Pith. sign in

REVIEW 2 major objections 4 minor 1 cited by

Priority-Aware Model-Distributed Inference at Edge Networks

T0 review · 2 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read A single priority weight per source, folded into each layer-offload decision, is enough to make model-distributed inference favor high-priority tasks without any central coordinator.

desk verdict A practical priority-aware MDI scheduler with real testbed results, but the evaluation confounds topology with priority and the optimization derivation is hand-wavy. read the letter →

arxiv 2412.12371 v1 pith:CS2TXW7Z submitted 2024-12-16 cs.DC cs.LG

Teng Li , Hulya Seferoglu This is my paper · ORCID
classification cs.DCcs.LG
keywords model-distributedinferenceedgecomputingpriority-awareschedulingtaskoffloadingmodelparallelismlatencyheterogeneousdeviceslargelanguagemodels
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

Model-distributed inference splits a neural network across edge devices so no single device holds the whole model. This paper asks how to allocate those devices when data sources have different importance, and it claims that a single priority weight per source is enough to steer the system. The proposed PA-MDI algorithm picks, for each layer partition, the neighbor that minimizes an estimated cost (communication plus queued plus compute time) divided by that source's priority weight. On real edge hardware and a public wireless testbed running ResNet-50, ResNet-56, and GPT-2, the paper reports up to 75% lower average inference time for time-sensitive sources than two earlier distributed-inference schemes, with all decisions made locally. The practical stakes are that a decentralized edge network can favor critical tasks without a central scheduler.

What carries the argument

The load-bearing object is the per-task offload ratio $\frac{d_{n,j}+\delta(T_k^m(d))+F(T_k^m(d))F_j+Q_j}{\gamma_m \alpha_m(d)}$, the estimated cost of sending, queueing behind, and computing a layer partition on neighbor j, normalized by the source's priority weight and accuracy contribution. It emerges from decomposing the Lagrangian of the weighted-accuracy-minus-delay objective, which turns a network-wide allocation problem into independent local decisions. In the ratio, $\gamma_m$ sits in the denominator, so higher-priority sources win favorable offload choices automatically, and the local rule with an RTC/CTC handshake replaces any central coordinator.

What would settle it

Run PA-MDI and MS-MDI on identical topology and identical partition counts with all source priorities equal; if PA-MDI still shows a large latency advantage, the measured improvement cannot be attributed to its priority weighting.

Watch

Extended reading notes

Core claim

PA-MDI models each source m with a priority γm and formulates model allocation as maximizing weighted inference accuracy minus a delay penalty. The optimization decomposes by task, so each layer partition should be sent to the worker j minimizing the ratio of estimated task delay to γm αm(d). Each worker runs this local ratio rule on neighbor status information alone, using a request-to-compute/clear-to-compute handshake to avoid several sources flooding one receiver. The paper reports that this scheme reduces the average inference time of the time-sensitive source by up to 75.3% against AR-MDI and 73.2% against MS-MDI in the first Jetson scenario, with comparable reductions in multi-hop and GPT-2 experiments. It also reports that giving the lower-priority model too many partitions creates congestion that erodes the priority advantage.

Load-bearing premise

The load-bearing premise is that the comparison is fair: PA-MDI runs on a fully connected mesh while AR-MDI and MS-MDI are restricted to circular or chain topologies, so the reported reductions could partly reflect extra connectivity rather than the priority weighting itself.

Editorial extensions

If this is right

  • High-priority sources can receive faster inference without any central scheduler, purely from each worker comparing neighbor costs.
  • The number of partitions for low-priority models matters: more low-priority tasks create congestion and weaken the priority advantage, as seen when PA-MDI (4,2) underperforms other splits.
  • The same ratio rule transfers across workloads and scales: Jetson mesh, heterogeneous multi-hop Jetsons, and GPT-2 on a larger testbed all show reduced time-sensitive inference time versus AR-MDI, MS-MDI, and local processing.
  • When the high-priority source's model is small, local processing can be the best policy, and PA-MDI's local decision rule is able to approach that outcome rather than forcing distribution.

