Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

Rec-AD: An Efficient Computation Framework for FDIA Detection Based on Tensor Train Decomposition and Deep Learning Recommendation Model

T0 review · 4 major / 6 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 Replacing dense embedding rows with tensor-train cores compresses DLRM embedding memory by 5-74x and speeds training up to 3x at negligible accuracy cost.

desk verdict A real engineering core with a self-inflicted wound: the headline 3x speedup is contradicted by the paper's own tables. read the letter →

arxiv 2507.14668 v2 pith:AKXENOMY submitted 2025-07-19 cs.LG

Yunfeng Li , Junhong Liu , Zhaohui Yang , Guofu Liao , Chuyun Zhang This is my paper
classification cs.LG
keywords falsedatainjectionattackdetectiontensortraindecompositiondeeplearningrecommendationmodelembeddingcompressionindexreorderingpipelinetrainingsmartgridsecurity
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper sets out to show that false data injection attack detection, which normally relies on large deep models with huge embedding tables, can be made fast enough for real-time and edge use by compressing the embedding layer into a tensor-train decomposition and by reordering sparse indices to maximize reuse of shared intermediate products. It claims that this drop-in replacement, plus a pipeline that overlaps host-memory fetches with GPU computation, shrinks embedding memory by 5 to 74 times and speeds training up to 3 times while keeping accuracy within 0.1 percent on large click datasets and even improving FDIA accuracy on a 118-bus power system benchmark. If correct, the result matters because it gives resource-limited substations a practical path to millisecond-level attack response without abandoning standard deep-learning training stacks.

What carries the argument

The load-bearing mechanism is the Eff-TT embedding table: an embedding matrix is reshaped into a high-dimensional tensor and factored into a chain of small TT cores, so an embedding row is reconstructed by multiplying slices of the cores. Because real training data follows a power-law index distribution with local co-occurrence, many indices in a batch share TT-index prefixes; a Reuse Buffer stores the intermediate products of the first two cores, and a parallel pointer-preparation kernel feeds the reusable products into batched GEMMs. Around this core, the paper builds three supporting mechanisms: an index bijection that maps frequently co-occurring indices into nearby positions using modularity-based community detection on a batch co-occurrence graph; a backward pass that aggregates duplicate gradients before multiplying with TT cores and fuses the core update; and a pipeline training system with a GPU embedding cache that prefetches parameters from host memory while overlapping MLP computation.

What would settle it

Take the same Rec-AD pipeline and the same model, but flatten the training data's index distribution so every embedding is accessed equally often and randomly permute indices within each batch to destroy local co-occurrence; if Rec-AD's per-batch training time is not close to the dense DLRM baseline and most of the speedup persists, then the reuse/reordering mechanism is not the source of the gains; if the speedup largely vanishes, the skewness assumption is confirmed as load-bearing.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that the embedding table in a deep recommendation model can be replaced by an 'Efficient Tensor Train' (Eff-TT) table without changing the model's interface, and that this replacement is not merely a memory-saving trick but a computational accelerator when the index distribution is skewed. The key quantitative claims are a 74x memory compression on the largest benchmark, an average 3x end-to-end training speedup over the unmodified recommendation model on a V100 GPU, a 1.3-1.4x speedup over the tensor-train baseline and the host-memory baseline, and FDIA detection at 97.5% accuracy and 96.2% recall versus 94.1% and 92.2% for the unmodified model.

Load-bearing premise

The speedups depend on the training indices being highly skewed and locally clustered—a few embeddings dominate, and related embeddings tend to appear in the same batches—so tensor-train indices share prefixes and intermediate results can be reused; if a workload has uniform or scattered index access, the reuse and reordering gains shrink and only the pipeline savings remain.

