Pith. sign in

REVIEW 6 major objections 7 minor 53 references

SparseMap: A Sparse Tensor Accelerator Framework Based on Evolution Strategy

T0 review · 6 major / 7 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 SparseMap claims that jointly evolving dataflow mapping and sparse-compression strategy finds sparse-tensor accelerator designs with large energy-delay reductions.

desk verdict SparseMap is a genuinely new co-search framework with sensible encodings, but its headline 10x–100x EDP gains rest entirely on one simulator with no silicon check, no seeds, and no code. read the letter →

arxiv 2508.12906 v1 pith:BRR7PRWX submitted 2025-08-18 cs.LG

Boran Zhao , Haiming Zhai , Zihang Yuan , Hetian Liu , Tian Xia , Wenzhe Zhao , Pengju Ren This is my paper · ORCID
classification cs.LG
keywords sparsetensoracceleratordesignspaceexplorationevolutionstrategygeneticencodingmappingoptimizationenergy-delayproductalgebra
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

SparseMap is the paper's answer to a simple problem: when designing a sparse tensor accelerator, the choice of how to tile and order computations (the mapping) and the choice of how to compress and skip zeros (the sparse strategy) are mutually entangled, yet existing design-space explorations optimize only one of them. The paper claims that encoding both choices into a single genome makes the combined space, around $O(10^{41})$ designs for a representative SpMM workload, tractable for an evolution-strategy search. The experimental claim is that the resulting framework reports average energy-delay product reductions of 26.8x, 19.2x, and 171.4x over a sparse-strategy-only baseline and 8.8x, 4.5x, and 158.9x over a mapping-only baseline, on edge, mobile, and cloud hardware budgets. A sympathetic reader would take the contribution to be a demonstration that the joint design space can be searched effectively, provided the analytical cost model it relies on is faithful.

What carries the argument

The load-bearing device is the genome and its two encoding tricks. Prime factors encoding turns each tensor dimension into a sequence of prime factors and assigns each factor to one of the five mapping levels (L1 temporal through L3 spatial), so the constraint $M_1\cdots M_5 = M$ is satisfied automatically and the search never wastes evaluations on invalid tilings. Cantor encoding represents the loop-order permutation at each level by a single integer whose magnitude tracks phenotypic similarity, so small mutations correspond to similar mappings. Around these, the framework wraps a Monte-Carlo sensitivity calibration that identifies high-impact genes, a hypercube-divided initialization, an annealing mutation schedule that shifts from high-sensitivity to low-sensitivity genes, and a crossover that cuts only at segment boundaries to reduce dead offspring. Every fitness call is scored by the analytical cost model.

What would settle it

Take the best SparseMap design and the best baseline design for one workload and platform from the paper's tables, implement both in RTL with the same memory systems and sparse patterns, and compare measured cycles and energy; if the measured EDP gap is much smaller than the reported 10x to 100x, or if the model's error on those points exceeds its claimed single-digit percent range, the central claim fails.

Watch

Extended reading notes

Core claim

The central discovery is that the combined optimization landscape can be flattened into a well-formed search space by construction. The paper encodes every accelerator design as a one-dimensional genome: the mapping half stores loop permutations as Cantor integers and distributes the prime factors of each tensor dimension across five memory-loop levels, so dimension-tiling constraints hold automatically; the sparse-strategy half stores compression formats for each tensor and a skipping/gating code for each memory level and the compute unit. Because every decoded design satisfies the tiling constraints by construction, the evolution strategy spends most of its 20,000-sample budget on valid points, and customized operators, namely high-sensitivity hypercube initialization, annealing mutation, and sensitivity-aware crossover, keep the population diverse and local search effective. On the paper's evaluation model, this consistently yields lower EDP than particle swarm optimization, Monte Carlo tree search, TBPSA, PPO, DQN, and the two single-sided baselines.

Load-bearing premise

Everything depends on the analytical cost model used for sensitivity calibration, fitness scores, and final comparisons predicting real hardware energy and latency accurately for the sparse patterns SparseMap selects.

Editorial extensions

If this is right

  • If the central claim holds, designers can automate the full mapping-plus-sparsity decision for a new workload or new hardware budget, turning a labor-intensive manual step into a search run.
  • The joint search should dominate any pipeline that optimizes mapping and sparse strategy separately, because those pipelines are subsets of the joint space and the encoding keeps valid designs reachable.
  • The encoding techniques, prime-factor tiling and Cantor permutations, could be applied to other design-space searches where constraints are multiplicative or where ordering effects dominate, including dense accelerator mapping searches.
  • Because the paper evaluates edge, mobile, and cloud hardware budgets, the claimed gains are not tied to one platform; the same tool adjusts to both resource-constrained accelerators and large cloud accelerators.

