Pith. sign in

REVIEW 4 major objections 5 minor 53 references

NMS: Efficient Edge DNN Training via Near-Memory Sampling on Manifolds

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read No-DNN sample picker beats coresets and cuts edge training energy 74x

desk verdict Hardware story is plausible but the central algorithmic claim—that t-SNE perplexity search needs differential evolution—is mathematically false; paper overreaches. read the letter →

arxiv 2508.02313 v1 pith:PZJPUAX3 submitted 2025-08-04 cs.LG

Boran Zhao , Haiduo Huang , Qiwei Dang , Wenzhe Zhao , Tian Xia , Pengju Ren This is my paper
classification cs.LG
keywords edgeDNNtrainingcoresetselectiont-SNEdifferentialevolutionnear-memorycomputingdatasetcompressiongeneralizationDRAMenergy
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

Training a deep network on an edge device is impractical if every image in a large dataset must cross a long PCB bus from DRAM to the compute chip, and existing dataset-compression methods add a second cost: they use a DNN to judge which samples matter, so the chosen subset inherits that network's bias and transfers poorly to other architectures. This paper proposes a sampling pipeline that avoids DNNs altogether: an improved t-SNE embedding, DE-SNE, maps images into a low-dimensional manifold, grids that space, and picks representative samples that cover the whole feature distribution. The same routine is implemented in logic placed next to DRAM cells, so only the selected images travel to the training accelerator. The claim is that this removes the generalization gap of DNN-based coresets while cutting DRAM energy by up to 74x relative to the closest prior system, and the paper reports accuracy gains on ImageNet-1K of 11.9% over DQ, 9.7% over DQAS, and 4.7% over NeSSA.

What carries the argument

The load-bearing machinery is DE-SNE, defined as t-SNE with differential evolution substituted for the binary search that fixes each point's Gaussian width $\sigma_i$. t-SNE builds conditional probabilities $p_{j|i}$ from $\sigma_i$ and targets a fixed perplexity; DE-SNE evolves a population of candidate $\sigma$ values through mutation, crossover, and selection on perplexity error, which the paper argues is needed because perplexity is non-monotonic in $\sigma$. The embedding output is gridded and sampled in parallel, converting the manifold coordinates back into a covered subset of original images. Around this sits the near-memory sampling circuit, in which processing-element arrays compute the distance matrix, probability matrices, and gradients inside DRAM, with tree elements handling comparisons, so that data movement over the board is limited to the final selected images.

What would settle it

Take random samples from CIFAR-10, CIFAR-100, and ImageNet-1K, and for each image sweep $\sigma_i$ over a fine grid from $10^{-20}$ to $10^3$, plotting the resulting perplexity $P_i = 2^{H_i}$. If every curve is monotonically increasing, the paper's stated reason for replacing binary search is false and the reported DE-SNE gains would need another explanation; if any curve has a local maximum or multiple crossings at the target perplexity, the paper's mechanism is directly confirmed.

Watch

Extended reading notes

Core claim

The central discovery, stated on the paper's own terms, is that representative samples for DNN training can be chosen from the raw data distribution alone, with no forward passes through the target network. t-SNE is used to collapse high-dimensional images into a low-dimensional manifold while preserving local distances, and the usual binary search for the Gaussian width $\sigma$ is replaced by differential evolution, because the authors find perplexity to be a non-monotonic function of $\sigma$ and therefore unsuited to bisection. The embedding is then gridded and sampled in parallel so the selected subset covers the feature space. Since no DNN is involved, the same subset transfers across ResNet, ShuffleNet, MobileNet, and ViT. Implementing the DE-SNE arithmetic in 3D-stacked near-memory DRAM logic means only the chosen images cross the PCB, yielding the reported average Top-1 improvements on ImageNet-1K of 11.9% over DQ, 9.7% over DQAS, and 4.7% over NeSSA, and DRAM energy reductions up to 74x at a 10% keeping ratio.

Load-bearing premise

The argument rests on the claim that t-SNE's perplexity is a non-monotonic function of the Gaussian width $\sigma$, so binary search can fail and an evolutionary search is needed; the paper asserts this without proof.