Editorial extensions

If this is right

  • Embedding tables that exceed GPU memory (the paper constructs a 19 GB table) can be trained on a single GPU instead of sharded across devices, removing inter-GPU synchronization from the critical path.
  • FDIA detection latency on a streaming single-sample workload drops from 25 ms to 21.5 ms and throughput rises about 16%, shrinking the window in which an attacker can corrupt readings before detection.
  • Because the compressed embedding uses the standard embedding-bag interface, existing training pipelines can adopt it by swapping the embedding layer, with no changes to the surrounding model code.
  • Tables with more rows benefit more from index reordering, so the speedup should grow with industrial-scale data rather than flattening.

Reading between the lines

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

  • We infer that the same compression plus reuse recipe would transfer to other sparse high-dimensional detection tasks whose categorical features have a similar skewed co-occurrence structure, such as network intrusion detection or sensor fault diagnosis.
  • We infer that the efficiency gains are workload-dependent: under near-uniform index access, the paper's own complexity analysis suggests TT lookup would cost more than a dense table, so the practical win is conditional on the power-law assumption.
  • We infer that the reported 3.4-point FDIA accuracy improvement over the unmodified model is not yet established as a general property of tensor-train regularization, since the paper gives no error bars or run counts; a controlled repeated-seed study would be needed.
  • A testable extension is to sample a training batch's index distribution and compute the expected prefix-sharing rate before adopting Rec-AD, letting an operator predict speedup without running full training.
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 / 6 minor

Summary. This paper proposes Rec-AD, a framework that combines tensor-train (TT) decomposition with the deep learning recommendation model (DLRM) for false data injection attack (FDIA) detection. It introduces an Eff-TT embedding table, index reordering based on local/global structure, a reuse buffer for intermediate TT products, and a pipelined training mechanism with host-memory expansion. Experiments on Avazu, Criteo Terabyte, Criteo Kaggle, and IEEE118-Bus compare Rec-AD against DLRM, TT-Rec, FAE, HugeCTR, and TorchRec, reporting 5.33–74.19× embedding memory compression, up to 3× training speedup, and FDIA detection accuracy of 97.5%. The claimed contribution is a drop-in PyTorch-compatible embedding replacement that enables large DLRM training and real-time FDIA detection on resource-constrained hardware.

Significance. If the efficiency claims hold, Rec-AD would be a practically useful systems contribution: a single-GPU-compatible TT embedding with optimized kernels and pipelining that preserves DLRM accuracy within 0.1% on standard recommendation benchmarks. The paper's strengths include a reasonably honest complexity accounting in Sec. III.B, a clean ablation study in Fig. 12, and a broad baseline selection. The central efficiency claim, however, is currently not internally consistent, and the FDIA detection result rests on a small, single-split evaluation without uncertainty quantification. These issues are fixable within the manuscript's scope, but they must be addressed before the central claims can be accepted.