Reading between the lines

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

  • The same genome structure extends naturally to workloads with more than three tensor dimensions, which the paper sketches with a batch-dimension example, so the method should carry over to tensor contractions, batched GEMMs, and attention or MLP sparse patterns without new search machinery.
  • The sensitivity-calibration procedure is a standalone recipe: for any evolutionary hardware search, measuring each gene's fitness variation under Monte-Carlo sampling can identify which variables deserve diversity first and which can be annealed late.
  • The strongest untested assumption is the cost model itself; a natural next experiment is to implement several reported best designs in RTL and measure cycle counts and energy, which would either confirm the 10x to 100x claims or bound the model's error regime.
  • When a dimension is a large prime, the paper pads it to the nearest larger composite number, which changes the tensor shape and could interact with real memory alignment behavior, so prime-sized workloads may need a shape-aware validity check before synthesis.
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

6 major / 7 minor

Summary. The manuscript proposes SparseMap, an evolution-strategy framework that jointly explores mapping and sparse-strategy design spaces for sparse tensor accelerators. It introduces a genome with Cantor-encoded loop permutations and prime-factor-based dimension tiling, a Monte-Carlo sensitivity calibration that identifies high-influence genes, a hypercube initialization, annealing mutation, and sensitivity-aware crossover. The authors evaluate on SpMM and SpConv workloads across edge, mobile, and cloud platform resource constraints using TimeloopV2/Sparseloop as the cost model. They report average EDP reductions of 26.8x, 19.2x, and 171.4x over a SAGE-like baseline and of 8.8x, 4.5x, and 158.9x over Sparseloop Mapper, plus ablations showing that each proposed component improves convergence on two example workloads.

Significance. If the reported simulator results are robust, SparseMap is a useful contribution: it is, to my knowledge, the first framework in this line to co-optimize mapping and sparse strategy, and the prime-factor plus Cantor encoding scheme is a clean way to respect tiling constraints while keeping the genome compact. The paper also provides a constructive ablation showing that each proposed component helps on the two reported workloads. However, the headline claims currently rest on single unrepeated runs and on a single analytical cost model, so the hardware-level significance is not yet established. The framework-level ideas are sound and potentially reproducible in principle, but the evaluation falls short of supporting the abstract's claim that SparseMap 'consistently finds superior solutions.'

major comments (6)
  1. [IV-I, Table IV] All EDP numbers are produced by TimeloopV2/Sparseloop, and the cited 0.1%-8% average error is not validated for the specific mapping and sparse-strategy combinations SparseMap discovers. Because the same cost model is the fitness function, the sensitivity calibrator, and the final comparator, the reported orders-of-magnitude gains could be artifacts of modeling bias rather than real hardware improvements. The authors should either provide an independent check (RTL or FPGA measurements on a few representative designs, or a comparison with a second cycle-accurate or measurement-based model) or explicitly restrict the claims to 'simulator-reported' results and remove the word 'consistently.'
  2. [V, Table IV] No repeated runs, random seeds, or variance estimates are reported for any stochastic algorithm. Evolution strategy, PSO, MCTS, PPO, DQN, and TBPSA are all stochastic, so single-number comparisons cannot support the central claim that SparseMap consistently finds superior solutions. The authors should run multiple seeds, report median and interquartile range (or mean with standard deviation), and use a paired statistical test.
  3. [IV-D, Eqs. (2)-(5)] The sensitivity calibration is a methodological feedback loop: it uses the same cost model that defines the fitness and the final comparison. More importantly, the threshold 3/4, the hypercube count of 100, and the per-hypercube budget of 20 are empirical constants with no sensitivity analysis. Equation (2) also has an unstable normalization, since the denominator |v1-v2|·min{EDP(v1),EDP(v2)} can be zero or produce huge ratios; the paper should justify these choices or replace them with a more robust criterion.
  4. [V.D-V.E, Table IV] The reported 'average EDP reduction' of 26.8x/19.2x/171.4x and 8.8x/4.5x/158.9x appears to be the arithmetic mean of per-workload ratios, which is dominated by a few workloads (for example, mm7 on cloud shows a ratio around 1700x). The authors should report geometric means or medians and provide the full per-workload ratio distribution; otherwise the headline numbers are not statistically meaningful.
  5. [IV.H, V.C] The algorithm description is not complete enough to reproduce the reported results. Population size, parent selection mechanism, offspring count, the base mutation probability for low-sensitivity genes, and the crossover application rate are not specified; only the total budget of 20,000 samples and the annealing schedule are given. Similarly, the 'SAGE-like' and 'Sparseloop Mapper' baseline implementations are described only verbally, without the exact search space or sampling procedure. Pseudocode, hyperparameter tables, or a code release are needed.
  6. [IV.B] Replacing large prime dimensions with the nearest larger composite number changes the workload definition in both tensor sizes and sparsity patterns; the claim that this is 'without materially changing the optimization target' is not supported. For any workload that requires padding, the authors should report the sensitivity of the final design to the padding decision, or explicitly state that the results are for the padded problem only.
minor comments (7)
  1. [I, IV.A] There are typos that should be fixed: 'SpareMap' in Section I, 'Chanllenges' in Section IV.A, and 'acclerators' in Section I.
  2. [Abstract, III.B] Several superscripts are missing or garbled in the PDF text, e.g., 'O(1041)' in the abstract and 'O(1028)'/'O(1013)' in Section III.B; these should be typeset correctly.
  3. [IV.D, Eq. (2)] The notation in Eq. (2) is ambiguous: the summation over v1,v2 in V_i^d suggests all pairs, while the prose says two values are randomly selected; specify the sampling procedure and define the normalization with respect to the number of pairs.
  4. [IV.E] The 'natural boundaries of high-sensitivity gene segments' used for sensitivity-aware crossover are never defined concretely; give an explicit genome-segmentation rule so the operation is reproducible.
  5. [II, IV-I] Reference [18] is cited both for the cost model and for the claimed 0.1%-8% error, but the text calls the model both 'TimeloopV2' and 'Sparseloop'; clarify the relationship between TimeloopV2 and Sparseloop and give the exact source of the error figure.
  6. [Fig. 2] The x-axis labels of Fig. 2 are duplicated ('OS + CSR OS + RLE IS + CSR IS + RLE'); make the legend and tick labels unambiguous.
  7. [V.F] The ablation experiment is limited to two workloads (conv3 and mm3) on a single platform; state this limitation explicitly and, ideally, extend the ablation to at least a few additional workloads across platforms.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the EDP objective is defined externally by TimeloopV2/Sparseloop, and the reported improvements are direct evaluations of that model, not fits to it; the sole self-citation (REMAP) is non-load-bearing.