Editorial extensions

If this is right

  • Because sampling no longer depends on a DNN, the same DE-SNE subset can be reused across heterogeneous architectures and can be computed once per dataset rather than once per model.
  • Moving the selector into DRAM removes most of the energy cost of reading the full dataset: reported DRAM energy savings range from about 32x to 74x depending on the keeping ratio.
  • The approach scales in its advantage: the largest accuracy gains over DQ, DQAS, and NeSSA appear on ImageNet-1K, the largest dataset tested.
  • DE-SNE's evolutionary search makes t-SNE's $\sigma$ computation stable enough for fixed-point hardware, enabling manifold sampling on battery-powered devices.
  • At a 60% keeping ratio, training on the sampled set loses less than 1% accuracy compared with full data while the combined sampling-plus-training system improves energy efficiency by 1.2x over a sparse training accelerator.

Reading between the lines

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

  • The paper leaves implicit that the same differential-evolution fix could stabilize $\sigma$ search in other t-SNE applications, such as single-cell RNA visualization or word-embedding plots, where the binary-search bottleneck is identical.
  • The near-memory blueprint generalizes: any DNN-free selection rule cheap enough to run inside DRAM logic could make dataset pruning, active learning, and replay-buffer construction nearly energy-free for edge devices.
  • A natural but untested extension is to check whether DE-SNE subsets also transfer across optimizers, augmentation policies, and label shifts, where DNN-based coresets are known to be brittle.
  • If the non-monotonic perplexity claim is confirmed, it would suggest that many published t-SNE embeddings have silently used failed or imprecise $\sigma$ searches, so DE-SNE could improve visualization fidelity independently of its use in coresets.
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. The paper proposes NMS, a system for edge DNN training that combines DE-SNE, a t-SNE variant in which differential evolution replaces binary search for perplexity optimization, with a near-memory computing architecture. The authors claim that DE-SNE is a DNN-free data sampling method that improves generalization over DNN-based coresets, and that placing the sampling circuit near DRAM reduces DDR energy consumption. They report accuracy gains over DQ, DQAS, and NeSSA on six datasets, plus large DDR energy reductions and improved hardware efficiency when integrated with the THETA training accelerator.

Significance. If the claims held, the paper would address two real problems in edge training: generalization bias from DNN-based coresets and data-movement energy. The hardware portion is concrete, with a 28nm synthesis flow, CACTI-based energy estimates, and cycle-accurate simulation, which are strengths relative to purely algorithmic submissions. However, the core algorithmic motivation is mathematically incorrect: perplexity is monotonic in the Gaussian width, so the stated reason for replacing binary search is false. The accuracy comparisons lack error bars and code, and the hardware comparison mixes sampling-accelerator metrics with training-accelerator metrics. The central claims are therefore not supported in their current form.

major comments (4)
  1. [Section IV-B and Fig. 6] The premise that perplexity is non-monotonic in sigma is mathematically false under Eqs. (1)-(3). Writing beta = 1/(2 sigma^2) and p_{j|i} = exp(-beta d_{ij})/Z_i, the entropy H_i satisfies dH_i/dbeta = -beta * Var_p(d_{ij}) < 0 whenever the distances are not all equal. Since dbeta/dsigma < 0, H_i and the perplexity P_i = 2^{H_i} are strictly increasing in sigma. Binary search on sigma therefore converges to any target perplexity in exact arithmetic. The stated motivation for replacing binary search with differential evolution collapses, and the accuracy gap in Table II cannot be attributed to perplexity search error without additional evidence.
  2. [Table II and Section X] The ablation does not control for search convergence. Algorithm 1 specifies epsilon = 1e-10 and max iter = 10000; if binary search were run to the same tolerance on a monotonic continuous function, the resulting sigma values would agree to that tolerance and the downstream embeddings should be nearly identical. The reported 4-6 percentage point accuracy improvements are therefore unexplained. The claim in Section X that DE reduces perplexity error by 3, 16, 14, 10, and 12 orders of magnitude relative to binary search and other methods is implausible for a one-dimensional monotonic search and is not supported by any convergence diagnostics or code.
  3. [Tables I-II and Fig. 12] All accuracy numbers are reported as single values with no standard deviations, number of seeds, or code/data release. Since the claimed improvements over DQ, DQAS, and NeSSA are often only 1-3 percentage points, run-to-run variance could account for the differences. Additionally, the target perplexity of 15 and the DE hyperparameters are described as determined experimentally on the same datasets, so part of the reported gain may encode these choices. This is load-bearing because the central claim is superior accuracy.
  4. [Table III and Section XII] The hardware comparison mixes a sampling accelerator with training accelerators. The 'Our Work' row reports 66.87 TOPS and 176.12 TOPS/W, but the system under discussion integrates the DE-SNE sampling circuit with the THETA training accelerator. No end-to-end power, latency, or energy-per-trained-epoch measurement is provided for the integrated system, so the statement that the system improves efficiency by 1.2x over THETA lacks support. The comparison to DAC'19, DAC'20, SparTANN, GANPU, and other training accelerators is not apples-to-apples unless all rows execute the same workload.
minor comments (5)
  1. [Section VII-A] There is a typo: 'THATA' should be 'THETA' in the hardware setup paragraph, and 'AMD EPYCT M' should be 'AMD EPYC'.
  2. [Section X] The sentence 'the DE algorithm achieves error reductions of 3, 16, 14, 10, and 12 orders of magnitude on six commonly used datasets' lists five numbers but six datasets and does not identify which comparison corresponds to which dataset.
  3. [Fig. 5] The claim that the bisection search needs up to 5000 iterations is atypical; standard t-SNE implementations use on the order of 50 bisection steps. Please justify this search budget, since it is central to the runtime-bottleneck argument.
  4. [Fig. 16 and Section XI] The CPU and GPGPU energy comparison lacks a stated methodology: no workload details, power models, or measurement sources are given for the four-orders-of-magnitude and three-orders-of-magnitude energy reduction claims.
  5. [Algorithm 1, line 8] Standard differential evolution samples three distinct indices a, b, and c that differ from i; the current pseudocode only excludes i and could choose a = b, making the mutation degenerate. Please clarify the sampling rule.