major comments (4)
  1. [Sec. V.H, Fig. 10] The headline claim of an average 3× speedup over DLRM is contradicted by the paper's own numerical results. Table III implies 1.35× (1 GPU) and 1.61× (4 GPU) speedups on IEEE118-Bus; Sec. V.H's own text later states that on the 1 GPU setting, "DLRM slightly outperforms Rec-AD in raw throughput"; and Fig. 14 reports an average 2.44× speedup for the pipeline variant. Fig. 10 is presented only as a bar chart with no numeric table, so the 3× figure cannot be audited. Please provide per-dataset, per-hardware numeric results, reconcile them with Table III and Sec. V.H, and specify exactly what "average" is computed over.
  2. [Sec. III.B, Algorithm 1, and Table IV] The efficiency mechanism is premised on power-law, locally structured index distributions, but the paper never reports the measured reuse rate, the values of R and n used in the experiments, or the sensitivity of the speedup to index distribution. The complexity expression O_TT = O(k n^2 R^2) with "R >> n" is asserted without supporting ranges; without actual TT ranks, Table IV's compression ratios and the claimed speedups cannot be verified. Please report TT ranks, core dimensions, and either measured reuse rates or a sensitivity analysis under near-uniform index access.
  3. [Table III and Secs. V.F–V.G] The FDIA detection advantage (97.5% accuracy and 96.2% recall versus 94.1% and 92.2% for DLRM) is presented without error bars, number of repeated runs, or statistical significance testing. On a dataset of 24,800 samples, a 3.4-point accuracy difference may be within run-to-run noise for deep models. Please provide repeated-run statistics and, ideally, comparison with FDIA-specific detectors to support the claim that TT compression improves detection robustness rather than merely fitting this particular split.
  4. [Sec. V.I and Fig. 13] The large-embedding-table experiment reports only relative speedups (1.35× over TorchRec, 1.07× over HugeCTR) and does not state whether the DLRM baseline was included in this setup; further, the conclusion that Rec-AD "enables the training of ultra-large embeddings within a single GPU" is not demonstrated by a memory-footprint curve or a comparison of GPU memory usage. Please add absolute throughputs, memory measurements, and a clear statement of which baselines were run on which GPU count.
minor comments (6)
  1. [Eq. (1)] The notation W(i1, · · ·, ik, id) is inconsistent and the text refers to D(d) as the k-th core; please standardize the dimension and core indexing.
  2. [Algorithm 1] The variable name "Bufe" appears repeatedly and should be corrected to "Buf".
  3. [Algorithm 2, line 8] The loss-function remark in line 8 is unrelated to the graph-generation algorithm and is grammatically incomplete; move this information to the experimental setup section.
  4. [Sec. V.D and Table VI] Table VI reports results on an RTX 2060, but Sec. V.D only describes V100 and T4 platforms; please add the RTX 2060 setup and describe how the 100MB dataset was constructed.
  5. [Sec. V.H] The statement that "the TT embedding dimension is set to 128 for V100 and 64 for T4" is not justified; please state whether these values were tuned and how they affect accuracy and speed.
  6. [Sec. V.K] The ablation text says disabling "forward-pass gradient aggregation" causes a 52% throughput drop, but the surrounding discussion describes backward-pass gradient aggregation; please correct the terminology.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation; the efficiency claims rest on standard TT algebra and external baselines, though the reported 3x speedup is internally inconsistent with Table III.

full rationale

The paper's derivation chain is not circular. Section II.B quotes the tensor-train decomposition (Eqs. 1-2) from Oseledets [34] and the embedding-table factorization from prior TT literature; Section III.B gives an explicit complexity comparison in which the uncompressed TT lookup (O(k n^2 R^2)) is admittedly more expensive than the dense lookup (O(k n^3)), so the claimed efficiency gain is not achieved by definition but depends on the reuse and reordering mechanisms. The reuse mechanisms (Algorithm 1, index reordering, gradient aggregation) are engineering heuristics whose payoff rests on the empirical power-law and co-occurrence premises stated in Section II.C; if those premises fail the speedup shrinks, which is a correctness risk, not a circularity. No fitted parameter is renamed as a prediction: Table V reports test accuracy from independently trained baselines, and Table III compares Rec-AD against external DLRM and TT-Rec baselines. The only overlapping-author citation is [47], used to describe the IEEE118-Bus dataset configuration; it is a data-preprocessing reference, not a load-bearing theorem, and it does not constrain the possible design choices. The headline 'average speedup of 3x' in Section V.H conflicts with Table III (DLRM normalized to 1.00 vs Rec-AD 0.74/0.62) and with Section V.I's statement that DLRM slightly outperforms Rec-AD in single-GPU raw throughput; this is an internal inconsistency in the evaluation, not a circular step. Score 2 reflects the rubric's allowance for one minor, non-load-bearing self-citation; no circular step was found.

