Pith. sign in

REVIEW 4 major objections 5 minor 34 references

Adaptive Configuration Selection for Multi-Model Inference Pipelines in Edge Computing

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper claims that a policy-gradient reinforcement-learning agent, fed by LSTM workload forecasts and residual-network features, can select model variants, replica counts, and batch sizes for multi-model edge inference pipelines so as…

desk verdict A plausible RL-based pipeline configurator on a real Kubernetes testbed, but the accuracy metric and headline claims don't hold up under inspection. read the letter →

arxiv 2506.02814 v2 pith:G3BYNKUP submitted 2025-06-03 cs.DC

Jinhao Sheng , Zhiqing Tang , Jianxiong Guo , Tian Wang This is my paper
classification cs.DC
keywords edgecomputingmulti-modelinferencepipelinereinforcementlearningpolicygradientProximalOptimizationLSTMworkloadpredictionKubernetesonlineconfigurationselection
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that an online reinforcement-learning agent can make configuration decisions for multi-model inference pipelines on edge devices—choosing which model variant, how many replicas, and what batch size at each stage—in a way that improves QoS and lowers cost compared with random, greedy, and solver-based baselines, while keeping decision time flat as pipelines grow. The authors model the problem as a Markov Decision Process, predict incoming load with an LSTM, extract node and pipeline features with residual blocks, and train a policy-gradient (PPO-style) controller. They evaluate in a real Kubernetes cluster under steady and fluctuating workloads. If the results hold, edge operators could automate pipeline tuning without hand-crafted rules.

What carries the argument

The load-bearing object is the OPD algorithm, a policy-gradient reinforcement-learning controller whose action is the full configuration vector $a_t=[(z_{1,t},f_{1,t},b_{1,t}),\ldots,(z_{n,t},f_{n,t},b_{n,t})]$—model variant, replica count, and batch size for every stage—and whose reward is $r_t=Q-\beta C-\gamma B$. The state is built from node resources, observed and LSTM-predicted incoming load, and per-stage performance, refined through residual blocks before entering the policy network. This design is what lets the algorithm coordinate decisions across stages and keep decision-time complexity linear in the number of tasks.

What would settle it

Record true end-to-end latency, throughput, and accuracy for the same workloads while the agent runs, then check whether the configurations OPD selects actually win on those direct measurements and not just on Eq. (3); if the direct ranking contradicts the Eq. (3) ranking, the central claim collapses.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central claim is that the Online Pipeline Decision (OPD) algorithm balances cost and QoS for multi-stage inference pipelines by treating configuration selection as an MDP and learning a policy that maps node and pipeline state—including predicted load—to per-stage choices of model variant, replica count, and batch size. The policy is trained with a clipped policy-gradient objective, aided by an expert-guided warm start, and the state representation combines an LSTM workload predictor with residual-network feature extraction. In cluster experiments, OPD reports higher QoS than the cost-minimizing greedy baseline and lower cost than the accuracy-priority IPA baseline, with decision times that grow far more slowly than IPA's as pipeline complexity increases.

Load-bearing premise

The argument rests on assuming the QoS score $Q=\alpha V+\beta T-L-\gamma E$ (with unstated weights, and $V$ computed as the sum of per-model accuracies) faithfully reflects real end-to-end service quality; if that score is not a faithful proxy, the reported improvements are improvements only against an arbitrary objective.