Circularity Check

1 steps flagged · score 6.0 of 10

Perplexity-error 'prediction' is DE's own fitness function, making the error-reduction claim tautological; held-out accuracy keeps the core result independent.

  1. self definitional [Section VII-A (Software setup); Algorithm 1 lines 2 and 16-19; Section X and Fig. 15]
    "These algorithms are employed to solve the σ parameters on the above datasets, obtaining predicted perplexity values and comparing them with the input ground truth values to calculate the error. / Algorithm 1: best ind ← pop[arg min(|f (ind) − tgt| for ind in pop)]; if |trial val − tgt| < |f (pop[i]) − tgt| then pop[i] ← trial"

    DE-SNE's reported 'perplexity error' is the same quantity that Algorithm 1 optimizes: initialization selects the population member with argmin |f(ind)−tgt|, and every replacement is accepted only when |f(trial)−tgt| is smaller than the current value. Therefore the 3–16 orders-of-magnitude error reduction over binary search, simulated annealing, GA, PSO, and ACO is a property of the fitness function by construction, not an independent measurement of sample quality. The paper then relies on this tautologically small error to explain the Table II accuracy gap ('the superior training accuracy of DE-SNE arises from the inability of other algorithms to effectively address the perplexity issue'), so the causal story reduces to DE's objective.

full rationale

The main accuracy claim is not circular: DE-SNE-selected subsets are trained and evaluated on held-out test data and compared against external coreset methods (DQ, DQAS, NeSSA), so Table I and Fig. 12 are externally falsifiable. The circularity is in the supporting causal chain: Section X explains the accuracy gain by a perplexity-error comparison whose metric is identical to Algorithm 1's selection criterion. That makes the reported 3–16 orders-of-magnitude error reduction a construction, not evidence. The paper's premise that t-SNE perplexity is non-monotonic in sigma (Section IV-B, Fig. 6) is also mathematically doubtful under Eq. (3), where perplexity is strictly increasing in sigma for distinct pairwise distances; I treat that as a correctness risk rather than circularity. The target perplexity 15 is stated to be chosen by experimentation, so quantitative gains should be read as conditional on that tuning choice, though this too is a tuning caveat rather than a definitional reduction. Overall: one auxiliary prediction reduces by construction, so partial circularity (6), while the central accuracy comparison retains independent content.