full rationale

The paper's central claim is that SparseMap finds superior mapping/sparse-strategy designs, with fitness evaluated by TimeloopV2/Sparseloop [18]. The EDP objective is defined externally by that cost model; SparseMap does not fit constants into the model or construct the cost function to reproduce the reported improvements. The high-sensitivity calibration (Eqs. 2-5) uses the same cost model to steer the evolutionary search, but that is algorithmic feedback, not a fitted parameter being renamed as a prediction: the reported Table IV EDP values are direct evaluations of the external model on the discovered designs. The only self-citation is REMAP [23], used in Table I and as an example of a mapping-only exploration framework; it does not justify SparseMap's encoding, operators, or EDP results, so it is not load-bearing. The skeptic's concern that TimeloopV2 may be inaccurate for the discovered sparse-strategy/mapping combinations is a correctness/validation risk about the external benchmark, not circular reasoning under the review rules. Accordingly, no circular step is identified; the score of 2 reflects only the minor non-load-bearing self-citation.

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

The framework's improvements rest on a simulation model and on several hand-tuned hyperparameters. The sensitivity threshold (3/4), number of hypercubes (100), per-hypercube budget (20), total search budget (20,000), and annealing-schedule constants are empirical choices, and no sensitivity analysis shows robustness to them. The TimeloopV2 model is assumed accurate for all designs explored, and the fixed three-level architecture template bounds the design space. Large-prime dimensions are padded to the next composite, which changes the workload and may change sparsity. No invented physical entities are introduced.

free parameters (5)
  • EDP variation sensitivity threshold coefficient = 3/4
    Equations (4) and (5) classify genes as high- or low-sensitivity using a 3/4 threshold between Smin and Smax; the paper says this is empirically derived and validated on the tested cases.
  • Number of hypercubes in high-sensitivity hypercube initialization = 100
    Section IV-D states the number of hypercubes is empirically set to 100, with no robustness analysis reported.
  • Random search budget per hypercube = 20
    Section IV-D sets a budget of 20 random samples per hypercube to find a valid individual.
  • Total search budget = 20,000 samples
    Section V states the total search budget is 20,000 samples based on engineering experience; all comparison results depend on this budget.
  • Annealing mutation schedule maximum probability = 0.8
    Equation (6) uses Ph(g) = 0.8 * e^{-phi} * (1 - phi) with phi = g/G; the 0.8 maximum mutation probability is a hand-chosen constant.
assumptions (3)
  • domain assumption TimeloopV2/Sparseloop EDP model accuracy transfers to all search-discovered designs
    All fitness evaluations, sensitivity estimates, and final comparisons use this analytical model; the paper repeats the claimed 0.1% to 8% average error from [18] but does not verify it for the specific designs SparseMap produces. See Section IV-I and Section V.
  • domain assumption The fixed three-level memory hierarchy with five mapping levels spans relevant sparse tensor accelerator designs
    The genome encodes L1_T, L2_T, L2_S, L3_T, and L3_S loops on a DRAM-GLB-PE buffer-MAC architecture only; accelerators with different hierarchies or loop structures are outside the search space. See Section III-A.
  • ad hoc to paper Prime dimensions are padded to the nearest larger composite number without materially changing the optimization target
    Section IV-B replaces large prime dimensions with the nearest larger composite to enable factorization. Padding changes tensor dimensions and the zero distribution, so the EDP reported may be for a workload different from the one named.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SparseMap: A Sparse Tensor Accelerator Framework Based on Evolution Strategy." pith.science (2026). https://pith.science/paper/BRR7PRWX