Reading between the lines

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

  • A natural test the paper does not report: equalize all γm and run PA-MDI and MS-MDI on the same topology; if PA-MDI still wins, its advantage comes partly from its more flexible mesh-based operation rather than from priority weighting.
  • The ratio rule suggests a simple dynamic-priority extension: if γm changes over time (e.g., a detected event raises a camera's importance), the same local formula can track the change without re-solving the network-wide problem.
  • Because the offload rule is parameter-light, it could be combined with early-exit or layer-pruning methods to trade accuracy for latency without altering the decision structure.
  • The experiments use two sources; scaling to many sources with overlapping tasks would test whether the single-hop ratio remains near-optimal or needs congestion pricing across multiple hops.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 4 minor

Summary. The paper considers model-distributed inference (MDI) in an edge network with multiple data sources that have different priorities. It formulates an optimization problem that trades accuracy, weighted by source priority, against inference delay, and derives a decentralized routing rule in which each worker sends a task to the neighbor minimizing (delay plus queue plus compute time) divided by the source priority weight gamma_m and accuracy improvement alpha_m. This rule is implemented as PA-MDI, with an RTC/CTC handshake to avoid overloading a selected worker. The paper evaluates PA-MDI against AR-MDI, MS-MDI, and local processing on a five-node Jetson testbed, a heterogeneous six-node multi-hop Jetson testbed, and the Colosseum testbed with GPT-2, reporting reductions in average inference time for the time-sensitive source of up to about 75% in the Jetson experiments.

Significance. If the claimed gains are caused by the priority weighting, the paper would make a useful contribution: a single scalar priority per source is enough to steer decentralized model-parallel inference. The experiments are broad and the algorithm is not post-hoc fitted to the measurements; the evaluation uses real devices and multiple models, and the proposed mechanism is fully decentralized. However, the empirical separation between the priority mechanism and the greater connectivity available to PA-MDI is not established, and the analytic derivation connecting the objective to the deployed rule is incomplete. The core idea is plausible and the evidence is suggestive, but the current manuscript does not support the headline causal claim.

major comments (2)
  1. [V-A, V-B] The experimental comparison is confounded with topology flexibility. In Section V-A the testbed is described as a full mesh ('all devices are connected to each other, forming a mesh topology'), while AR-MDI and MS-MDI are explicitly restricted to a circular topology over the same five workers. In Section V-B the baselines are restricted to fixed chains [A, B, E, D, F, C] and [D, F, C, A, B, E] while PA-MDI can offload to any neighbor in the available topology. PA-MDI therefore has a strictly larger action space in every experiment, so the reported reductions (up to 75.3% in Fig. 3, and the corresponding reductions in Figs. 4, 5, 7, 8, 9, and 10) could be caused by the extra connectivity rather than by the priority-aware term in Eq. (8). A topology-controlled comparison, in which the baselines are given the same mesh connectivity or PA-MDI is constrained to the same circular/chain topologies, is needed to support the paper's central claim that prioritizing by gamma_m/alpha_m is what reduces high-priority inference time.
  2. [IV-B, Eqs. (4)-(7)] The derivation of the optimization solution is not rigorous. Equation (4) is described as a convex optimization problem, but the decision variable pi is a discrete assignment of tasks to workers, so standard convexity does not apply. The maximization over beta is also not well defined: with unconstrained beta and positive delays, the objective is either unbounded or collapses to beta = 0. The statement 'choosing the minimum Lagrange multiplier beta' is not defined. Furthermore, the reduction from Eq. (6) to Eq. (7) silently drops the product over success probabilities, product_{k <= K_m} (1 - P(pi^k_m(d))), even though that product is part of the stated objective in Eq. (1) and appears in the denominator of Eq. (6). Because the deployed rule (8) is derived from this reduction, the formal link between the optimization problem and the algorithm is not established. The authors should either provide a correct derivation or explicitly present PA-MDI as a heuristic motivated by Eq. (4) and validate it as such.
minor comments (4)
  1. [V-B, Fig. 8] The caption of Fig. 8 appears to duplicate the caption of Fig. 7: it says Worker A hosts the 'Non-Time-Sensitive' data source, while the body text says that in the second scenario Worker A hosts the 'Time-Sensitive' data source. Please correct the caption or the text so they agree.
  2. [V-C, Eq. (8)] The notation in Eq. (8) is ambiguous: the term written as F(T^k_m(d)) F_j should presumably be F(T^k_m(d)) / F_j, since F_j is defined as computing performance and the expression is supposed to represent computation delay. Please clarify the units and the intended arithmetic.
  3. [IV-B] The term 'Lagrange multiplier' for beta is used without any corresponding constraint; beta appears to be a trade-off or regularization parameter. Using a different name would avoid the impression of a missing constraint.
  4. [V] The reported average inference times in Figs. 3-5 and 7-10 are given without error bars or a statement of the number of trials per configuration. Adding this information would make the claimed reductions more convincing.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: PA-MDI's priority weighting is explicitly inserted into the optimization objective (Eqs. 4-8) and then evaluated on real testbeds; the cited prior baselines are used as competitors, not as justification, and the main concerns (unproven convexity, topology mismatch) are correctness risks rather than circular derivation.