Assumptions & free parameters 6 free parameters · 3 assumptions · 0 invented entities

The central claim rests on hand-tuned algorithmic parameters, an unproven property of perplexity, and energy assumptions from prior literature. The hardware itself is a design not a physical entity.

free parameters (6)
  • target perplexity = 15
    Chosen via experimentation on the same datasets used for evaluation (Section VII-A).
  • DE population size = 30
    Algorithm 1 sets pop size to 30 without sensitivity analysis.
  • DE scaling factor F = 0.5
    Algorithm 1, line 9; standard default, not tuned.
  • DE crossover rate CR = 0.7
    Algorithm 1, line 13; standard default, not tuned.
  • grid resolution for gridding step = unspecified
    The gridding of the manifold space is described in words but no grid size is given; it affects how many samples are kept per cell.
  • keeping ratio for THETA system comparison = 60%
    Section XII assumes a 60% keeping ratio with less than 1% accuracy loss, but no experiment at 60% is shown.
assumptions (3)
  • domain assumption t-SNE correctly preserves local similarities of raw images in low-dimensional space
    The whole sampling method relies on t-SNE embeddings being meaningful for image similarity; no validation against semantic labels is provided.
  • ad hoc to paper Perplexity is non-monotonic in sigma, making binary search unreliable
    The paper asserts this in Section IV-B without proof, and it contradicts standard t-SNE practice.
  • domain assumption Near-memory computing reduces per-bit energy by about 20x
    Taken from UCIe/PCIe references [44], [45]; used as the basis for all DDR energy savings calculations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of NMS: Efficient Edge DNN Training via Near-Memory Sampling on Manifolds." pith.science (2026). https://pith.science/paper/PZJPUAX3

@misc{pith2026250802313,
  author       = {Pith},
  title        = {Pith review of: NMS: Efficient Edge DNN Training via Near-Memory Sampling on Manifolds},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PZJPUAX3}},
  note         = {Machine review of arXiv:2508.02313}
}
read the original abstract

Training deep neural networks (DNNs) on edge devices has attracted increasing attention due to its potential to address challenges related to domain adaptation and privacy preservation. However, DNNs typically rely on large datasets for training, which results in substantial energy consumption, making the training in edge devices impractical. Some dataset compression methods have been proposed to solve this challenge. For instance, the coreset selection and dataset distillation reduce the training cost by selecting and generating representative samples respectively. Nevertheless, these methods have two significant defects: (1) The necessary of leveraging a DNN model to evaluate the quality of representative samples, which inevitably introduces inductive bias of DNN, resulting in a severe generalization issue; (2) All training images require multiple accesses to the DDR via long-distance PCB connections, leading to substantial energy overhead. To address these issues, inspired by the nonlinear manifold stationary of the human brain, we firstly propose a DNN-free sample-selecting algorithm, called DE-SNE, to improve the generalization issue. Secondly, we innovatively utilize the near-memory computing technique to implement DE-SNE, thus only a small fraction of images need to access the DDR via long-distance PCB. It significantly reduces DDR energy consumption. As a result, we build a novel expedited DNN training system with a more efficient in-place Near-Memory Sampling characteristic for edge devices, dubbed NMS. As far as we know, our NMS is the first DNN-free near-memory sampling technique that can effectively alleviate generalization issues and significantly reduce DDR energy caused by dataset access. The experimental results show that our NMS outperforms the current state-of-the-art (SOTA) approaches, namely DQ, DQAS, and NeSSA, in model accuracy.

Figures

Figures reproduced from arXiv: 2508.02313 by the authors.

