Pith. sign in

REVIEW 3 major objections 6 minor 51 references

NestQuant: Post-Training Integer-Nesting Quantization for On-Device DNN

T0 review · 3 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read One quantized model stores two bitwidths and switches by paging lower bits in or out.

desk verdict A genuinely new post-training bit-nesting method with broad, solid ImageNet accuracy evidence; the headline switching-overhead saving is file-size arithmetic, not measured latency, so read it as a data-volume gain. read the letter →

arxiv 2506.17870 v1 pith:MORYV3Y4 submitted 2025-06-22 cs.LG cs.AIcs.DC

Jianhang Xie , Chuntao Ding , Xiaqing Li , Shenyuan Ren , Yidong Li , Zhichao Lu This is my paper · ORCID
classification cs.LGcs.AIcs.DC
keywords post-trainingquantizationintegerweightdecompositionmodelswitchingbitwidthnestingon-devicedeeplearningresourceadaptationHessian-basedroundingpacked-bittensors
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

NestQuant claims that a single quantized neural network file can serve two models at once: a full-bit model that keeps nearly full accuracy, and a lower-bit part-bit model extracted by dropping the low-order bits of the integer weights. Because the higher-bit weights closely resemble the original weights, they can be optimized with the same Hessian-based adaptive rounding used in post-training quantization, and the lower bits can later be paged in or out to switch between the two modes. On ImageNet-1K, the paper reports that ResNet-101 with INT8 nesting INT6 reaches 78.1% and 77.9% top-1 accuracy for the full-bit and part-bit models, while cutting numerically computed switching overhead by about 78.1% compared with storing and swapping two separate quantized models. If the method holds, on-device services could adapt to fluctuating power and memory without retraining, special hardware, or storing multiple models.

What carries the argument

The load-bearing object is the integer weight decomposition $w_{\mathrm{int}} = w_{\mathrm{high}} \cdot 2^l + w_{\mathrm{low}}$, obtained by right and left bit shifts on signed integer tensors. It turns a single quantized weight tensor into two tensors that can be stored or loaded independently, and because $w_{\mathrm{low}}/2^l$ acts as a known quantization perturbation of $w_{\mathrm{int}}$, the higher part can be optimized by the same Hessian-based adaptive rounding used for the original weights. The auxiliary pieces are the one-bit compensation range for $w_{\mathrm{low}}$, which makes recomposition exact, and packed-bit tensors, which store sub-8-bit values inside INT64 containers because current mobile inference libraries lack native sub-8-bit types.

What would settle it

Run NestQuant's INT8 nesting at the predicted critical bit on a model whose higher-bit and full-bit weights have near-zero correlation (for example, a randomly initialized network), and check whether part-bit accuracy collapses; alternatively, measure actual page-in/out wall-clock latency on a Raspberry Pi 4B to see whether the real switching-speed reduction matches the reported 78.1% disk-size-based figure.

Watch

Extended reading notes

Core claim

The paper's central claim is that the high-order bits of an already-quantized integer weight tensor form a usable lower-bit weight model, so one stored model can switch between two precisions. The identity is $w_{\mathrm{int}} = w_{\mathrm{high}} \cdot 2^l + w_{\mathrm{low}}$: an $n$-bit integer weight is split by bit-shifting into an $h$-bit higher part and an $l$-bit lower part, with $n = h + l$. The higher part is statistically similar to the full weight, so the same Hessian-based adaptive rounding that quantizes the original model can be applied to the higher part. An extra one-bit range is added to the lower part so that recomposing the two parts reproduces the full-bit weights without numerical error, and the effective nested bit $h$ follows a model-size rule of thumb, with a performance cliff below the critical combination. In deployment, paging out the lower-bit weights downgrades the full-bit model to the part-bit model, and paging them back in upgrades it, with full-bit accuracy unchanged and part-bit accuracy only slightly lower.

Load-bearing premise

The optimizer that rounds the full-bit quantized weights is assumed to work just as well on the extracted higher-bit weights, even though the error distribution of those higher-bit weights is different.