full rationale

The derivation chain is self-contained. The paper defines total accuracy I(pi) with source weights gamma_m (Eq. 2) and minimizes delay Delta(pi) (Eq. 3), forming J(pi) = I(pi) - beta*Delta(pi) (Eq. 4). It decomposes this into a per-task ratio rho(pi)/(gamma_m alpha_m(d)) (Eqs. 6-7), and PA-MDI's offloading rule (Eq. 8) is explicitly the distributed implementation of that ratio, with the numerator d_nj + delta + F(T)/F_j + Q_j identified as rho(pi). No free parameter is fitted to the experimental outcomes; gamma_m is set qualitatively to 'a large positive number' for time-sensitive sources, and alpha_m(d) are set equal. The empirical claim that high-priority sources finish faster is therefore a test of whether this derived rule works in a real system, not a tautology. The baselines AR-MDI [1] and MS-MDI [2] are prior publications by the same group, but they are used as external competing algorithms and as related work, not as the justification for PA-MDI's design; no uniqueness theorem or ansatz is imported from these citations. The paper does assert without proof that Eq. (4) is convex and decomposable, and the experimental comparison gives PA-MDI a full-mesh topology while restricting the baselines to circular/chain topologies; both are substantive validity concerns, but they are not cases where a result is equivalent to its inputs by construction. Hence no circular step is identified.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central claim depends on a small number of hand-chosen parameters (priority weights, accuracy weights, partition counts), on an unproven decomposition of the optimization problem, and on the assumption that failure probability can be ignored in the scheduler. No new physical or mathematical entities are introduced.

free parameters (4)
  • γm priority weight = unspecified (quoted as 'large positive number')
    The priority of each source is set by hand; the exact value is not reported in Section V-A.1, yet it directly determines task selection in Eq. (8). Without the value, the experiment cannot be reproduced and the priority effect cannot be quantified.
  • αm(d) accuracy improvement = assumed equal across sources and data points
    Section V-A.1 states 'We consider the accuracy parameters αm(d) equal to each other for different data sources and data points', removing accuracy from the objective and reducing the claim to inference time only.
  • β Lagrange multiplier = not specified; 'minimum' is not defined
    Used in Eq. (4) to combine accuracy and delay, but the proposed solution 'choosing the minimum Lagrange multiplier' is not a well-defined procedure and the algorithm never uses β.
  • partition counts (μ, η) = selected from {2,4} in experiments
    The number of partitions per model is an experiment choice (PA-MDI (μ,η)). The paper observes that partition count affects performance, e.g., PA-MDI (4,2) hurts priority, but provides no selection rule.
assumptions (4)
  • ad hoc to paper Eq. (4) is a convex optimization problem and is decomposable per source and data point.
    Stated in Section IV-B without proof; the decision variables are discrete worker assignments, so convexity is not obvious and the decomposition is what makes the greedy algorithm appear optimal.
  • domain assumption A new task is generated only after the previous task succeeds, so the failure-probability product in Eq. (1) cancels from the per-task decision.
    In Section IV-B, the reduction from Eq. (6) to Eq. (7) drops P(π^k_m(d)); the scheduler therefore ignores task failure even though the stated objective in Eq. (1) includes it.
  • domain assumption All workers can execute any partition of any model and communicate over an ad-hoc topology.
    Section III assumes workers are capable of processing assigned tasks and are connected via WiFi in ad-hoc mode; the algorithm depends on one-hop status exchange and multi-hop offloading.
  • ad hoc to paper The RTC/CTC handshake avoids overloading a selected worker without significant latency overhead.
    Section IV-C adds this mechanism inspired by CSMA/CA but does not model its delay cost in the objective or measure its overhead separately.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Priority-Aware Model-Distributed Inference at Edge Networks." pith.science (2026). https://pith.science/paper/CS2TXW7Z