Figure 1
Figure 1. The gap between the energy required for training on edge devices [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The human brain processes information through [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 5
Figure 5. Breakdown of the running times for various operations in the t [PITH_FULL_IMAGE:figures/full_fig_p004_5.png] view at source ↗
Figures from the paper (11 more)
Figure 4
Figure 4. Figure 4: The mismatch phenomenon is evident between samples using [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 7
Figure 7. Figure 7: Illustration of the brain-inspired DE-SNE dataset sampling compres [PITH_FULL_IMAGE:figures/full_fig_p005_7.png]
Figure 8
Figure 8. Figure 8: The accelerator of DE-SNE. 2 a x b x X ' 2 ' 2 2 R  x  x  X  2XX  ( X ) a b 1 1 1 1 2 2 2 2 3 3 3 3 4 4 4 4 X i0 j0 4 16 36 64 2 X Dataflow: for(i0=0:2:1) for(j1=0:2:1) pfor(i0=0:2:1)→n pfor(j0=0:2:1)→m HW Parallel clk-0： 1 1 2 2 =12+12 = 2 =22+22 = 8 clk-1： 1 1 2…
Figure 9
Figure 9. Figure 9: Dataflow of calculating distance matrix. [PITH_FULL_IMAGE:figures/full_fig_p006_9.png]
Figure 10
Figure 10. Figure 10: Dataflow of processing differential evolution. [PITH_FULL_IMAGE:figures/full_fig_p007_10.png]
Figure 11
Figure 11. Figure 11: Near-memory sampling architecture. VII. EXPERIMENTS AND ANALYSIS A. Experimental Setup Datasets: Following the experimental setups outlined in the literature [13], [30], [46], we selected a total of six datasets: CIFAR10, CINIC10, SVHN, CIFAR100, ImageNet-100, and Ima…
Figure 12
Figure 12. Figure 12: (a), (b) and (c): The Top-1 accuracy comparison with the SOTA coreset methods in different keeping ratios on complex datasets. (d): End-to-end [PITH_FULL_IMAGE:figures/full_fig_p009_12.png]
Figure 13
Figure 13. Figure 13: The visual results of our improved DE-SNE algorithm in the CIFAR10 dataset with a 10% keeping ratio. [PITH_FULL_IMAGE:figures/full_fig_p010_13.png]
Figure 14
Figure 14. Figure 14: DDR energy comparison of NeSSA, DQ and our NMS. [PITH_FULL_IMAGE:figures/full_fig_p010_14.png]
Figure 16
Figure 16. Figure 16: Normalized energy of AMD 64 Core CPU, NVIDIA 4090 GPU and [PITH_FULL_IMAGE:figures/full_fig_p010_16.png]
Figure 17
Figure 17. Figure 17: Power and area breakdown of proposed DE-SNE accelerator. [PITH_FULL_IMAGE:figures/full_fig_p011_17.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

53 extracted references · 8 linked inside Pith

  1. [2]

    Mask r-cnn,

    K. He, G. Gkioxari, P. Doll ´ar, and R. Girshick, “Mask r-cnn,” in Proceedings of the IEEE international conference on computer vision , 2017, pp. 2961–2969

    2017

  2. [3]

    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 et al. , “An image is worth 16x16 words: Transformers for image recognition at scale,” arXiv preprint arXiv:2010.11929 , 2020

    arXiv 2010

  3. [4]

    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,” in Proceedings of the European conference on computer vision (ECCV) , 2018, pp. 116–131

    2018

  4. [5]

    An energy-and-area-efficient cnn accelerator for universal powers-of- two quantization,

    T. Xia, B. Zhao, J. Ma, G. Fu, W. Zhao, N. Zheng, and P. Ren, “An energy-and-area-efficient cnn accelerator for universal powers-of- two quantization,” IEEE Transactions on Circuits and Systems I: Regular Papers, vol. 70, no. 3, pp. 1242–1255, 2022

    2022

  5. [6]

    Remap: A spatiotemporal cnn accelerator optimization methodology and toolkit thereof,

    B. Zhao, T. Xia, H. Zhai, F. Ma, Y . Du, H. Chang, W. Zhao, and P. Ren, “Remap: A spatiotemporal cnn accelerator optimization methodology and toolkit thereof,” IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems , vol. 42, no. 5, pp. 1691–1704, 2022

    2022

  6. [7]

    Hipu: A hybrid intelligent processing unit with fine-grained isa for real-time deep neural network inference applications,

    W. Zhao, G. Yang, T. Xia, F. Chen, N. Zheng, and P. Ren, “Hipu: A hybrid intelligent processing unit with fine-grained isa for real-time deep neural network inference applications,” IEEE Transactions on Very Large Scale Integration (VLSI) Systems , vol. 31, no. 12, pp. 1980–1993, 2023. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 12

    1980

  7. [8]

    Efficient dataset distillation using random feature approximation,

    N. Loo, R. Hasani, A. Amini, and D. Rus, “Efficient dataset distillation using random feature approximation,” Advances in Neural Information Processing Systems, vol. 35, pp. 13 877–13 891, 2022

    2022

  8. [9]

    Scaling up dataset distillation to imagenet-1k with constant memory,

    J. Cui, R. Wang, S. Si, and C.-J. Hsieh, “Scaling up dataset distillation to imagenet-1k with constant memory,” in International Conference on Machine Learning. PMLR, 2023, pp. 6565–6590