Editorial extensions

If this is right

  • Storing one NestQuant file replaces storing two separate bitwidth models, with measured size reductions close to ideal bitwidth arithmetic: roughly 22-40% for INT8 nesting and 32-37% for INT6 nesting across the tested ResNet and lightweight models.
  • Upgrading from part-bit to full-bit costs only the page-in of the lower-bit weights, and downgrading costs only their page-out, which is why the reported switching-overhead reduction reaches 78.1% for INT8 nesting INT6 and up to 86.6% for wider nesting gaps.
  • Full-bit accuracy after recomposition is preserved, and the part-bit model stays within a small margin of the full-bit model, as in ResNet-101 with INT8 nesting INT6 (78.1% versus 77.9%).
  • The critical nested bit is model-size dependent: roughly $h = n/2 + 1$ for models under 30 MB, $h = n/2$ between 30 and 300 MB, and $h = n/2 - 1$ above 300 MB, so the method comes with a selection rule rather than a single universal setting.
  • Because the whole procedure is post-training and data-free, it avoids retraining and special hardware, which is the practical advantage over dynamic-precision and mixed-precision approaches.

Reading between the lines

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

  • If the model-size rule of thumb generalizes, the 30 MB and 300 MB cutoffs predict critical nesting bits for untested architectures; a quick check would be to run INT8 nesting on a mid-size CNN around 100 MB and see whether the cliff sits at $h = 4$ as predicted.
  • The paper computes switching overhead from disk size rather than measured page-in/out latency, so an obvious next experiment is wall-clock switching time on real IoT hardware to see whether the 78.1% reduction transfers to practice.
  • If mobile inference libraries ever add native 1-7 bit data types, NestQuant would no longer need the packed-bit INT64 compromise, and the same nesting idea could extend to activations or to much larger on-device models.
  • The sharp performance cliff at the critical nested bit suggests a possible diagnostic use: the critical bit of a model could serve as a cheap proxy for how sensitive that model is to weight perturbation.
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

3 major / 6 minor

Summary. The paper proposes NestQuant, a post-training quantization scheme that decomposes an n-bit integer weight into an h-bit higher-bit weight w_high and an l-bit lower-bit weight w_low, stores both components, and switches between a full-bit model (recomposed n-bit weights) and a part-bit model (h-bit weights) by paging w_low in or out. The higher-bit weights are optimized with Hessian-based adaptive rounding using SQuant, and an extra 1-bit range is used for w_low so that the recomposed full-bit model is exactly the original n-bit quantized model. The paper evaluates top-1 accuracy on ImageNet-1K across ResNet, DenseNet, ResNeXt, lightweight CNNs, and ViTs, and reports reductions in network traffic, storage size, and switching overhead relative to storing and switching between diverse-bitwidth PTQ models. The headline result is ResNet-101 with INT8 nesting INT6 achieving 78.1% and 77.9% accuracy for full-bit and part-bit models and a roughly 78.1% reduction in switching overhead compared with storing separate INT8 and INT6 models.

Significance. If the claims are validated, NestQuant addresses a real practical need: resource-adaptive on-device inference without retraining, special hardware, or multiple model copies. The paper has notable strengths: the full-bit model is preserved exactly by construction through the 1-bit compensation, the part-bit accuracies are often close to full-bit accuracies across quite different architectures, the method is data-free when built on SQuant, and the authors provide a code link. The main weakness is that the headline switching-overhead reduction is not a measured on-device cost but a numerical ratio of packed-bit model sizes. Because that number is central to the resource-adaptation claim, the manuscript needs either hardware measurements or a substantially weaker interpretation of the result before publication.