Editorial extensions

If this is right

  • Edge operators could automatically adapt model variants, replica counts, and batch sizes to changing workloads without manual tuning.
  • Decision-making time stays nearly flat as the number of pipeline stages and model variants grows, unlike the solver-based IPA baseline, so the approach is positioned for larger pipelines.
  • The same controller can balance cost and QoS under steady low load, fluctuating load, and steady high load, which is the regime edge deployments typically face.
  • The combination of LSTM load prediction and policy-gradient decision-making could be reused for other sequential resource-allocation problems at the edge.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper, the practical value of OPD hinges on how the QoS weights $\alpha,\beta,\gamma,\delta$ are chosen; a natural extension is to learn these weights from user-level service objectives rather than fixing them.
  • The paper sums per-model accuracies to get pipeline accuracy (Eq. 1), but real end-to-end accuracy often behaves more like a product or a stage-dependent composition; testing OPD with a product-style accuracy model would show whether the method's advantage survives.
  • Because training is warm-started by an expert model, an ablation that varies the expert's quality and frequency would reveal how much of the reported performance comes from the expert guidance rather than the policy-gradient learning itself.
  • The same algorithm could be adapted to GPU allocation, which the paper names as future work, and to heterogeneous edge clusters where node capabilities differ across devices.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. This paper considers online configuration selection for multi-model inference pipelines deployed at the edge. The authors model the problem as a constrained MDP whose objective is to maximize a hand-defined QoS measure minus cost, design an LSTM workload predictor and a residual-network feature extractor, and propose OPD, a PPO-based policy-gradient algorithm with expert-guided warm-start training. The evaluation is carried out in a real three-node Kubernetes cluster and compares OPD with Random, Greedy, and IPA baselines under steady and fluctuating workloads, reporting QoS, cost, and decision-time figures. The main claimed contribution is an adaptive policy that balances QoS and cost and outperforms all baselines.

Significance. If the evaluation were valid, the paper would provide a practically relevant demonstration of RL-based configuration control for edge inference pipelines, including a real Kubernetes deployment, a workload predictor, and a complexity analysis. The strongest assets are the concrete system implementation and the attempt to handle resource constraints that many prior autoscaling works ignore. However, the validity of the evaluation depends on the accuracy-aggregation definition in Eq. (1), the unreported weighting parameters in Eqs. (3), (4), and (7), and the consistency of the claims with the reported tradeoffs; each of these is problematic. The result is that the central 'outperforms all baselines' claim is not established.

major comments (4)
  1. [Section III.B, Eq. (1)] The end-to-end accuracy of a linear multi-model pipeline is defined in Eq. (1) as a sum of per-stage model accuracies. For a chain of models, end-to-end accuracy is the product of per-stage accuracies under the usual independence assumption; the proposed sum can exceed 1 and increases with the number of stages even when every stage is imperfect, so it is not a valid accuracy measure and is not monotonically related to true end-to-end accuracy across pipelines of different lengths. Because V enters the QoS objective in Eq. (3), the optimization objective in Eq. (4), and the reward in Eq. (7), all experimental QoS comparisons in Section VI.B are anchored to an invalid accuracy proxy. The manuscript states that the definition is inspired by [13], but it does not show that [13] uses a sum, and no external validation of Eq. (1) against labeled end-to-end accuracy is provided.
  2. [Section III.B, Eq. (3), and Section IV.B, Eq. (7)] The QoS metric Q and the reward r_t contain weighting parameters α, β, γ, δ (and the objective in Eq. (4) contains λ), but none of these values is reported anywhere, so the optimization target and the experimental comparison are not reproducible. In addition, β and γ denote different quantities in Eq. (3) (throughput weight and excess-load penalty) and in Eq. (7) (cost weight and batch-size penalty), which makes the definitions inconsistent. The authors should disclose all weights and justify their choices.
  3. [Section VI.B] The conclusion that OPD 'outperforms all baseline algorithms' is not supported by the paper's own numbers. Under steady low load, OPD's cost is 120% higher than Greedy while improving QoS by 36%, and relative to IPA it reduces cost by 16% at a 3.8% QoS loss; under fluctuating load, OPD costs 37% more than Greedy and has 3% lower QoS than IPA. These results position OPD on a cost-QoS tradeoff frontier between Greedy and IPA, not as a dominant policy. The claim should be either withdrawn or replaced by a properly framed multi-criteria comparison.
  4. [Section VI.B, Fig. 6] The decision-time evaluation compares OPD only with IPA, yet the abstract and conclusion claim that the approach 'significantly' shortens decision-making time and that OPD outperforms all baseline algorithms. No decision-time data for the Random or Greedy baselines are reported, so the sweeping decision-time claim is unsupported. At minimum, the manuscript should restrict the claim to the IPA comparison.
minor comments (5)
  1. [Section VI.A] The citation for SMAPE points to reference [14] (He et al., ResNet); please cite a proper SMAPE source.
  2. [Section IV.C, Eq. (8)] Eq. (8) uses γ for a trajectory while γ is already a weight in Eqs. (3) and (7); rename one of the two uses.
  3. [Section III.C, Eq. (5)] The node features (u_t, p_t, m_t) are repeated verbatim in every task component of the state vector in Eq. (5); factoring them out would clarify the state space.
  4. [Section VI.B, Figs. 4-5] The cost and QoS figures appear to show a single workload cycle, and no error bars or repeated-run statistics are reported; given the fixed-seed statement, the reader cannot assess variance.
  5. [Abstract] The abstract contains grammatical errors ('significantly improve QoS while reducing costs and shorten decision-making time'); please proofread the text.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the optimization objective is author-defined, but OPD is benchmarked against baselines on the same metric, and no load-bearing step reduces to its own input.