    2023

Show all 53 references
  1. [10]

    Improved distribution matching for dataset condensation,

    G. Zhao, G. Li, Y . Qin, and Y . Yu, “Improved distribution matching for dataset condensation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2023, pp. 7856–7865

    2023

  2. [11]

    Gen- eralizing dataset distillation via deep generative prior,

    G. Cazenavette, T. Wang, A. Torralba, A. A. Efros, and J.-Y . Zhu, “Gen- eralizing dataset distillation via deep generative prior,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2023, pp. 3739–3748

    2023

  3. [12]

    Dataset condensation with distribution match- ing,

    B. Zhao and H. Bilen, “Dataset condensation with distribution match- ing,” in Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , 2023, pp. 6514–6523

    2023

  4. [13]

    Dataset quantization,

    D. Zhou, K. Wang, J. Gu, X. Peng, D. Lian, Y . Zhang, Y . You, and J. Feng, “Dataset quantization,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , 2023, pp. 17 205–17 216

    2023

  5. [14]

    Nonlinear manifold learning in functional magnetic resonance imaging uncovers a low-dimensional space of brain dynamics,

    S. Gao, G. Mishne, and D. Scheinost, “Nonlinear manifold learning in functional magnetic resonance imaging uncovers a low-dimensional space of brain dynamics,” Human brain mapping , vol. 42, no. 14, pp. 4510–4524, 2021

    2021

  6. [15]

    The manifold ways of perception,

    H. S. Seung and D. D. Lee, “The manifold ways of perception,” science, vol. 290, no. 5500, pp. 2268–2269, 2000

    2000

  7. [16]

    Nonlinear dimensionality reduction by locally linear embedding,

    S. T. Rowes, “Nonlinear dimensionality reduction by locally linear embedding,” Science, vol. 290, p. 232, 2000

    2000

  8. [17]

    A global geometric framework for nonlinear dimensionality reduction,

    J. B. Tenenbaum, V . d. Silva, and J. C. Langford, “A global geometric framework for nonlinear dimensionality reduction,” science, vol. 290, no. 5500, pp. 2319–2323, 2000

    2000

  9. [18]

    Decoding brain states on the intrinsic manifold of human brain dynamics across wakefulness and sleep,

    J. Ru ´e-Queralt, A. Stevner, E. Tagliazucchi, H. Laufs, M. L. Kringel- bach, G. Deco, and S. Atasoy, “Decoding brain states on the intrinsic manifold of human brain dynamics across wakefulness and sleep,” Communications biology, vol. 4, no. 1, p. 854, 2021

    2021

  10. [19]

    Visualizing data using t-sne

    L. Van der Maaten and G. Hinton, “Visualizing data using t-sne.” Journal of machine learning research , vol. 9, no. 11, 2008

    2008

  11. [20]

    Dataset distillation,

    T. Wang, J.-Y . Zhu, A. Torralba, and A. A. Efros, “Dataset distillation,” arXiv preprint arXiv:1811.10959 , 2018

    2018 arXiv

  12. [21]

    Remember the past: Distilling datasets into addressable memories for neural networks,

    Z. Deng and O. Russakovsky, “Remember the past: Distilling datasets into addressable memories for neural networks,” Advances in Neural Information Processing Systems , vol. 35, pp. 34 391–34 404, 2022

    2022

  13. [22]

    Dataset condensation with gradient matching,

    B. Zhao, K. R. Mopuri, and H. Bilen, “Dataset condensation with gradient matching,” arXiv preprint arXiv:2006.05929 , 2020

    2006 arXiv

  14. [23]

    Dataset condensation via efficient synthetic-data parame- terization,

    J.-H. Kim, J. Kim, S. J. Oh, S. Yun, H. Song, J. Jeong, J.-W. Ha, and H. O. Song, “Dataset condensation via efficient synthetic-data parame- terization,” in International Conference on Machine Learning . PMLR, 2022, pp. 11 102–11 118

    2022

  15. [24]

    Dream: Efficient dataset distillation by representative matching,

    Y . Liu, J. Gu, K. Wang, Z. Zhu, W. Jiang, and Y . You, “Dream: Efficient dataset distillation by representative matching,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , 2023, pp. 17 314–17 324

    2023

  16. [25]

    Active learning for convolutional neural networks: A core-set approach,

    O. Sener and S. Savarese, “Active learning for convolutional neural networks: A core-set approach,” arXiv preprint arXiv:1708.00489 , 2017