major comments (3)
  1. [Section 4.3.3, Table 11] The claimed reduction of roughly 78.1% in switching overhead is not measured but computed from packed-bit disk sizes; the text states that 'the overheads of model switching can be numerically calculated by disk size' and assumes P(in) is proportional to disk size. On real hardware, page-in/page-out latency includes fixed costs independent of bytes moved, and the NestQuant upgrade path additionally requires reading a byte range of w_low and recomposing w_high and w_low at runtime, which may add CPU time and peak memory. Because the abstract and introduction present this number as a headline benefit, please add on-device measurements of switching latency (and, if possible, energy) on the Raspberry Pi 4B testbed, or substantially weaken the claim to 'reduced data volume moved during switching.'
  2. [Section 3.2.1, Eq. (9)] The claim that the higher-bit weights w_high can be optimized by the same Hessian-based adaptive rounding as the original weights is asserted rather than established. Equation (9) minimizes δw_intᵀ H(w_int) δw_int over the perturbation δw_int = w_low/2^l, but the Hessian of the task loss is evaluated at the original full-precision weights, not at the decomposed higher-bit weights; there is no derivation showing that the minimizer of this surrogate is the optimal w_high. The distributional statistics in Section 3.2.2 demonstrate similarity between w_high and w_int, but they do not prove that the optimization transfer holds. Please provide a derivation, an ablation against direct optimization of w_high or simpler rounding baselines, or an explicit statement that this is an empirical design choice.
  3. [Section 3.3.1, Eq. (12)] The 'Emerging Patterns' rule is derived from and evaluated on the same architectures whose critical nested combinations are reported in Section 4, so the 30 MB and 300 MB cut-offs are fit to the validation set rather than tested on held-out models. Since Algorithm 1 step (i) uses this rule to select h, generalizing NestQuant to a new model family currently rests on an unvalidated regression. The suggested fallback heuristic of trying INT(n|n/2) first mitigates this, but the paper should either test Eq. (12) on unseen architectures or clearly label it as a post-hoc observation rather than a predictive law.
minor comments (6)
  1. [Section 4.2] The statement that 'all part-bit models can upgrade to the full-bit ones without any performance degradation' is a construction property of the 1-bit compensation, not an empirical finding; consider stating this explicitly early so readers do not interpret it as a measured result.
  2. [Section 3.3.3] The paper should state clearly that the NestQuant model stores h + (l+1) = n+1 bits per weight, so its model size is slightly larger than a single n-bit model; the reported storage reductions are relative to storing two separate models, which is the correct comparison for the switching scenario but should be made explicit.
  3. [Equation (11)] The argument order in Equation (11), Clip(..., max_low, min_low), contradicts Equation (2)'s Clip(..., min, max); please align the notation.
  4. [Algorithm 1] Algorithm 1 does not specify how the 'adaptive rounding' of w_high is performed or whether it reuses SQuant's optimization with the same hyperparameters; please provide the objective and the data requirements (none, according to Section 4.1) for reproducibility.
  5. [Table 13] The model-size entries for the QAT baselines are given as '%FP32' without a definition; please clarify whether these methods store a single FP32 model or multiple bitwidth copies, since that determines the fairness of the storage comparison.
  6. [Section 3.3.3] There is a typo in the fourth paragraph: 'NestQaunt' should be 'NestQuant'.

Circularity Check

2 steps flagged · score 4.0 of 10

Accuracy claims are externally grounded, but the model-selection rule and the headline switching-overhead saving are post-hoc or by-construction.

  1. fitted input called prediction [Section 3.3.1 'Emerging Patterns' / Eq. (12); Section 4.2.2]
    "Evidently, two clear cut-offs at model sizes of 30 MB and 300 MB emerged. Accordingly, given FP32 model sizeDfp32, we can identify the following patterns of the critical nested combination:"

    The 30 MB and 300 MB cut-offs are read off Fig. 7, which plots exactly the models whose critical nested combinations were measured in Section 4.2 (ResNet/DenseNet/ResNeXt, lightweight models, ViTs). Equation (12) therefore encodes the observed outcomes (h=n/2+1 below 30 MB, h=n/2 in 30-300 MB, h=n/2-1 above 300 MB) rather than predicting them from an independent rule. Algorithm 1 Step 7 then uses this rule to select h, so the 'pattern' is fitted to the same data it is used to select. No held-out model validates the thresholds.

  2. renaming known result [Section 4.3.3 'Numerical Computation of Switching Overheads', Table 11]
    "The overheads of model switching can be numerically calculated by disk size, and the premise of numerical computations is that the size of packed-bit weights indeed approximates the ideal quantized model size. ... P(in)high = Dhigh ≈ h/(h+l+1)·D; the upgrading page-in overheads of wlow can be calculated by: P(in)low = Dlow ≈ (l+1)/(h+l+1)·D."

    Here 'page-in overhead' is defined as a disk-size fraction, so Table 11's 'Reduced Overhead' is the same arithmetic as the storage reductions in Section 4.3.2: for INT(8|6), NestQuant moves w_low ≈ 3 bits per weight, while diverse bitwidths move INT8+INT6 = 14 bits, yielding the 78.1% by construction. The table is a numerical computation, not a measured page-in/out latency or energy; the headline 'switching overhead' saving is a renamed storage/transfer saving, not an independent on-device result.

full rationale

The core NestQuant contribution—that INT(n|h) nesting with adaptive rounding keeps full-bit accuracy and usable part-bit accuracy—is supported by ImageNet-1K experiments on external pretrained models using the external SQuant rounding optimizer; it does not reduce to the paper's own definitions. The integer decomposition in Eq. (6) is an identity, and the 1-bit compensation makes recomposition exact by construction, which the paper states rather than disguises. Self-citations ([5], [28]-[30]) appear only in related work and are not load-bearing. The two localized circular elements are: (i) the 'Emerging Patterns' thresholds of Eq. (12) are fitted to the same models used to demonstrate them, so as a selection rule it is post-hoc; (ii) Section 4.3.3 defines switching overhead as packed-bit disk size, so the 78.1% overhead reduction is a restatement of storage arithmetic, not a measured hardware saving. These issues affect the deployment-rule and overhead-claim strength, not the independent accuracy measurements.

