks,” inProc. NeurIPS, vol. 1, 2015, pp. 1135–1143

    2015

  42. [42]

    Neurosurgeon: Collaborative intelligence between the cloud and mobile edge,

    Y . Kang, J. Hauswald, C. Gao, A. Rovinski, T. Mudge, J. Mars, and L. Tang, “Neurosurgeon: Collaborative intelligence between the cloud and mobile edge,” inProc. ACM ASPLOS, 2017, pp. 615–629

    2017

  43. [43]

    Distributed DNN inference with fine-grained model partitioning in mobile edge computing networks,

    H. Li, X. Li, Q. Fan, Q. He, X. Wang, and V . C. M. Leung, “Distributed DNN inference with fine-grained model partitioning in mobile edge computing networks,”IEEE Trans. Mob. Comput., vol. 23, pp. 9060– 9074, 2024

    2024

  44. [44]

    Hastening stream offloading of inference via multi-exit dnns in mobile edge computing,

    Z. Liu, J. Song, C. Qiu, X. Wang, X. Chen, Q. He, and H. Sheng, “Hastening stream offloading of inference via multi-exit dnns in mobile edge computing,”IEEE Trans. Mob. Comput., vol. 23, pp. 535–548, 2024

    2024

  45. [45]

    Joint DNN model deployment, selection, and configuration for heterogeneous inference services toward edge intelligence,

    H. Huang, J. Liang, and G. Min, “Joint DNN model deployment, selection, and configuration for heterogeneous inference services toward edge intelligence,”IEEE Trans. Mob. Comput., vol. 24, pp. 12 726– 12 741, 2025

    2025

  46. [46]

    Video offloading in mobile edge comput- ing: Dealing with uncertainty,

    W. Ma and L. Mashayekhy, “Video offloading in mobile edge comput- ing: Dealing with uncertainty,”IEEE Trans. Mob. Comput., vol. 23, pp. 10 251–10 264, 2024

    2024

  47. [47]

    SplitLoRA: A split parameter-efficient fine-tuning framework for large language models.arXiv preprint arXiv:2407.00952, 2024

    Z. Lin, X. Hu, Y . Zhang, Z. Chen, Z. Fang, X. Chen, A. Li, P. Vepakomma, and Y . Gao, “SplitLoRA: A Split Parameter-Efficient Fine-Tuning Framework for Large Language Models,”arXiv preprint arXiv:2407.00952, 2024

    work page arXiv 2024

  48. [48]

    Hfedmoe: Resource-aware heterogeneous federated learning with mixture-of-experts,

    Z. Fang, Z. Lin, S. Hu, Y . Ma, Y . Tao, Y . Deng, X. Chen, and Y . Fang, “Hfedmoe: Resource-aware heterogeneous federated learning with mixture-of-experts,”arXiv preprint arXiv:2601.00583, 2026

    work page arXiv 2026

  49. [49]

    Split Learning in 6G Edge Networks,

    Z. Lin, G. Qu, X. Chen, and K. Huang, “Split Learning in 6G Edge Networks,”IEEE Wirel. Commun., 2024

    2024

  50. [50]

    Mobile edge intelligence for large language models: A contemporary survey,

    G. Qu, Q. Chen, W. Wei, Z. Lin, X. Chen, and K. Huang, “Mobile edge intelligence for large language models: A contemporary survey,”IEEE Communications Surveys & Tutorials, 2025

    2025

  51. [51]

    Automated Federated Pipeline for Parameter-Efficient Fine-Tuning of Large Lan- guage Models,

    Z. Fang, Z. Lin, Z. Chen, X. Chen, Y . Gao, and Y . Fang, “Automated Federated Pipeline for Parameter-Efficient Fine-Tuning of Large Lan- guage Models,”IEEE Trans. Mobile Comput., 2025

    2025

  52. [52]

    Hi- erarchical Split Federated Learning: Convergence Analysis and System Optimization,

    Z. Lin, W. Wei, Z. Chen, C.-T. Lam, X. Chen, Y . Gao, and J. Luo, “Hi- erarchical Split Federated Learning: Convergence Analysis and System Optimization,”IEEE Trans. Mobile Comput., 2025

    2025

  53. [53]

    Optimizing Split Federated Learning with Unstable Client Participation

    W. Wei, Z. Lin, X. Liu, H. Du, D. Niyato, and X. Chen, “Optimizing split federated learning with unstable client participation,”arXiv preprint arXiv:2509.17398, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  54. [54]

    Nsc-sl: A bandwidth-aware neural subspace compression for communication-efficient split learning,

    Z. Fang, M. Yang, Z. Lin, Z. Lin, Z. Fang, Z. Zhang, T. Duan, D. Huang, and S. Zhu, “Nsc-sl: A bandwidth-aware neural subspace compression for communication-efficient split learning,”arXiv preprint arXiv:2602.02696, 2026

    work page arXiv 2026

  55. [55]

    HASFL: Heterogeneity- aware Split Federated Learning over Edge Computing Systems,

    Z. Lin, Z. Chen, X. Chen, W. Ni, and Y . Gao, “HASFL: Heterogeneity- aware Split Federated Learning over Edge Computing Systems,”IEEE Trans. Mobile Comput., 2026

    2026

  56. [56]

    Accelerating Federated Learning with Model Segmentation for Edge Networks,

    M. Hu, J. Zhang, X. Wang, S. Liu, and Z. Lin, “Accelerating Federated Learning with Model Segmentation for Edge Networks,”IEEE Trans. Green Commun. Netw., 2024

    2024

  57. [57]

    Aggregation alignment for federated learning with mixture-of-experts under data heterogeneity,

    Z. Fang, Q. Wang, H. An, Z. Lin, Y . Deng, X. Chen, and Y . Fang, “Aggregation alignment for federated learning with mixture-of-experts under data heterogeneity,”arXiv preprint arXiv:2603.21276, 2026

    work page arXiv 2026

  58. [58]

    Optimal resource allocation for u-shaped parallel split learning,

    S. Lyu, Z. Lin, G. Qu, X. Chen, X. Huang, and P. Li, “Optimal resource allocation for u-shaped parallel split learning,” in2023 IEEE Globecom Workshops (GC Wkshps), 2023, pp. 197–202

    2023

  59. [59]

    Adaptsfl: Adaptive Split Federated Learning in Resource-Constrained Edge Networks,

    Z. Lin, G. Qu, W. Wei, X. Chen, and K. K. Leung, “Adaptsfl: Adaptive Split Federated Learning in Resource-Constrained Edge Networks,” IEEE Trans. Netw., 2025

    2025