@misc{pith2026241212371,
  author       = {Pith},
  title        = {Pith review of: Priority-Aware Model-Distributed Inference at Edge Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CS2TXW7Z}},
  note         = {Machine review of arXiv:2412.12371}
}
read the original abstract

Distributed inference techniques can be broadly classified into data-distributed and model-distributed schemes. In data-distributed inference (DDI), each worker carries the entire Machine Learning (ML) model but processes only a subset of the data. However, feeding the data to workers results in high communication costs, especially when the data is large. An emerging paradigm is model-distributed inference (MDI), where each worker carries only a subset of ML layers. In MDI, a source device that has data processes a few layers of ML model and sends the output to a neighboring device, i.e., offloads the rest of the layers. This process ends when all layers are processed in a distributed manner. In this paper, we investigate the design and development of MDI when multiple data sources co-exist. We consider that each data source has a different importance and, hence, a priority. We formulate and solve a priority-aware model allocation optimization problem. Based on the structure of the optimal solution, we design a practical Priority-Aware Model- Distributed Inference (PA-MDI) algorithm that determines model allocation and distribution over devices by taking into account the priorities of different sources. Experiments were conducted on a real-life testbed of NVIDIA Jetson Xavier and Nano edge devices as well as in the Colosseum testbed with ResNet-50, ResNet- 56, and GPT-2 models. The experimental results show that PA-MDI performs priority-aware model allocation successfully while reducing the inference time as compared to baselines.

Figures

Figures reproduced from arXiv: 2412.12371 by the authors.

Figure 1
Figure 1. Model-distributed inference and model parallelism. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Model-distributed inference for multiple data sources. One of the [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Worker A, who hosts “Non-Time-Sensitive” data, has dataset CIFAR￾10 (224x224) which is processed by ResNet-50, while Worker D, who hosts “Time-Sensitive” data, has dataset CIFAR-10 (32x32) which is processed by ResNet-56 [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Worker A has a small data set (CIFAR-10 (32x32), which is processed by distributed ResNet-56) while Worker D has a larger dataset (CIFAR-10 (224x224), which is processed by distributed ResNet-50). Worker A hosts “Non-Time-Sensitive” data, while Worker D hosts “Time-Sen…
Figure 5
Figure 5. Figure 5: Both Workers A and D have larger dataset (CIFAR-10 (224x224), which is processed by ResNet-50). Similar to the previous scenarios, Worker A hosts “Non-Time-Sensitive” data, while Worker D hosts “Time-Sensitive” data. 2) Results: We evaluate the performance of PA-MDI fo…
Figure 6
Figure 6. Figure 6: Multi-hop topology. B. Heterogeneous and Multi-Hop Setup with NVIDIA Jetsons 1) Setup: We created a multi-hop topology of heteroge￾neous NVIDIA Jetsons. The topology shown in [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 8
Figure 8. Figure 8: Average inference time for the multi-hop topology in Fig. 6. Worker [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 10
Figure 10. Figure 10: Worker A has a smaller input size (Batch Size = 12) and D have larger (Batch Size = 16). Worker A hosts “Non-Time-Sensitive” data, while Worker D hosts “Time-Sensitive” data. In the second scenario, PA-MDI demonstrates a significant advantage over all baselines in the…
Figure 9
Figure 9. Figure 9: Workers A has a larger input size (Batch Size = 16) and D have smaller (Batch Size = 12). Worker A hosts “Non-Time-Sensitive” source, while Worker D hosts “Time-Sensitive” source. The experimental results are similar to those observations in the first scenario of the J…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Model-Distributed Inference for Large Language Models at the Edge