Assumptions & free parameters 2 free parameters · 7 assumptions · 0 invented entities

The central claims rest on standard arithmetic, standard PTQ Hessian assumptions, the transfer of Hessian rounding to decomposed weights, a post-hoc selection rule for nested bits, and an unverified disk-size proxy for switching overhead. The method introduces one design hyperparameter (nested bit h) plus fitted thresholds for the selection rule, but no new physical or mathematical entities.

free parameters (2)
  • Nested bitwidth h = per-model values (e.g., 4 for ResNet-101, 5 for MobileNet, 3 for Swin-B)
    The choice of h is determined by accuracy sweeps across candidate values for each model, making it a per-model hyperparameter rather than a derived quantity.
  • Critical nested combination thresholds in Eq. (12) = 30 MB and 300 MB FP32 model size cutoffs
    The 3e1 MB and 3e2 MB thresholds are chosen based on the evaluated models' sizes; they are fit post hoc from the same data used to validate the rule.
assumptions (7)
  • standard math Binary decomposition identity: w_int = w_high * 2^l + w_low holds for the bit-split of an integer tensor.
    Used in Eq. (6) and Eq. (7); this is a standard property of binary arithmetic and requires no proof.
  • domain assumption Symmetric linear quantization with signed INT weights and zero-point ignored is sufficient for the evaluated models.
    Section 3.1 states this simplification; it excludes asymmetric quantization, which is common in some deployment stacks, but is standard in the cited PTQ literature.
  • domain assumption The Hessian-based task loss approximation in Eq. (1) is valid: gradients vanish at convergence and the second-order term dominates.
    Adopted from AdaRound, BRECQ, OBQ, and SQuant; the paper relies on it without re-derivation.
  • ad hoc to paper Hessian-based optimization transfers to the decomposed higher-bit weights as stated in Eq. (9).
    Section 3.2.1 claims that because the perturbation forms are similar, w_high can be optimized with the same Hessian objective; this is an analogy, not a proof, and is not flagged as an assumption.
  • domain assumption SQuant's data-free diagonal Hessian approximation is accurate for all evaluated architectures.
    Algorithm 1 uses SQuant as the adaptive rounding engine; the paper does not independently quantify approximation error per model.
  • ad hoc to paper The critical nested combination rule of Eq. (12) generalizes to models beyond those evaluated.
    The 30 MB and 300 MB thresholds are derived from the same models used to present the rule, with no held-out validation.
  • domain assumption Disk size of packed-bit tensors is proportional to memory page-in/page-out overhead.
    Section 4.3.3 numerically computes switching overheads from disk sizes; actual page-in/out latency is not measured.

how reviews work

0 comments
Cite this review

Pith. "Pith review of NestQuant: Post-Training Integer-Nesting Quantization for On-Device DNN." pith.science (2026). https://pith.science/paper/MORYV3Y4

@misc{pith2026250617870,
  author       = {Pith},
  title        = {Pith review of: NestQuant: Post-Training Integer-Nesting Quantization for On-Device DNN},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MORYV3Y4}},
  note         = {Machine review of arXiv:2506.17870}
}
read the original abstract

Deploying quantized deep neural network (DNN) models with resource adaptation capabilities on ubiquitous Internet of Things (IoT) devices to provide high-quality AI services can leverage the benefits of compression and meet multi-scenario resource requirements. However, existing dynamic/mixed precision quantization requires retraining or special hardware, whereas post-training quantization (PTQ) has two limitations for resource adaptation: (i) The state-of-the-art PTQ methods only provide one fixed bitwidth model, which makes it challenging to adapt to the dynamic resources of IoT devices; (ii) Deploying multiple PTQ models with diverse bitwidths consumes large storage resources and switching overheads. To this end, this paper introduces a resource-friendly post-training integer-nesting quantization, i.e., NestQuant, for on-device quantized model switching on IoT devices. The proposed NestQuant incorporates the integer weight decomposition, which bit-wise splits quantized weights into higher-bit and lower-bit weights of integer data types. It also contains a decomposed weights nesting mechanism to optimize the higher-bit weights by adaptive rounding and nest them into the original quantized weights. In deployment, we can send and store only one NestQuant model and switch between the full-bit/part-bit model by paging in/out lower-bit weights to adapt to resource changes and reduce consumption. Experimental results on the ImageNet-1K pretrained DNNs demonstrated that the NestQuant model can achieve high performance in top-1 accuracy, and reduce in terms of data transmission, storage consumption, and switching overheads. In particular, the ResNet-101 with INT8 nesting INT6 can achieve 78.1% and 77.9% accuracy for full-bit and part-bit models, respectively, and reduce switching overheads by approximately 78.1% compared with diverse bitwidths PTQ models.