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

The central claims rest on standard tensor-train mathematics and the DLRM architecture, plus domain assumptions about index skew, row exchangeability, and cache consistency. No new physical entities are introduced; the Eff-TT table, Reuse Buffer, index bijection, and embedding cache are software constructs whose contributions are measured in the ablation (Fig. 12) and pipeline (Fig. 14) experiments. All hand-set parameters (TT ranks, Hot ratio, LC, granularity) are numerically unreported except the TT embedding dimension, which materially limits independent replication.

free parameters (5)
  • TT ranks (R) = Not reported in the paper.
    TT ranks in Eq. 1 are user-set hyperparameters that trade compression versus accuracy and directly set the lookup cost O(k n^2 R^2) in Sec. III.B; the experimental values are never stated.
  • TT embedding dimension and core granularity = 128 (V100), 64 (T4); m_i and n_i not reported.
    Sec. V.H fixes the embedding dimension per GPU; the factorization M=m1*m2*m3 and N=n1*n2*n3 determines reuse opportunities, but the chosen factorizations are not given.
  • Hot ratio = Not reported in the paper.
    Algorithm 2 uses the Hot ratio to exempt frequent embeddings from reordering; the threshold shapes the index bijection and the resulting reuse rate.
  • LC (Load Capacity) = Not reported in the paper.
    Sec. IV.B uses LC to bound the prefetch and gradient queues and govern cache eviction; the value trades memory footprint against CPU-GPU communication.
  • Prefetch queue length = 1 in the sequential ablation; the production value is not reported.
    Sec. V.L sets the queue length to 1 to disable pipelining; the production length used for the reported 2.44x pipeline speedup is not stated.
assumptions (5)
  • domain assumption The categorical feature space is low-rank, so TT decomposition with modest ranks preserves task-relevant information.
    The Eff-TT design (Sec. III.A) and Eq. 2 assume the embedding matrix can be factorized into small cores without losing structure needed for FDIA and CTR accuracy; the paper tests this empirically (Tables III and V) but derives no rank bound.
  • domain assumption Sparse indices follow a power-law distribution with local co-occurrence structure.
    Sec. II.C asserts global power-law popularity and local within-batch structure; the reuse buffer (Algorithm 1), gradient aggregation (Sec. III.E), and index bijection (Algorithm 2) all rely on this premise.
  • domain assumption Embedding rows are exchangeable at initialization, so frequency-based permutation does not change the model.
    Sec. III.G states that all embedding rows are initially equivalent to justify position-based index growth sorting; the claim is asserted, not proven for later training stages.
  • domain assumption The GPU-side embedding cache resolves read-after-write hazards without stale reads.
    Sec. IV.B describes the adaptive filling policy and lifecycle control but provides no consistency proof; the correctness of the pipeline depends on this mechanism.
  • standard math Standard modularity-based community detection assigns co-occurring indices to nearby positions.
    Sec. III.H applies modularity clustering (Eq. 10) from references [40]-[42] as a well-defined objective; the locality benefit is evaluated empirically, not derived.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Rec-AD: An Efficient Computation Framework for FDIA Detection Based on Tensor Train Decomposition and Deep Learning Recommendation Model." pith.science (2026). https://pith.science/paper/AKXENOMY

@misc{pith2026250714668,
  author       = {Pith},
  title        = {Pith review of: Rec-AD: An Efficient Computation Framework for FDIA Detection Based on Tensor Train Decomposition and Deep Learning Recommendation Model},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AKXENOMY}},
  note         = {Machine review of arXiv:2507.14668}
}
read the original abstract

Deep learning models have been widely adopted for False Data Injection Attack (FDIA) detection in smart grids due to their ability to capture unstructured and sparse features. However, the increasing system scale and data dimensionality introduce significant computational and memory burdens, particularly in large-scale industrial datasets, limiting detection efficiency. To address these issues, this paper proposes Rec-AD, a computationally efficient framework that integrates Tensor Train decomposition with the Deep Learning Recommendation Model (DLRM). Rec-AD enhances training and inference efficiency through embedding compression, optimized data access via index reordering, and a pipeline training mechanism that reduces memory communication overhead. Fully compatible with PyTorch, Rec-AD can be integrated into existing FDIA detection systems without code modifications. Experimental results show that Rec-AD significantly improves computational throughput and real-time detection performance, narrowing the attack window and increasing attacker cost. These advancements strengthen edge computing capabilities and scalability, providing robust technical support for smart grid security.