    2017 arXiv

  17. [26]

    Super-samples from kernel herding,

    Y . Chen, M. Welling, and A. Smola, “Super-samples from kernel herding,” arXiv preprint arXiv:1203.3472 , 2012

    2012 arXiv

  18. [27]

    An empirical study of example forgetting during deep neural network learning,

    M. Toneva, A. Sordoni, R. T. d. Combes, A. Trischler, Y . Bengio, and G. J. Gordon, “An empirical study of example forgetting during deep neural network learning,” arXiv preprint arXiv:1812.05159 , 2018

    2018 arXiv

  19. [28]

    Active learning by acquiring contrastive examples,

    K. Margatina, G. Vernikos, L. Barrault, and N. Aletras, “Active learning by acquiring contrastive examples,” arXiv preprint arXiv:2109.03764 , 2021

    2021 arXiv

  20. [29]

    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 , 2016, pp. 770–778

    2016

  21. [30]

    Nessa: Near-storage data selection for accelerated machine learning training,

    N. Prakriya, Y . Yang, B. Mirzasoleiman, C.-J. Hsieh, and J. Cong, “Nessa: Near-storage data selection for accelerated machine learning training,” in Proceedings of the 15th ACM Workshop on Hot Topics in Storage and File Systems , 2023, pp. 8–15

    2023

  22. [31]

    Principal components analysis (pca),

    A. Ma ´ckiewicz and W. Ratajczak, “Principal components analysis (pca),” Computers & Geosciences , vol. 19, no. 3, pp. 303–342, 1993

    1993

  23. [32]

    Linear discriminant analysis,

    P. Xanthopoulos, P. M. Pardalos, T. B. Trafalis, P. Xanthopoulos, P. M. Pardalos, and T. B. Trafalis, “Linear discriminant analysis,” Robust data mining, pp. 27–33, 2013

    2013

  24. [33]

    Development of a multi-dimensional scale for measuring the perceived value of a service,

    J. F. Petrick, “Development of a multi-dimensional scale for measuring the perceived value of a service,” Journal of leisure research , vol. 34, no. 2, pp. 119–134, 2002

    2002

  25. [34]

    Selection of the optimal parameter value for the isomap algorithm,

    O. Samko, A. D. Marshall, and P. L. Rosin, “Selection of the optimal parameter value for the isomap algorithm,” Pattern Recognition Letters , vol. 27, no. 9, pp. 968–979, 2006

    2006

  26. [35]

    Nonlinear dimensionality reduction by locally linear embedding,

    S. T. Roweis and L. K. Saul, “Nonlinear dimensionality reduction by locally linear embedding,” science, vol. 290, no. 5500, pp. 2323–2326, 2000

    2000

  27. [36]

    Laplacian eigenmaps for dimensionality reduction and data representation,

    M. Belkin and P. Niyogi, “Laplacian eigenmaps for dimensionality reduction and data representation,” Neural computation , vol. 15, no. 6, pp. 1373–1396, 2003

    2003

  28. [37]

    Feoktistov, Differential evolution

    V . Feoktistov, Differential evolution. Springer, 2006

    2006

  29. [38]

    Resource-aware distributed dif- ferential evolution for training expensive neural-network-based controller in power electronic circuit,

    X.-F. Liu, Z.-H. Zhan, and J. Zhang, “Resource-aware distributed dif- ferential evolution for training expensive neural-network-based controller in power electronic circuit,” IEEE Transactions on Neural Networks and Learning Systems, vol. 33, no. 11, pp. 6286–6296, 2021

    2021

  30. [39]

    Hardware implemen- tation of multi-objective differential evolution algorithm: a case study of spectrum allocation in cognitive radio networks,

    K. K. Anumandla, R. Peesapati, and S. L. Sabat, “Hardware implemen- tation of multi-objective differential evolution algorithm: a case study of spectrum allocation in cognitive radio networks,” International Journal of Innovative Computing and Applications , vol. 8, no. 4, pp....