Figures

Figures reproduced from arXiv: 2506.17870 by the authors.

Figure 1
Figure 1. Existing system architecture for on-device deploy [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Illustration of integer weight decomposition. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Distribution of wˆ, wˆhigh and wˆlow. 0.00 0.02 0.04 0.06 0.08 0.10 Value 0 1 2 3 4 5 6 7 8 Density 1e2 0.004 0.063 KDE of high KDE of low 95% CI of high 95% CI of low UB of high: 0.004 UB of low: 0.063 (a) INT(8|5) 0.00 0.02 0.04 0.06 0.08 0.10 Value 0.0 0.5 1.0 1.5 2.0 2.5 Density 1e2 0.008 0.064 KDE of high KDE of low 95% CI of high 95% CI of low UB of high: 0.008 UB of low: 0.064 (b) INT(8|4) 0.00 0.02 0.04 0.06… view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: Kernel Density Estimation (KDE) and Upper Bound (UB) of 95% Confidence Intervals (CIs) of [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Overview of the proposed NestQuant. Algorithm 1: NestQuant Procedures Input: A floating-point DNN model Mfp Output: An INT(n|h) model Mint(n|h) 1 for weights w(ℓ) of layer 1, 2, . . . , ℓ in Mfp do 2 /* Step ①: INTn Hessian-based quantization for w (ℓ) int */ 3 Compute…
Figure 6
Figure 6. Figure 6: Performance cliff of the PTQ model. in Mfp, we can obtain the NestQuant model Mint(n|h) , which can switch between full-bit and part-bit models. 3.3.1 Part-Bit Model: Effective Nested Combinations As shown in [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 8
Figure 8. Figure 8: NestQuant effective nested combinations, compensa [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: The case of decomposed signed INT data with [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 10
Figure 10. Figure 10: Performance of INT8 nesting quantization. [PITH_FULL_IMAGE:figures/full_fig_p009_10.png]
Figure 11
Figure 11. Figure 11: Performance of INT6 nesting quantization. [PITH_FULL_IMAGE:figures/full_fig_p009_11.png]
Figure 12
Figure 12. Figure 12: Performance of INT8 nesting in lightweight CNNs. [PITH_FULL_IMAGE:figures/full_fig_p009_12.png]
Figure 13
Figure 13. Figure 13: Comparison in ResNet series network traffic. [PITH_FULL_IMAGE:figures/full_fig_p010_13.png]
Figure 14
Figure 14. Figure 14: Comparison in lightweight models network traffic. [PITH_FULL_IMAGE:figures/full_fig_p011_14.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

51 extracted references · 47 canonical work pages

  1. [1]

    Deep- wear: Adaptive local offloading for on-wearable deep learning,

    M. Xu, F. Qian, M. Zhu, F. Huang, S. Pushp, and X. Liu, “Deep- wear: Adaptive local offloading for on-wearable deep learning,” IEEE Transactions on Mobile Computing, vol. 19, no. 2, pp. 314–330, 2020

    work page 2020

  2. [2]

    A new deep learning-based food recognition system for dietary assessment on an edge computing service infrastructure,

    C. Liu, Y. Cao, Y. Luo, G. Chen, V . Vokkarane, M. Yunsheng, S. Chen, and P . Hou, “A new deep learning-based food recognition system for dietary assessment on an edge computing service infrastructure,”IEEE Transactions on Services Computing, vol. 11, no. 2, pp. 249–261, 2018

    work page 2018

  3. [3]

    A first look at deep learning apps on smartphones,

    M. Xu, J. Liu, Y. Liu, F. X. Lin, Y. Liu, and X. Liu, “A first look at deep learning apps on smartphones,” inProceedings of the World Wide Web Conference, 2019, p. 2125–2136

    work page 2019

  4. [4]

    A comprehensive deep learning library benchmark and optimal library selection,

    Q. Zhang, X. Che, Y. Chen, X. Ma, M. Xu, S. Dustdar, X. Liu, and S. Wang, “A comprehensive deep learning library benchmark and optimal library selection,”IEEE Transactions on Mobile Computing, vol. 23, no. 5, pp. 5069–5082, 2024

    work page 2024

  5. [5]

    Secaas-based partially observable defense model for iiot against advanced persistent threats,