full rationale

The paper's derivation chain is self-contained rather than circular. The QoS definition in Eq. (3) and the reward in Eq. (7) are author-constructed, but OPD does not derive its reported QoS from that definition; it is trained with PPO on a reward that includes Q, and Section VI.B compares OPD against Random, Greedy, and IPA using the same Q and C for all algorithms. Such an evaluation measures whether the learned policy optimizes the declared objective, which is a normal RL evaluation, not a circular prediction. Eq. (1)'s sum-of-accuracies proxy is inherited from external work [13] and is a heuristic; while it raises correctness and reproducibility concerns (especially since the weights in Eq. (3) are unreported), it is not defined in terms of OPD's decisions and does not make the comparison an identity. The LSTM workload predictor is trained on actual load series and evaluated by SMAPE, so it is not a fitted parameter renamed as a prediction. Self-citations (e.g., [7], [27], [29]) appear only in related-work context and are not load-bearing for the central claim. No uniqueness theorem is imported from the authors' prior work, and no ansatz is smuggled in via self-citation. The central claim is therefore not forced by construction; concerns about the accuracy proxy and unreported weights belong to correctness and validity, not circularity.

Assumptions & free parameters 5 free parameters · 3 assumptions · 0 invented entities

The central optimization relies on several hand-defined weights and ad hoc metric definitions. The QoS, reward, and objective functions are not derived from first principles and their parameters are not reported. The accuracy approximation by summation is a domain assumption with no validation. No new physical or conceptual entities are introduced.

free parameters (5)
  • α (QoS accuracy weight) = not reported
    Weight on accuracy in Eq. (3); value never stated, directly controls what QoS means.
  • β (QoS throughput weight / reward cost weight) = not reported
    β appears in Eq. (3) as throughput weight and in Eq. (7) as cost weight; no value given.
  • γ (QoS excess-load penalty / reward batch penalty) = not reported
    γ appears in Eqs. (3) and (7) with different roles; no value given.
  • δ (QoS spare-capacity weight) = not reported
    Weight on unused capacity in Eq. (3); no value given.
  • λ (objective cost weight) = not reported
    Weight balancing QoS and cost in Eq. (4); never specified, so the reported 'best' balances are not reproducible.
assumptions (3)
  • ad hoc to paper End-to-end pipeline accuracy is the sum of per-model accuracies (Eq. 1)
    The paper states this is a heuristic approach inspired by [13], but summing accuracies over pipeline stages does not generally equal end-to-end accuracy; no justification is given.
  • ad hoc to paper The QoS function Q in Eq. (3) with user-chosen weights α, β, γ, δ captures true service quality
    No empirical validation or user study supports this weighted sum of accuracy, throughput, latency, and excess load; the weights are unreported.
  • domain assumption Workloads over the next 20 seconds can be predicted from the past 2 minutes with sufficient accuracy (SMAPE 6%) for configuration decisions
    The LSTM predictor is trained and evaluated on data from the test workload; no sensitivity analysis of prediction error on RL performance is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Adaptive Configuration Selection for Multi-Model Inference Pipelines in Edge Computing." pith.science (2026). https://pith.science/paper/G3BYNKUP

@misc{pith2026250602814,
  author       = {Pith},
  title        = {Pith review of: Adaptive Configuration Selection for Multi-Model Inference Pipelines in Edge Computing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/G3BYNKUP}},
  note         = {Machine review of arXiv:2506.02814}
}
read the original abstract