@misc{pith2026250812906,
  author       = {Pith},
  title        = {Pith review of: SparseMap: A Sparse Tensor Accelerator Framework Based on Evolution Strategy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BRR7PRWX}},
  note         = {Machine review of arXiv:2508.12906}
}
abstract

The growing demand for sparse tensor algebra (SpTA) in machine learning and big data has driven the development of various sparse tensor accelerators. However, most existing manually designed accelerators are limited to specific scenarios, and it's time-consuming and challenging to adjust a large number of design factors when scenarios change. Therefore, automating the design of SpTA accelerators is crucial. Nevertheless, previous works focus solely on either mapping (i.e., tiling communication and computation in space and time) or sparse strategy (i.e., bypassing zero elements for efficiency), leading to suboptimal designs due to the lack of comprehensive consideration of both. A unified framework that jointly optimizes both is urgently needed. However, integrating mapping and sparse strategies leads to a combinatorial explosion in the design space(e.g., as large as $O(10^{41})$ for the workload $P_{32 \times 64} \times Q_{64 \times 48} = Z_{32 \times 48}$). This vast search space renders most conventional optimization methods (e.g., particle swarm optimization, reinforcement learning and Monte Carlo tree search) inefficient. To address this challenge, we propose an evolution strategy-based sparse tensor accelerator optimization framework, called SparseMap. SparseMap constructing a more comprehensive design space with the consideration of both mapping and sparse strategy. We introduce a series of enhancements to genetic encoding and evolutionary operators, enabling SparseMap to efficiently explore the vast and diverse design space. We quantitatively compare SparseMap with prior works and classical optimization methods, demonstrating that SparseMap consistently finds superior solutions.

Figures

Figures reproduced from arXiv: 2508.12906 by the authors.

Figure 1
Figure 1. Different sparse strategies prefer different mappings. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Impact of different mappings (OS represents output stationary and IS indicates input stationary) and sparse strategies (CSR represents compressed sparse row and RLE indicates run length encoding) on normalized latency and energy. The left plot shows normalized latency, while the right plot depicts normalized energy across varying sparsity levels. It can be observed that no single sparse strategy is optimal for all w… view at source ↗
Figure 3
Figure 3. (a) A basic sparse tensor accelerator architecture with 3-level memory. [PITH_FULL_IMAGE:figures/full_fig_p002_3.png] view at source ↗
Figures from the paper (13 more)
Figure 4
Figure 4. Figure 4: This figure demonstrates the complete mapping of a matrix multi [PITH_FULL_IMAGE:figures/full_fig_p003_4.png]
Figure 5
Figure 5. Figure 5: Hierarchical compression schemes for sparse data representation. The figure illustrates 1D, 2D, and tiled 2D (3D) data structures. Each level [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]
Figure 6
Figure 6. Figure 6: Illustration of the skipping/gating mechanism within a three-level [PITH_FULL_IMAGE:figures/full_fig_p004_6.png]
Figure 9
Figure 9. Figure 9: Illustration of prime factors encoding for [PITH_FULL_IMAGE:figures/full_fig_p006_9.png]
Figure 10
Figure 10. Figure 10: (a) Cantor encoding and (b) random encoding for permutations of mapping. (c) Convergence curve of different encoding. The workload is SpMM with ID: mm3. The hardware platform is cloud and the optimization objective is EDP. decreases. Correspondingly, in accelerator ma…
Figure 11
Figure 11. Figure 11: Illustration of high-sensitivity hypercube initialization in our work. The design space is divided into several quadrants, with high-sensitivity segment genes (Gene A/B/C) used as coordinate axes (only three coordinate axes are shown for clarity). Mapping Sp a rs e S …
Figure 12
Figure 12. Figure 12: (a) Design space without cantor encoding, where the fitness variation [PITH_FULL_IMAGE:figures/full_fig_p008_12.png]
Figure 13
Figure 13. Figure 13: At the top is the genome of an individual. Permutations are [PITH_FULL_IMAGE:figures/full_fig_p009_13.png]
Figure 14
Figure 14. Figure 14: This figure presents the hardware accelerator design corresponding to the mapping and sparse strategy in Fig. 13. The right subfigure depicts the [PITH_FULL_IMAGE:figures/full_fig_p009_14.png]
Figure 15
Figure 15. Figure 15: An example of SparseMap supporting a SpTA workload involving [PITH_FULL_IMAGE:figures/full_fig_p010_15.png]
Figure 16
Figure 16. Figure 16: Overview of the SparseMap framework. The framework com [PITH_FULL_IMAGE:figures/full_fig_p010_16.png]
Figure 17
Figure 17. Figure 17: (a) Comparison of the search results between our work and various [PITH_FULL_IMAGE:figures/full_fig_p011_17.png]
Figure 18
Figure 18. Figure 18: Convergence curves of different evolution strategies under different [PITH_FULL_IMAGE:figures/full_fig_p012_18.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