    Z. Zhang, C. Ding, Y. Li, J. Yu, and J. Li, “Secaas-based partially observable defense model for iiot against advanced persistent threats,”IEEE Transactions on Services Computing, vol. 17, no. 6, pp. 4267–4280, 2024

    work page 2024

  6. [6]

    Multi-dimensional qos evaluation and optimization of mobile edge computing for iot: A survey,

    J. Huang, F. Liu, and J. Zhang, “Multi-dimensional qos evaluation and optimization of mobile edge computing for iot: A survey,” Chinese Journal of Electronics, vol. 33, no. 4, pp. 859–874, 2024

    work page 2024

  7. [7]

    An intelligent privacy protection scheme for efficient edge computation offloading in iov,

    L. Yao, X. Xu, W. Dou, and M. Bilal, “An intelligent privacy protection scheme for efficient edge computation offloading in iov,”Chinese Journal of Electronics, vol. 33, no. 4, pp. 910–919, 2024

    work page 2024

  8. [8]

    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,” inProceedings of the International Conference on Learning Representations, 2016, pp. 1–14

    work page 2016

Show all 51 references
  1. [9]

    Quantized neural networks: Training neural networks with low precision weights and activations,

    I. Hubara, M. Courbariaux, D. Soudry, R. EI-Yaniv, and Y. Bengio, “Quantized neural networks: Training neural networks with low precision weights and activations,”Journal of Machine Learning Research, vol. 18, no. 187, pp. 1–30, 2018. IEEE TRANSACTIONS ON MOBILE COMPUTING 14 T...

    2018

  2. [10]

    Post training 4-bit quan- tization of convolutional networks for rapid-deployment,

    R. Banner, Y. Nahshan, and D. Soudry, “Post training 4-bit quan- tization of convolutional networks for rapid-deployment,” inAd- vances in Neural Information Processing Systems, 2019, pp. 7950–7958

    2019

  3. [11]

    Adabits: Neural network quanti- zation with adaptive bit-widths,

    Q. Jin, L. Yang, and Z. Liao, “Adabits: Neural network quanti- zation with adaptive bit-widths,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 2143–2153

    2020

  4. [12]

    Any-precision deep neural networks,

    H. Yu, H. Li, H. Shi, T. S. Huang, and G. Hua, “Any-precision deep neural networks,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 35, no. 12, 2021, pp. 10 763–10 771

    2021

  5. [13]

    Eq-net: Elastic quantization neural networks,

    K. Xu, L. Han, Y. Tian, S. Yang, and X. Zhang, “Eq-net: Elastic quantization neural networks,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 1505–1514

    2023

  6. [14]

    Spark: Scalable and precision-aware acceleration of neural net- works via efficient encoding,

    F. Liu, N. Yang, H. Li, Z. Wang, Z. Song, S. Pei, and L. Jiang, “Spark: Scalable and precision-aware acceleration of neural net- works via efficient encoding,” inProceedings of the IEEE Interna- tional Symposium on High-Performance Computer Architecture, 2024, pp. 1029–1042

    2024

  7. [15]

    Up or down? adaptive rounding for post-training quantization,

    M. Nagel, R. A. Amjad, M. Van Baalen, C. Louizos, and T. Blankevoort, “Up or down? adaptive rounding for post-training quantization,” inProceedings of the International Conference on Ma- chine Learning, 2020, pp. 7197–7206

    2020

  8. [16]

    Accurate post training quantization with small calibration sets,

    I. Hubara, Y. Nahshan, Y. Hanani, R. Banner, and D. Soudry, “Accurate post training quantization with small calibration sets,” inProceedings of the International Conference on Machine Learning, 2021, pp. 4466–4475

    2021

  9. [17]

    Brecq: Pushing the limit of post-training quantization by block reconstruction,

    Y. Li, R. Gong, X. Tan, Y. Yang, P . Hu, Q. Zhang, F. Yu, W. Wang, and S. Gu, “Brecq: Pushing the limit of post-training quantization by block reconstruction,” inProceedings of the International Confer- ence on Learning Representations, 2021

    2021

  10. [18]

    Optimal brain compres- sion: A framework for accurate post-training quantization and pruning,

    E. Frantar, S. P . Singh, and D. Alistarh, “Optimal brain compres- sion: A framework for accurate post-training quantization and pruning,” inAdvances in Neural Information Processing Systems, 2022, pp. 4475–4488

    2022

  11. [19]