    cs.LG 2025-05 conditional novelty 5.0 of 10

    Model-distributed inference with recurrent pipeline parallelism lets edge devices share an LLM, cutting per-device memory and increasing token generation throughput when more devices join.

Reference graph

Works this paper leans on

38 extracted references · 28 canonical work pages · cited by 1 Pith paper

  1. [1]

    Adaptive and resilient model- distributed inference in edge computing systems,

    P. Li, E. Koyuncu, and H. Seferoglu, “Adaptive and resilient model- distributed inference in edge computing systems,” IEEE Open Journal of the Communications Society , 2023

    work page 2023

  2. [2]

    Model-distributed inference in multi-source edge networks,

    P. Li, H. Seferoglu, and E. Koyuncu, “Model-distributed inference in multi-source edge networks,” in 2023 IEEE International Conference on Acoustics, Speech, and Signal Processing Workshops (ICASSPW) . IEEE, 2023, pp. 1–5

    2023

  3. [3]

    Nvidia jetson xavier

    “Nvidia jetson xavier.” [Online]. Available: https://www.nvidia.com/ en-us/autonomous-machines/embedded-systems/jetson-xavier-series/

    work page

  4. [4]

    Nvidia jetson nano

    “Nvidia jetson nano.” [Online]. Available: https: //www.nvidia.com/en-us/autonomous-machines/embedded-systems/ jetson-nano/product-development/

    work page

  5. [5]

    Colosseum testbed,

    “Colosseum testbed,” 2024, available at https://www.northeastern.edu/colosseum/

    work page 2024

  6. [6]

    Gpipe: Efficient training of giant neural networks using pipeline parallelism,

    Y . Huang, Y . Cheng, A. Bapna, O. Firat, D. Chen, M. Chen, H. Lee, J. Ngiam, Q. V . Le, Y . Wu et al. , “Gpipe: Efficient training of giant neural networks using pipeline parallelism,” in Advances in Neural Information Processing Systems , 2019, pp. 103–112

    work page 2019

  7. [7]

    Pipedream: Fast and efficient pipeline parallel dnn training,

    A. Harlap, D. Narayanan, A. Phanishayee, V . Seshadri, N. Devanur, G. Ganger, and P. Gibbons, “Pipedream: Fast and efficient pipeline parallel dnn training,” arXiv preprint arXiv:1806.03377 , 2018

    arXiv 2018

  8. [8]

    Improving ml applications in shared computing environ- ments,

    A. Harlap, “Improving ml applications in shared computing environ- ments,” Ph.D. dissertation, Carnegie Mellon University, 2019

    work page 2019

Show all 38 references
  1. [9]

    Efficient and robust parallel dnn training through model parallelism on multi-gpu platform,

    C.-C. Chen, C.-L. Yang, and H.-Y . Cheng, “Efficient and robust parallel dnn training through model parallelism on multi-gpu platform,” arXiv preprint arXiv:1809.02839, 2018

    2018 arXiv

  2. [10]

    Respipe: Resilient model- distributed dnn training at edge networks,

    P. Li, E. Koyuncu, and H. Seferoglu, “Respipe: Resilient model- distributed dnn training at edge networks,” in IEEE ICASSP, 2021

    2021

  3. [11]

    Model-distributed dnn training for memory-constrained edge computing devices,

    P. Li, H. Seferoglu, V . R. Dasari, and E. Koyuncu, “Model-distributed dnn training for memory-constrained edge computing devices,” in 2021 IEEE International Symposium on Local and Metropolitan Area Net- works (LANMAN). IEEE, 2021, pp. 1–6