53 extracted references · 1 linked inside Pith

  1. [1]

    Tensorflow: a system for large-scale machine learning,

    M. Abadi, P. Barham, J. Chen, Z. Chen, A. Davis, J. Dean, M. Devin, S. Ghemawat, G. Irving, M. Isard et al. , “Tensorflow: a system for large-scale machine learning,” in 12th USENIX symposium on operating systems design and implementation (OSDI 16) , 2016, pp. 265–283

    2016

  2. [2]

    Pytorch: An imperative style, high-performance deep learning library,

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga et al. , “Pytorch: An imperative style, high-performance deep learning library,” Advances in neural information processing systems , vol. 32, 2019

    2019

  3. [3]

    A survey of recommendation systems: recommendation models, techniques, and application fields,

    H. Ko, S. Lee, Y . Park, and A. Choi, “A survey of recommendation systems: recommendation models, techniques, and application fields,” Electronics, vol. 11, no. 1, p. 141, 2022

    2022

  4. [4]

    Tiirec: A tensor approach for tag-driven item recommendation with sparse user generated content,

    L. Yu, J. Huang, G. Zhou, C. Liu, and Z.-K. Zhang, “Tiirec: A tensor approach for tag-driven item recommendation with sparse user generated content,” Information Sciences, vol. 411, pp. 122–135, 2017

    2017

  5. [5]

    Hierarchical analysis of power distribution networks,

    M. Zhao, R. V . Panda, S. S. Sapatnekar, T. Edwards, R. Chaudhry, and D. Blaauw, “Hierarchical analysis of power distribution networks,” in Proceedings of the 37th Annual Design Automation Conference , 2000, pp. 150–155

    2000

  6. [6]

    Feasta: A flexible and efficient accelerator for sparse tensor algebra in machine learning,

    K. Zhong, Z. Zhu, G. Dai, H. Wang, X. Yang, H. Zhang, J. Si, Q. Mao, S. Zeng, K. Hong et al. , “Feasta: A flexible and efficient accelerator for sparse tensor algebra in machine learning,” in Proceedings of the 29th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 3 , 2024, pp. 349–366

    2024

  7. [7]

    Eyeriss: An energy- efficient reconfigurable accelerator for deep convolutional neural net- works,

    Y . H. Chen, T. Krishna, J. S. Emer, and V . Sze, “Eyeriss: An energy- efficient reconfigurable accelerator for deep convolutional neural net- works,” Solid-state Circuits Conference , 2016

    2016

  8. [8]

    Cnvlutin: Ineffectual-neuron-free deep neural network computing,

    J. Albericio, P. Judd, T. Hetherington, T. Aamodt, N. E. Jerger, and A. Moshovos, “Cnvlutin: Ineffectual-neuron-free deep neural network computing,” ACM SIGARCH Computer Architecture News , vol. 44, no. 3, pp. 1–13, 2016

    2016

Show all 53 references
  1. [9]

    Eyeriss v2: A flexible accelerator for emerging deep neural networks on mobile devices,

    Y .-H. Chen, T.-J. Yang, J. Emer, and V . Sze, “Eyeriss v2: A flexible accelerator for emerging deep neural networks on mobile devices,” IEEE Journal on Emerging and Selected Topics in Circuits and Systems, vol. 9, no. 2, pp. 292–308, 2019

    2019

  2. [10]

    Extensor: An accelerator for sparse tensor algebra,

    K. Hegde, H. Asghari-Moghaddam, M. Pellauer, N. Crago, A. Jaleel, E. Solomonik, J. Emer, and C. W. Fletcher, “Extensor: An accelerator for sparse tensor algebra,” in Proceedings of the 52nd Annual IEEE/ACM International Symposium on Microarchitecture , 2019, pp. 319–333

    2019

  3. [11]

    Matraptor: A sparse-sparse matrix multiplication accelerator based on row-wise product,

    N. Srivastava, H. Jin, J. Liu, D. Albonesi, and Z. Zhang, “Matraptor: A sparse-sparse matrix multiplication accelerator based on row-wise product,” in 2020 53rd Annual IEEE/ACM International Symposium on Microarchitecture (MICRO). IEEE, 2020, pp. 766–780