The growing demand for real-time processing tasks is driving the need for multi-model inference pipelines on edge devices. However, cost-effectively deploying these pipelines while optimizing Quality of Service (QoS) and costs poses significant challenges. Existing solutions often neglect device resource constraints, focusing mainly on inference accuracy and cost efficiency. To address this, we develop a framework for configuring multi-model inference pipelines. Specifically: 1) We model the decision-making problem by considering the pipeline's QoS, costs, and device resource limitations. 2) We create a feature extraction module using residual networks and a load prediction model based on Long Short-Term Memory (LSTM) to gather comprehensive node and pipeline status information. Then, we implement a Reinforcement Learning (RL) algorithm based on policy gradients for online configuration decisions. 3) Experiments conducted in a real Kubernetes cluster show that our approach significantly improve QoS while reducing costs and shorten decision-making time for complex pipelines compared to baseline algorithms.

Figures

Figures reproduced from arXiv: 2506.02814 by the authors.

Figure 1
Figure 1. System Overview. III. SYSTEM DESIGN AND PROBLEM FORMULATION A. System Design This section provides an overview of the system’s main components, as illustrated in [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Algorithm Overview. pipeline state records each task’s current model index, num￾ber of replicas, batch size, costs, latency, and throughput performance. Thus, the state st is defined as: st =[(ut, pt, mt, l1,t, t1,t, z1,t, f1,t, b1,t, c1,t), . . . , (ut, pt, mt, ln,t, tn,t, zn,t, fn,t, bn,t, cn,t)]. (5) Action: The action at encompasses the configuration for each pipeline task, including the model variant index, num… view at source ↗
Figure 3
Figure 3. Lstm Prediction. V. SYSTEM IMPLEMENTATION A multi-model inference pipeline is deployed on a Kuber￾netes cluster, a popular open-source platform for orchestrat￾ing containerized applications. A. Implementation Issues Network Communication: The system processes infer￾ence requests sequentially across multiple models, which can introduce latency. To mitigate this, it uses gRPC [32] for inter-stage communication, optimi… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Temporal analysis under different workloads. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 6
Figure 6. Figure 6: Different pipelines decision time. while OPD’s remains steady. As the number of tasks and model variants per stage increases, OPD’s decision-making efficiency significantly surpasses IPA, with improvements of 32.5%, 53.5%, 111.6%, and 212.8% in processing a single work…
Figure 7
Figure 7. Figure 7: Training Loss, Value Loss, and Reward of the OPD algorithm. [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

34 extracted references · 29 canonical work pages

  1. [13]

    [solution] ipa: Inference pipeline adaptation to achieve high accuracy and cost-efficiency,

    S. Ghafouri, K. Razavi, M. Salmani, A. Sanaee, T. L. Botran, L. Wang, J. Doyle, and P. Jamshidi, “[solution] ipa: Inference pipeline adaptation to achieve high accuracy and cost-efficiency,” Journal of Systems Research, vol. 4, no. 1, 2024

    work page 2024

  2. [1]

    Internet of things (iot): A literature review,

    S. Madakam, R. Ramaswamy, and S. Tripathi, “Internet of things (iot): A literature review,” Journal of Computer and Communications, vol. 3, no. 5, pp. 164–173, 2015

    work page 2015

  3. [2]

    Power-steering control architecture for automatic driving,

    J. E. Naranjo, C. Gonz ´alez, R. Garc ´ıa, T. de Pedro, and R. E. Haber, “Power-steering control architecture for automatic driving,” Ieee transactions on intelligent transportation systems , vol. 6, no. 4, pp. 406–415, 2005

    work page 2005

  4. [3]

    Healthcare 4.0,

    M. Wehde, “Healthcare 4.0,” IEEE Engineering Management Review, vol. 47, no. 3, pp. 24–28, 2019

    work page 2019

  5. [4]

    The characteristics of cloud computing,

    C. Gong, J. Liu, Q. Zhang, H. Chen, and Z. Gong, “The characteristics of cloud computing,” in 2010 39th International Conference on Parallel Processing Workshops, pp. 275–279, IEEE, 2010

    work page 2010

  6. [5]