    2021

  4. [12]

    Distributed deep neural networks over the cloud, the edge and end devices,

    S. Teerapittayanon, B. McDanel, and H.-T. Kung, “Distributed deep neural networks over the cloud, the edge and end devices,” in 2017 IEEE 37th international conference on distributed computing systems (ICDCS). IEEE, 2017, pp. 328–339

    2017

  5. [13]

    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,” ACM SIGARCH Computer Architecture News, vol. 45, no. 1, pp. 615–629, 2017

    2017

  6. [14]

    Jointdnn: an efficient training and inference engine for intelligent mobile cloud computing services,

    A. E. Eshratifar, M. S. Abrishami, and M. Pedram, “Jointdnn: an efficient training and inference engine for intelligent mobile cloud computing services,” IEEE Transactions on Mobile Computing , vol. 20, no. 2, pp. 565–576, 2019

    2019

  7. [15]

    Pipeline parallelism for inference on heterogeneous edge computing,

    Y . Hu, C. Imes, X. Zhao, S. Kundu, P. A. Beerel, S. P. Crago, and J. P. N. Walters, “Pipeline parallelism for inference on heterogeneous edge computing,” arXiv preprint arXiv:2110.14895 , 2021

    2021 arXiv

  8. [16]

    Adaptive neural networks for efficient inference,

    T. Bolukbasi, J. Wang, O. Dekel, and V . Saligrama, “Adaptive neural networks for efficient inference,” in International Conference on Ma- chine Learning. PMLR, 2017, pp. 527–536

    2017

  9. [17]

    Branchynet: Fast inference via early exiting from deep neural networks,

    S. Teerapittayanon, B. McDanel, and H.-T. Kung, “Branchynet: Fast inference via early exiting from deep neural networks,” in 2016 23rd International Conference on Pattern Recognition (ICPR) . IEEE, 2016, pp. 2464–2469

    2016

  10. [18]

    Shallow-deep networks: Under- standing and mitigating network overthinking,

    Y . Kaya, S. Hong, and T. Dumitras, “Shallow-deep networks: Under- standing and mitigating network overthinking,” in International Confer- ence on Machine Learning . PMLR, 2019, pp. 3301–3310

    2019

  11. [19]

    Early-exit meets model- distributed inference at edge networks,

    M. Colocrese, E. Koyuncu, and H. Seferoglu, “Early-exit meets model- distributed inference at edge networks,” in IEEE LANMAN, Jul. 2024

    2024

  12. [20]

    Dynamic neural networks: A survey,

    Y . Han, G. Huang, S. Song, L. Yang, H. Wang, and Y . Wang, “Dynamic neural networks: A survey,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021

    2021

  13. [21]

    Pruning algorithms to accelerate convolutional neural networks for edge applications: A survey,

    J. Liu, S. Tripathi, U. Kurup, and M. Shah, “Pruning algorithms to accelerate convolutional neural networks for edge applications: A survey,” arXiv preprint arXiv:2005.04275 , 2020

    2005 arXiv

  14. [22]

    Pruning early exit networks,

    A. Gormez and E. Koyuncu, “Pruning early exit networks,” in Workshop on Sparsity in Neural Networks , Jul. 2022

    2022

  15. [23]

    Compress- ing neural networks with the hashing trick,

    W. Chen, J. Wilson, S. Tyree, K. Weinberger, and Y . Chen, “Compress- ing neural networks with the hashing trick,” in International conference on machine learning . PMLR, 2015, pp. 2285–2294

    2015

  16. [24]

    Compressing deep convolutional networks using vector quantization,

    Y . Gong, L. Liu, M. Yang, and L. Bourdev, “Compressing deep convolutional networks using vector quantization,” arXiv preprint arXiv:1412.6115, 2014

    2014 arXiv

  17. [25]

    Deep compression: Compressing deep neural networks with pruning, trained quantization and huffman coding,