    2017

  31. [40]

    Field programmable gate arrays-based differential evolu- tion coprocessor: a case study of spectrum allocation in cognitive radio network,

    K. K. Anumandla, R. Peesapati, S. L. Sabat, S. K. Udgata, and A. Abraham, “Field programmable gate arrays-based differential evolu- tion coprocessor: a case study of spectrum allocation in cognitive radio network,” IET Computers & Digital Techniques , vol. 7, no. 5, pp. 221– 234, 2013

    2013

  32. [41]

    Soc based floating point implementation of differential evolution algorithm using fpga,

    K. K. Anumandla, R. Peesapati, S. L. Sabat, and S. K. Udgata, “Soc based floating point implementation of differential evolution algorithm using fpga,” Design Automation for Embedded Systems, vol. 16, pp. 221– 240, 2012

    2012

  33. [42]

    Performance evalua- tion of floating point differential evolution hardware accelerator on fpga,

    R. Peesapati, K. K. Anumandla, and S. L. Sabat, “Performance evalua- tion of floating point differential evolution hardware accelerator on fpga,” in 2016 IEEE Region 10 Conference (TENCON). IEEE, 2016, pp. 3173– 3178

    2016

  34. [43]

    Solving the nonlinear power flow equa- tions with an inexact newton method using gmres,

    A. J. Flueck and H.-D. Chiang, “Solving the nonlinear power flow equa- tions with an inexact newton method using gmres,” IEEE Transactions on Power Systems, vol. 13, no. 2, pp. 267–273, 1998

    1998

  35. [44]

    Universal chiplet interconnect express (ucie): An open industry standard for innovations with chiplets at package level,

    D. D. Sharma, G. Pasdast, Z. Qian, and K. Aygun, “Universal chiplet interconnect express (ucie): An open industry standard for innovations with chiplets at package level,” IEEE Transactions on Components, Packaging and Manufacturing Technology, vol. 12, no. 9, pp. 1423–1431, 2022

    2022

  36. [45]

    Pci express*(pcie*) 3.0 accelerator features,

    J. Ajanovic, “Pci express*(pcie*) 3.0 accelerator features,” Intel Corpo- ration, vol. 10, pp. 2–2, 2008

    2008

  37. [46]

    Dataset quantization with active learning based adaptive sampling,

    Z. Zhao, Y . Shang, J. Wu, and Y . Yan, “Dataset quantization with active learning based adaptive sampling,” arXiv preprint arXiv:2407.07268 , 2024

    2024 arXiv

  38. [47]

    Mobilenetv2: Inverted residuals and linear bottlenecks,

    M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L.-C. Chen, “Mobilenetv2: Inverted residuals and linear bottlenecks,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 4510–4520

    2018

  39. [48]

    Cacti 7: New tools for interconnect exploration in innovative off-chip memories,

    R. Balasubramonian, A. B. Kahng, N. Muralimanohar, A. Shafiee, and V . Srinivas, “Cacti 7: New tools for interconnect exploration in innovative off-chip memories,” ACM Transactions on Architecture and Code Optimization (TACO), vol. 14, no. 2, pp. 1–25, 2017

    2017

  40. [49]

    Theta: A high-efficiency training accel- erator for dnns with triple-side sparsity exploration,

    J. Lu, J. Huang, and Z. Wang, “Theta: A high-efficiency training accel- erator for dnns with triple-side sparsity exploration,” IEEE Transactions on Very Large Scale Integration (VLSI) Systems, vol. 30, no. 8, pp. 1034– 1046, 2022

    2022

  41. [50]

    Generalized binary search,

    R. Nowak, “Generalized binary search,” in 2008 46th annual Allerton conference on communication, control, and computing . IEEE, 2008, pp. 568–574

    2008

  42. [51]

    Simulated annealing,

    D. Bertsimas and J. Tsitsiklis, “Simulated annealing,” Statistical science, vol. 8, no. 1, pp. 10–15, 1993

    1993

  43. [52]

    Genetic algorithms,

    J. H. Holland, “Genetic algorithms,” Scientific american, vol. 267, no. 1, pp. 66–73, 1992

    1992

  44. [53]

    Particle swarm optimization,

    J. Kennedy and R. Eberhart, “Particle swarm optimization,” in Proceed- ings of ICNN’95-international conference on neural networks , vol. 4. ieee, 1995, pp. 1942–1948

    1995

  45. [54]

    Ant colony optimization,

    M. Dorigo, M. Birattari, and T. Stutzle, “Ant colony optimization,” IEEE computational intelligence magazine , vol. 1, no. 4, pp. 28–39, 2006

    2006

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.