    SQuant: On-the-fly data-free quantization via di- agonal hessian approximation,

    C. Guo, Y. Qiu, J. Leng, X. Gao, C. Zhang, Y. Liu, F. Yang, Y. Zhu, and M. Guo, “SQuant: On-the-fly data-free quantization via di- agonal hessian approximation,” inProceedings of the International Conference on Learning Representations, 2022

    2022

  12. [20]

    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,” inProceedings of 26th Annual International Conference on Mobile Computing and Networking, 2020, pp. 488–502

    2020

  13. [21]

    Cnnpc: End-edge-cloud collaborative cnn inference with joint model parti- tion and compression,

    S. Yang, Z. Zhang, C. Zhao, X. Song, S. Guo, and H. Li, “Cnnpc: End-edge-cloud collaborative cnn inference with joint model parti- tion and compression,”IEEE Transactions on Parallel and Distributed Systems, vol. 33, no. 12, pp. 4039–4056, 2022

    2022

  14. [22]

    Energy efficient federated learning over heterogeneous mobile devices via joint design of weight quantization and wireless transmission,

    R. Chen, L. Li, K. Xue, C. Zhang, M. Pan, and Y. Fang, “Energy efficient federated learning over heterogeneous mobile devices via joint design of weight quantization and wireless transmission,” IEEE Transactions on Mobile Computing, vol. 22, no. 12, pp. 7451– 7465, 2023

    2023

  15. [23]

    Lightfr: Lightweight federated recommendation with privacy-preserving matrix factor- ization,

    H. Zhang, F. Luo, J. Wu, X. He, and Y. Li, “Lightfr: Lightweight federated recommendation with privacy-preserving matrix factor- ization,”ACM Transactions on Information Systems, vol. 41, no. 4, pp. 1–28, 2023

    2023

  16. [24]

    Communication-efficient satellite-ground federated learn- ing through progressive weight quantization,

    C. Yang, J. Yuan, Y. Wu, Q. Sun, A. Zhou, S. Wang, and M. Xu, “Communication-efficient satellite-ground federated learn- ing through progressive weight quantization,”IEEE Transactions on Mobile Computing, vol. 23, no. 9, pp. 8999–9011, 2024

    2024

  17. [25]

    Nestdnn: Resource-aware multi- tenant on-device deep learning for continuous mobile vision,

    B. Fang, X. Zeng, and M. Zhang, “Nestdnn: Resource-aware multi- tenant on-device deep learning for continuous mobile vision,” in Proceedings of the 24th Annual International Conference on Mobile Computing and Networking, 2018, pp. 115–127

    2018

  18. [26]

    Flexdnn: Input- adaptive on-device deep learning for efficient mobile vision,

    B. Fang, X. Zeng, F. Zhang, H. Xu, and M. Zhang, “Flexdnn: Input- adaptive on-device deep learning for efficient mobile vision,” in Proceedings of 5th IEEE/ACM Symposium on Edge Computing, 2020, pp. 84–95

    2020

  19. [27]

    Adaptivenet: Post-deployment neural architecture adaptation for diverse edge environments,

    H. Wen, Y. Li, Z. Zhang, S. Jiang, X. Ye, Y. Ouyang, Y. Zhang, and Y. Liu, “Adaptivenet: Post-deployment neural architecture adaptation for diverse edge environments,” inProceedings of the 29th Annual International Conference on Mobile Computing and Net- working, 2023, pp. 408–424

    2023

  20. [28]

    Resource-aware feature extraction in mobile edge computing,

    C. Ding, A. Zhou, X. Liu, X. Ma, and S. Wang, “Resource-aware feature extraction in mobile edge computing,”IEEE Transactions on Mobile Computing, vol. 21, no. 1, pp. 321–331, 2022

    2022

  21. [29]

    A resource-efficient feature extraction framework for image processing in iot devices,

    C. Ding, Y. Li, Z. Lu, S. Wang, and S. Guo, “A resource-efficient feature extraction framework for image processing in iot devices,” IEEE Transactions on Mobile Computing, vol. 23, no. 1, pp. 42–55, 2024

    2024

  22. [30]

    Edge/cloud-assisted feature extrac- tion in iot devices,

    C. Ding, Y. Li, and S. Wang, “Edge/cloud-assisted feature extrac- tion in iot devices,”IEEE Internet of Things Journal, vol. 9, no. 21, pp. 21 594–21 606, 2022

    2022

  23. [31]

    [Online]

    The gflops/w of the various machines in the vmw research group. [Online]. Available: https://web.eece.maine. edu/∼vweaver/group/green machines.html

  24. [32]

    [Online]

    Nvidia jetson technical specifications. [Online]. Available: https://developer.nvidia.com/embedded/jetson-modules