    Edge computing: Vision and challenges,

    W. Shi, J. Cao, Q. Zhang, Y . Li, and L. Xu, “Edge computing: Vision and challenges,” IEEE internet of things journal , vol. 3, no. 5, pp. 637–646, 2016

    2016

  7. [6]

    Joint computing and caching in 5g-envisioned internet of vehicles: A deep reinforcement learning- based traffic control system,

    Z. Ning, K. Zhang, X. Wang, M. S. Obaidat, L. Guo, X. Hu, B. Hu, Y . Guo, B. Sadoun, and R. Y . Kwok, “Joint computing and caching in 5g-envisioned internet of vehicles: A deep reinforcement learning- based traffic control system,” IEEE Transactions on Intelligent Trans- portation Systems, vol. 22, no. 8, pp. 5201–5212, 2020

    work page 2020

  8. [7]

    Joint resource overbooking and container scheduling in edge computing,

    Z. Tang, F. Mou, J. Lou, W. Jia, Y . Wu, and W. Zhao, “Joint resource overbooking and container scheduling in edge computing,” IEEE Transactions on Mobile Computing , 2024

    work page 2024

Show all 34 references
  1. [8]

    A quality of service architecture,

    A. Campbell, G. Coulson, and D. Hutchison, “A quality of service architecture,” ACM SIGCOMM Computer Communication Review , vol. 24, no. 2, pp. 6–27, 1994

    1994

  2. [9]

    Rim: Offloading inference to the edge,

    Y . Hu, W. Pang, X. Liu, R. Ghosh, B. Ko, W.-H. Lee, and R. Govin- dan, “Rim: Offloading inference to the edge,” in Proceedings of the International Conference on Internet-of-Things Design and Implemen- tation, pp. 80–92, 2021

    2021

  3. [10]

    Inferline: latency-aware provisioning and scaling for prediction serving pipelines,

    D. Crankshaw, G.-E. Sela, X. Mo, C. Zumar, I. Stoica, J. Gonzalez, and A. Tumanov, “Inferline: latency-aware provisioning and scaling for prediction serving pipelines,” in Proceedings of the 11th ACM Symposium on Cloud Computing , pp. 477–491, 2020

    2020

  4. [11]

    Grandslam: Guaranteeing slas for jobs in microservices execution frameworks,

    R. S. Kannan, L. Subramanian, A. Raju, J. Ahn, J. Mars, and L. Tang, “Grandslam: Guaranteeing slas for jobs in microservices execution frameworks,” in Proceedings of the Fourteenth EuroSys Conference 2019, pp. 1–16, 2019

    2019

  5. [12]

    Fa2: Fast, accurate autoscaling for serving deep learning inference with sla guarantees,

    K. Razavi, M. Luthra, B. Koldehofe, M. M ¨uhlh¨auser, and L. Wang, “Fa2: Fast, accurate autoscaling for serving deep learning inference with sla guarantees,” in 2022 IEEE 28th Real-Time and Embedded Technology and Applications Symposium (RTAS), pp. 146–159, IEEE, 2022

    2022

  6. [14]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proceedings of the IEEE conference on computer vision and pattern recognition , pp. 770–778, 2016

    2016

  7. [15]

    Long short-term memory,

    S. Hochreiter, “Long short-term memory,” Neural Computation MIT- Press, 1997

    1997

  8. [16]

    Autopi- lot: workload autoscaling at google,

    K. Rzadca, P. Findeisen, J. Swiderski, P. Zych, P. Broniek, J. Kus- mierek, P. Nowak, B. Strack, P. Witusowski, S. Hand, et al., “Autopi- lot: workload autoscaling at google,” in Proceedings of the Fifteenth European Conference on Computer Systems , pp. 1–16, 2020

    2020

  9. [17]

    Reconciling high accuracy, cost-efficiency, and low latency of inference serving systems,

    M. Salmani, S. Ghafouri, A. Sanaee, K. Razavi, M. M ¨uhlh¨auser, J. Doyle, P. Jamshidi, and M. Sharifi, “Reconciling high accuracy, cost-efficiency, and low latency of inference serving systems,” in Proceedings of the 3rd Workshop on Machine Learning and Systems , pp. 78–86, 2023

    2023

  10. [18]

    Policy gradi- ent methods for reinforcement learning with function approximation,