    2020

  4. [12]

    Gospa: An energy- efficient high-performance globally optimized sparse convolutional neu- ral network accelerator,

    C. Deng, Y . Sui, S. Liao, X. Qian, and B. Yuan, “Gospa: An energy- efficient high-performance globally optimized sparse convolutional neu- ral network accelerator,” in 2021 ACM/IEEE 48th Annual International Symposium on Computer Architecture (ISCA) . IEEE, 2021, pp. 1110– 1123

    2021

  5. [13]

    Outerspace: An outer product based sparse matrix multiplication accelerator,

    S. Pal, J. Beaumont, D.-H. Park, A. Amarnath, S. Feng, C. Chakrabarti, H.-S. Kim, D. Blaauw, T. Mudge, and R. Dreslinski, “Outerspace: An outer product based sparse matrix multiplication accelerator,” in 2018 IEEE International Symposium on High Performance Computer Architectu...

    2018

  6. [14]

    Scnn: An accelerator for compressed-sparse convolutional neural networks,

    A. Parashar, M. Rhu, A. Mukkara, A. Puglielli, R. Venkatesan, JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 14 B. Khailany, J. Emer, S. W. Keckler, and W. J. Dally, “Scnn: An accelerator for compressed-sparse convolutional neural networks,” ACM SIGARCH computer arc...

    2021

  7. [15]

    Sigma: A sparse and irregular gemm ac- celerator with flexible interconnects for dnn training,

    E. Qin, A. Samajdar, H. Kwon, V . Nadella, S. Srinivasan, D. Das, B. Kaul, and T. Krishna, “Sigma: A sparse and irregular gemm ac- celerator with flexible interconnects for dnn training,” in 2020 IEEE International Symposium on High Performance Computer Architecture (HPCA). IE...

    2020

  8. [16]

    Sparten: A sparse tensor accelerator for convolutional neural networks,

    A. Gondimalla, N. Chesnut, M. Thottethodi, and T. Vijaykumar, “Sparten: A sparse tensor accelerator for convolutional neural networks,” in Proceedings of the 52nd Annual IEEE/ACM International Symposium on Microarchitecture, 2019, pp. 151–165

    2019

  9. [17]

    Sparsity-aware and re-configurable npu architecture for samsung flagship mobile soc,

    J.-W. Jang, S. Lee, D. Kim, H. Park, A. S. Ardestani, Y . Choi, C. Kim, Y . Kim, H. Yu, H. Abdel-Azizet al., “Sparsity-aware and re-configurable npu architecture for samsung flagship mobile soc,” in 2021 ACM/IEEE 48th Annual International Symposium on Computer Architecture (IS...

    2021

  10. [18]

    Sparseloop: An analytical, energy-focused design space exploration methodology for sparse tensor accelerators,

    Y . N. Wu, P. A. Tsai, A. Parashar, V . Sze, and J. S. Emer, “Sparseloop: An analytical, energy-focused design space exploration methodology for sparse tensor accelerators,” in 2021 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS) , 2021

    2021

  11. [19]

    NVDLA Deep Learning Accelerator,

    NVIDIA, “NVDLA Deep Learning Accelerator,” NVIDIA Corporation, Tech. Rep., 2020, white Paper. [Online]. Available: http://nvdla.org

    2020

  12. [20]

    Nvidia a100 tensor core gpu architecture,

    NVIDIA, “Nvidia a100 tensor core gpu architecture,” NVIDIA Corporation, Tech. Rep., 2020, white Paper. [Online]. Available: https://resources.nvidia.com/en-us-tensor-core/ nvidia-ampere-architecture-whitepaper

    2020

  13. [21]

    A low-power general matrix multiplication acceler- ator with sparse weight-and-output stationary dataflow,

    P. Liu and Y . Wang, “A low-power general matrix multiplication acceler- ator with sparse weight-and-output stationary dataflow,” Micromachines, vol. 16, no. 1, p. 101, 2025

    2025

  14. [22]

    Sparch: Efficient architecture for sparse matrix multiplication,

    Z. Zhang, H. Wang, S. Han, and W. J. Dally, “Sparch: Efficient architecture for sparse matrix multiplication,” in 2020 IEEE Interna- tional Symposium on High Performance Computer Architecture (HPCA). IEEE, 2020, pp. 261–274

    2020

  15. [23]

    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: A publication of the IEEE Circuits and...