  25. [33]

    [Online]

    Tensorflow lite: Tensorflow’s lightweight solution for mobile and embedded devices. [Online]. Available: https://www.tensorflow. org/lite

  26. [34]

    [Online]

    Pytorch mobile: End-to-end workflow from training to deployment for ios and android mobile devices. [Online]. Available: https://pytorch.org/mobile

  27. [35]

    [Online]

    Ncnn: A high-performance neural network inference computing framework optimized for mobile platforms. [Online]. Available: https://github.com/Tencent/ncnn

  28. [36]

    [Online]

    Onnx: Open standard for machine learning interoperability. [Online]. Available: https://github.com/onnx/onnx IEEE TRANSACTIONS ON MOBILE COMPUTING 15

  29. [37]

    [Online]

    Onnx runtime: Cross-platform, high performance ml inferencing and training accelerator. [Online]. Available: https://github.com/ microsoft/onnxruntime

  30. [38]

    Deep dif- ferentiable logic gate networks,

    F. Petersen, C. Borgelt, H. Kuehne, and O. Deussen, “Deep dif- ferentiable logic gate networks,” inAdvance in Neural Information Processing Systems, 2022, pp. 2006–2018

    2022

  31. [39]

    Petersen and T

    F. Petersen and T. Sutter. Distributional Quantization. [Online]. Available: https://github.com/Felix-Petersen/distquant

  32. [40]

    Im- agenet: A large-scale hierarchical image database,

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei, “Im- agenet: A large-scale hierarchical image database,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2009, pp. 248–255

    2009

  33. [41]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2016, pp. 770–778

    2016

  34. [42]

    Densely connected convolutional networks,

    G. Huang, Z. Liu, L. Van Der Maaten, and K. Q. Weinberger, “Densely connected convolutional networks,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 2261–2269

    2017

  35. [43]

    Aggregated residual transformations for deep neural networks,

    S. Xie, R. Girshick, P . Doll ´ar, Z. Tu, and K. He, “Aggregated residual transformations for deep neural networks,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 5987–5995

    2017

  36. [44]

    Mobilenets: Efficient convolutional neural networks for mobile vision applications,

    A. G. Howard, M. Zhu, B. Chen, D. Kalenichenko, W. Wang, T. Weyand, M. Andreetto, and H. Adam, “Mobilenets: Efficient convolutional neural networks for mobile vision applications,” in arXiv preprint arXiv:1704.04861, 2017

    2017 arXiv

  37. [45]

    Mobilenetv2: Inverted residuals and linear bottlenecks,

    M. Sandler, A. G. Howard, M. Zhu, A. Zhmoginov, and L.-C. Chen, “Mobilenetv2: Inverted residuals and linear bottlenecks,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018, pp. 4510–4520

    2018

  38. [46]

    Shufflenet: An extremely efficient convolutional neural network for mobile devices,

    X. Zhang, X. Zhou, M. Lin, and J. Sun, “Shufflenet: An extremely efficient convolutional neural network for mobile devices,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018, pp. 6848–6856

    2018

  39. [47]

    Shufflenet v2: Practical guidelines for efficient cnn architecture design,

    N. Ma, X. Zhang, H.-T. Zheng, and J. Sun, “Shufflenet v2: Practical guidelines for efficient cnn architecture design,” inProceedings of the European Conference on Computer Vision, 2018, pp. 116–131

    2018

  40. [48]

    EfficientNet: Rethinking model scaling for convolutional neural networks,

    M. Tan and Q. Le, “EfficientNet: Rethinking model scaling for convolutional neural networks,” inProceedings of the International Conference on Machine Learning, 2019, pp. 6105–6114

    2019

  41. [49]

    An image is worth 16x16 words: Transformers for image recognition at scale,

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, and N. Houlsby, “An image is worth 16x16 words: Transformers for image recognition at scale,” inProceedings of the International Conf...

    2021

  42. [50]

    Training data-efficient image transformers & distilla- tion through attention,

    H. Touvron, M. Cord, M. Douze, F. Massa, A. Sablayrolles, and H. Jegou, “Training data-efficient image transformers & distilla- tion through attention,” inProceedings of the International Conference on Machine Learning, 2021, pp. 10 347–10 357

    2021

  43. [51]

    Swin transformer: Hierarchical vision transformer using shifted windows,

    Z. Liu, Y. Lin, Y. Cao, H. Hu, Y. Wei, Z. Zhang, S. Lin, and B. Guo, “Swin transformer: Hierarchical vision transformer using shifted windows,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 9992–10 002

    2021

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.