    R. S. Sutton, D. McAllester, S. Singh, and Y . Mansour, “Policy gradi- ent methods for reinforcement learning with function approximation,” Advances in neural information processing systems , vol. 12, 1999

    1999

  11. [19]

    [Online]

    Kubernetes. [Online]. Available: https://kubernetes.io/docs/home/

  12. [20]

    Available: https://github.com/SeldonIO/seldon-core

    Seldon Core. Available: https://github.com/SeldonIO/seldon-core

  13. [21]

    Available: https://github.com/SeldonIO/MLServer

    MLServer. Available: https://github.com/SeldonIO/MLServer

  14. [22]

    [Online]

    Prometheus. [Online]. Available: https://prometheus.io

  15. [23]

    Aquatope: Qos-and- uncertainty-aware resource management for multi-stage serverless workflows,

    Z. Zhou, Y . Zhang, and C. Delimitrou, “Aquatope: Qos-and- uncertainty-aware resource management for multi-stage serverless workflows,” in Proceedings of the 28th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 1, p...

    2022

  16. [24]

    Cocktail: A multidimensional optimization for model serving in cloud,

    J. R. Gunasekaran, C. S. Mishra, P. Thinakaran, B. Sharma, M. T. Kan- demir, and C. R. Das, “Cocktail: A multidimensional optimization for model serving in cloud,” in 19th USENIX Symposium on Networked Systems Design and Implementation (NSDI 22), pp. 1041–1057, 2022

    2022

  17. [25]

    Spatio–temporal edge service placement: A bandit learning approach,

    L. Chen, J. Xu, S. Ren, and P. Zhou, “Spatio–temporal edge service placement: A bandit learning approach,” IEEE Transactions on Wire- less Communications, vol. 17, no. 12, pp. 8388–8401, 2018

    2018

  18. [26]

    Online service migration in mobile edge with incomplete system information: A deep recurrent actor-critic learning approach,

    J. Wang, J. Hu, G. Min, Q. Ni, and T. El-Ghazawi, “Online service migration in mobile edge with incomplete system information: A deep recurrent actor-critic learning approach,” IEEE Transactions on Mobile Computing, vol. 22, no. 11, pp. 6663–6675, 2022

    2022

  19. [27]

    Multi- user layer-aware online container migration in edge-assisted vehicular networks,

    Z. Tang, F. Mou, J. Lou, W. Jia, Y . Wu, and W. Zhao, “Multi- user layer-aware online container migration in edge-assisted vehicular networks,” IEEE/ACM Transactions on Networking , 2023

    2023

  20. [28]

    Dependent task offloading for edge computing based on deep rein- forcement learning,

    J. Wang, J. Hu, G. Min, W. Zhan, A. Y . Zomaya, and N. Georgalas, “Dependent task offloading for edge computing based on deep rein- forcement learning,” IEEE Transactions on Computers, vol. 71, no. 10, pp. 2449–2461, 2021

    2021

  21. [29]

    Latency-aware container scheduling in edge cluster upgrades: A deep reinforcement learning approach,

    H. Cui, Z. Tang, J. Lou, W. Jia, and W. Zhao, “Latency-aware container scheduling in edge cluster upgrades: A deep reinforcement learning approach,” IEEE Transactions on Services Computing, 2024

    2024

  22. [30]

    [Online]

    NVIDIA TensorRT. [Online]. Available: https://developer.nvidia.com/ tensorrt

  23. [31]

    A survey of quantization methods for efficient neural network inference,

    A. Gholami, S. Kim, Z. Dong, Z. Yao, M. W. Mahoney, and K. Keutzer, “A survey of quantization methods for efficient neural network inference,” in Low-Power Computer Vision , pp. 291–326, Chapman and Hall/CRC, 2022

    2022

  24. [32]

    [Online]

    GRPC. [Online]. Available: https://grpc.io/

  25. [33]

    [Online]

    Istio. [Online]. Available: https://istio.io/

  26. [34]

    Proximal policy optimization algorithms,

    J. Schulman, F. Wolski, P. Dhariwal, A. Radford, and O. Klimov, “Proximal policy optimization algorithms,” arXiv preprint arXiv:1707.06347, 2017

    2017 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.