    2023

  16. [24]

    Gamma: automating the hw mapping of dnn models on accelerators via genetic algorithm,

    T. Krishna and S. Kao, “Gamma: automating the hw mapping of dnn models on accelerators via genetic algorithm,” in ICCAD ’20: IEEE/ACM International Conference on Computer-Aided Design , 2020

    2020

  17. [25]

    Confuciux: Autonomous hardware resource assignment for dnn accelerators using reinforcement learning,

    S.-C. Kao, G. Jeong, and T. Krishna, “Confuciux: Autonomous hardware resource assignment for dnn accelerators using reinforcement learning,” in 2020 53rd Annual IEEE/ACM International Symposium on Microar- chitecture (MICRO). IEEE, 2020, pp. 622–636

    2020

  18. [26]

    Towards an auto- tuning system design for optimal sparse compression format selection with user expertise,

    I. Mehrez, O. Hamdi-Larbi, T. Dufaud, and N. Emad, “Towards an auto- tuning system design for optimal sparse compression format selection with user expertise,” in 2016 IEEE/ACS 13th International Conference of Computer Systems and Applications (AICCSA) . IEEE, 2016, pp. 1–6

    2016

  19. [27]

    Cdpu: Co-designing compression and decompression processing units for hyperscale sys- tems,

    S. Karandikar, A. N. Udipi, J. Choi, J. Whangbo, J. Zhao, S. Kanev, E. Lim, J. Alakuijala, V . Madduri, Y . S. Shaoet al., “Cdpu: Co-designing compression and decompression processing units for hyperscale sys- tems,” in Proceedings of the 50th Annual International Symposium on...

    2023

  20. [28]

    Extending sparse tensor accelerators to support multiple compression formats,

    E. Qin, G. Jeong, W. Won, S.-C. Kao, H. Kwon, S. Srinivasan, D. Das, G. E. Moon, S. Rajamanickam, and T. Krishna, “Extending sparse tensor accelerators to support multiple compression formats,” in 2021 IEEE International Parallel and Distributed Processing Symposium (IPDPS) . ...

    2021

  21. [29]

    Medea: A multi-objective evolutionary approach to dnn hardware mapping,

    E. Russo, M. Palesi, S. Monteleone, D. Patti, G. Ascia, and V . Catania, “Medea: A multi-objective evolutionary approach to dnn hardware mapping,” in 2022 Design, Automation & Test in Europe Conference & Exhibition (DATE) . IEEE, 2022, pp. 226–231

    2022

  22. [30]

    Learning by playing solving sparse reward tasks from scratch,

    M. Riedmiller, R. Hafner, T. Lampe, M. Neunert, J. Degrave, T. Wiele, V . Mnih, N. Heess, and J. T. Springenberg, “Learning by playing solving sparse reward tasks from scratch,” in International conference on machine learning . PMLR, 2018, pp. 4344–4353

    2018

  23. [31]

    Dealing with sparse rewards in reinforcement learning,

    J. Hare, “Dealing with sparse rewards in reinforcement learning,” arXiv preprint arXiv:1910.09281, 2019

    1910 arXiv

  24. [32]

    Dual- side sparse tensor core,

    Y . Wang, C. Zhang, Z. Xie, C. Guo, Y . Liu, and J. Leng, “Dual- side sparse tensor core,” in 2021 ACM/IEEE 48th Annual International Symposium on Computer Architecture (ISCA) . IEEE, 2021, pp. 1083– 1095

    2021

  25. [33]

    An overview of data compression techniques,

    S. B. Wright, “An overview of data compression techniques,” Ph.D. dissertation, University of Washington, 1989

    1989

  26. [34]

    [Online]

    (2016). [Online]. Available: https://github.com/baidu-research/ deepbench

    2016

  27. [35]

    Integrated particle swarm optimization (i-pso): An adaptive design space exploration framework for power- performance tradeoff in architectural synthesis,

    A. Sengupta and V . K. Mishra, “Integrated particle swarm optimization (i-pso): An adaptive design space exploration framework for power- performance tradeoff in architectural synthesis,” in Fifteenth Interna- tional Symposium on Quality Electronic Design . IEEE, 2014, pp. 60– 67

    2014

  28. [36]

    Autonomous design of noise- mitigating structures using deep reinforcement learning,

    S. B. Gebrekidan and S. Marburg, “Autonomous design of noise- mitigating structures using deep reinforcement learning,” The Journal of the Acoustical Society of America , vol. 156, no. 1, pp. 151–163, 2024

    2024

  29. [37]

    A fast and elitist multiobjective genetic algorithm: Nsga-ii,

    K. Deb, A. Pratap, S. Agarwal, and T. Meyarivan, “A fast and elitist multiobjective genetic algorithm: Nsga-ii,” IEEE transactions on evolu- tionary computation, vol. 6, no. 2, pp. 182–197, 2002

    2002

  30. [38]

    Automated docking using a lamarckian genetic algorithm and an empirical binding free energy function,

    G. M. Morris, D. S. Goodsell, R. S. Halliday, R. Huey, W. E. Hart, R. K. Belew, and A. J. Olson, “Automated docking using a lamarckian genetic algorithm and an empirical binding free energy function,” Journal of computational chemistry, vol. 19, no. 14, pp. 1639–1662, 1998