    S. Han, H. Mao, and W. J. Dally, “Deep compression: Compressing deep neural networks with pruning, trained quantization and huffman coding,” arXiv preprint arXiv:1510.00149 , 2015

    2015 arXiv

  18. [26]

    Deep learning with limited numerical precision,

    S. Gupta, A. Agrawal, K. Gopalakrishnan, and P. Narayanan, “Deep learning with limited numerical precision,” in International conference on machine learning . PMLR, 2015, pp. 1737–1746

    2015

  19. [27]

    Training deep neu- ral networks with low precision multiplications,

    M. Courbariaux, Y . Bengio, and J.-P. David, “Training deep neu- ral networks with low precision multiplications,” arXiv preprint arXiv:1412.7024, 2014

    2014 arXiv

  20. [28]

    Gradient sparsification for communication-efficient distributed optimization,

    J. Wangni, J. Wang, J. Liu, and T. Zhang, “Gradient sparsification for communication-efficient distributed optimization,” in Advances in Neural Information Processing Systems , 2018, pp. 1299–1309

    2018

  21. [29]

    The convergence of sparsified gradient methods,

    D. Alistarh, T. Hoefler, M. Johansson, N. Konstantinov, S. Khirirat, and C. Renggli, “The convergence of sparsified gradient methods,” in Advances in Neural Information Processing Systems , 2018, pp. 5973– 5983

    2018

  22. [30]

    Joint device-edge infer- ence over wireless links with pruning,

    M. Jankowski, D. G ¨und¨uz, and K. Mikolajczyk, “Joint device-edge infer- ence over wireless links with pruning,” in 2020 IEEE 21st International Workshop on Signal Processing Advances in Wireless Communications (SPAWC). IEEE, 2020, pp. 1–5

    2020

  23. [31]

    Improving device-edge cooperative inference of deep learning via 2-step pruning,

    W. Shi, Y . Hou, S. Zhou, Z. Niu, Y . Zhang, and L. Geng, “Improving device-edge cooperative inference of deep learning via 2-step pruning,” in IEEE INFOCOM WKSHPS , 2019, pp. 1–6

    2019

  24. [32]

    Performance evaluation and enhancement of the csma/ca mac protocol for 802.11 wireless lans,

    G. Bianchi, L. Fratta, and M. Oliveri, “Performance evaluation and enhancement of the csma/ca mac protocol for 802.11 wireless lans,” in Proceedings of PIMRC ’96 - 7th International Symposium on Personal, Indoor, and Mobile Communications , vol. 2, 1996, pp. 392–396 vol.2

    1996

  25. [33]

    Cifar dataset

    “Cifar dataset.” [Online]. Available: https://www.cs.toronto.edu/ ∼kriz/ cifar.html

  26. [34]

    Resnet-50

    “Resnet-50.” [Online]. Available: https://huggingface.co/microsoft/ resnet-50

  27. [35]

    Resnet-56

    “Resnet-56.” [Online]. Available: https://www.signalpop.com/tutorial resnet56 on cifar10/

  28. [36]

    Language models are unsupervised multitask learners,

    A. Radford, J. Wu, R. Child, D. Luan, D. Amodei, I. Sutskever et al., “Language models are unsupervised multitask learners,” OpenAI blog , vol. 1, no. 8, p. 9, 2019

    2019

  29. [37]

    Colos- seum: Large-scale wireless experimentation through hardware-in-the- loop network emulation,

    L. Bonati, P. Johari, M. Polese, S. D’Oro, S. Mohanti, M. Tehrani- Moayyed, D. Villa, S. Shrivastava, C. Tassie, K. Yoder et al. , “Colos- seum: Large-scale wireless experimentation through hardware-in-the- loop network emulation,” in 2021 IEEE International Symposium on Dynam...

    2021

  30. [38]

    OpenAI GPT2,

    Hugging Face, “OpenAI GPT2,” accessed on Aug 29, 2024. [Online]. Available: https://huggingface.co/docs/transformers/en/model doc/gpt2

    2024

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.