    1998

  31. [39]

    Develop- ment and validation of a genetic algorithm for flexible docking,

    G. Jones, P. Willett, R. C. Glen, A. R. Leach, and R. Taylor, “Develop- ment and validation of a genetic algorithm for flexible docking,” Journal of molecular biology , vol. 267, no. 3, pp. 727–748, 1997

    1997

  32. [40]

    Multiobjective optimization using nondominated sorting in genetic algorithms,

    J. Schaffer, “Multiobjective optimization using nondominated sorting in genetic algorithms,” in Proceedings of the First International Conference on Genetic Algorithms and Their Applications . Lawrence Erlbaum Associates, 1985, pp. 160–168

    1985

  33. [41]

    Genetic algorithms in search, optimization and machine learning,

    G. Zames, “Genetic algorithms in search, optimization and machine learning,” Inf Tech J, vol. 3, no. 1, p. 301, 1981

    1981

  34. [42]

    Mitchell, An introduction to genetic algorithms

    M. Mitchell, An introduction to genetic algorithms . MIT press, 1998

    1998

  35. [43]

    Gene expression programming: A survey,

    J. Zhong, L. Feng, and Y .-S. Ong, “Gene expression programming: A survey,” IEEE Computational Intelligence Magazine , vol. 12, no. 3, pp. 54–72, 2017

    2017

  36. [44]

    A review of population initial- ization techniques for evolutionary algorithms,

    B. Kazimipour, X. Li, and A. K. Qin, “A review of population initial- ization techniques for evolutionary algorithms,” in 2014 IEEE congress on evolutionary computation (CEC) . IEEE, 2014, pp. 2585–2592

    2014

  37. [45]

    Initialization strategies and diversity in evolutionary timetabling,

    E. K. Burke, J. P. Newall, and R. F. Weare, “Initialization strategies and diversity in evolutionary timetabling,” Evolutionary computation, vol. 6, no. 1, pp. 81–103, 1998

    1998

  38. [46]

    Natural evolutionary strategies for variational quantum computation,

    A. Anand, M. Degroote, and A. Aspuru-Guzik, “Natural evolutionary strategies for variational quantum computation,” Machine Learning: Science and Technology, vol. 2, no. 4, p. 045012, 2021

    2021

  39. [47]

    An advanced initialization technique for metaheuristic optimization: a fusion of latin hypercube sampling and evolutionary behaviors,

    H. Escobar-Cuevas, E. Cuevas, K. Avila, and O. Avalos, “An advanced initialization technique for metaheuristic optimization: a fusion of latin hypercube sampling and evolutionary behaviors,” Computational and Applied Mathematics, vol. 43, no. 4, p. 234, 2024

    2024

  40. [48]

    Latin hypercube initialization strategy for design space exploration of deep neural network architectures,

    H. R. Medeiros, D. M. Izidio, A. P. d. A. Ferreira, and E. N. da S. Barros, “Latin hypercube initialization strategy for design space exploration of deep neural network architectures,” in Proceedings of the Genetic and Evolutionary Computation Conference Companion , 2019, pp. 295–296

    2019

  41. [49]

    The initialization of evolutionary multi- objective optimization algorithms,

    M. Hamdan and O. Qudah, “The initialization of evolutionary multi- objective optimization algorithms,” in Advances in Swarm and Com- putational Intelligence: 6th International Conference, ICSI 2015, held in conjunction with the Second BRICS Congress, CCI 2015, Beijing, China, ...

    2015

  42. [50]

    A multistrategy differential evolution algorithm combined with latin hypercube sampling applied to a brain–computer interface to improve the effect of node displacement,

    H. Chang, Y . Sun, S. Lu, and D. Lin, “A multistrategy differential evolution algorithm combined with latin hypercube sampling applied to a brain–computer interface to improve the effect of node displacement,” Scientific Reports, vol. 14, no. 1, p. 20420, 2024

    2024

  43. [51]

    R. Y . Rubinstein and D. P. Kroese, Simulation and the Monte Carlo method. John Wiley & Sons, 2016

    2016

  44. [52]

    In-datacenter performance analysis of a tensor processing unit,

    N. P. Jouppi, C. Young, N. Patil, D. Patterson, G. Agrawal, R. Bajwa, S. Bates, S. Bhatia, N. Boden, A. Borchers et al. , “In-datacenter performance analysis of a tensor processing unit,” in Proceedings of the 44th annual international symposium on computer architecture , 2017...

    2017

  45. [53]

    Sparsegpt: Massive language models can be accurately pruned in one-shot,

    E. Frantar and D. Alistarh, “Sparsegpt: Massive language models can be accurately pruned in one-shot,” in International Conference on Machine Learning. PMLR, 2023, pp. 10 323–10 337

    2023

Pith tools

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