Figures

Figures reproduced from arXiv: 2507.14668 by the authors.

Figure 1
Figure 1. Rec-AD Overview II. FDIA DETECTION BASED ON DEEP LEARNING RECOMMENDATION MODEL WITH TENSOR-TRAIN DECOMPOSITION This section introduces the fundamental principles of DL￾RMs, TT decomposition, and embedding tables. Dense DNNs + Dense Input Sparse Input + … Dense & Sparse Interaction Layer Dense DNNs Embedding Tables … Prediction Bottom MLP [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. DLRM framework [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. The TT table is represented by an embedding table. [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: The difference between TT table lookup and [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: TT table back propagation includes two steps: (a) [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: (a) Eff-TT table query without reordering; (b) Eff-TT [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: The process of creating an index bijection is: 1) [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: Pipeline training system based on TT A. Pipeline training system framework Building upon TT decomposition, we further propose a hierarchical memory system to enhance the scalability of Rec-AD by leveraging host memory expansion. This strat￾egy is particularly critical …
Figure 9
Figure 9. Figure 9: Using GPU-side embedded cache design to solve [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 10
Figure 10. Figure 10: With limited GPU resources, single GPU can achieve end-to-end training speed-up ( [PITH_FULL_IMAGE:figures/full_fig_p012_10.png]
Figure 11
Figure 11. Figure 11: Training throughput with multiple GPUs enabled [PITH_FULL_IMAGE:figures/full_fig_p012_11.png]
Figure 12
Figure 12. Figure 12: Eff-TT table optimization decomposition We benchmarked three models on this large table setup: Rec-AD, HugeCTR, and TorchRec, each trained with varying numbers of GPUs. As shown in [PITH_FULL_IMAGE:figures/full_fig_p012_12.png]
Figure 14
Figure 14. Figure 14: Speedup of partial optimization training vs. overall [PITH_FULL_IMAGE:figures/full_fig_p013_14.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Clustered Federated Learning for Generalizable FDIA Detection in Smart Grids with Heterogeneous Data

    cs.LG 2025-07 reject novelty 4.0 of 10

    FedClusAvg is claimed to improve FDIA detection accuracy and communication efficiency under non-IID smart grid data, but the weighting formula contradicts the stated mechanism and the evidence is not reproducible.

Reference graph

Works this paper leans on

50 extracted references · 40 canonical work pages · cited by 1 Pith paper

  1. [1]

    Dual-hybrid intrusion detection system to detect false data injection in smart grids,

    S. H. Mohammed, M. S. J. Singh, A. Al-Jumaily, M. T. Islam, M. S. Islam, A. M. Alenezi, and M. S. Soliman, “Dual-hybrid intrusion detection system to detect false data injection in smart grids,” PloS One, vol. 20, no. 1, p. e0316536, 2025

    work page 2025

  2. [2]

    False data injection attack detection for virtual coupling systems of heavy-haul trains: A deep learning approach,

    X. Yu, W. Li, S. Li, Y . Yang, and J. Peng, “False data injection attack detection for virtual coupling systems of heavy-haul trains: A deep learning approach,” Journal of Advanced Computational Intelligence and Intelligent Informatics, vol. 29, no. 1, pp. 175–186, 2025

    work page 2025

  3. [3]

    Graph neural network-based approach for detecting false data injection attacks on voltage stability,

    S. R. Fahim, R. Atat, C. Kececi, A. Takiddin, M. Ismail, K. R. Davis, and E. Serpedin, “Graph neural network-based approach for detecting false data injection attacks on voltage stability,” IEEE Open Access Journal of Power and Energy , vol. 12, pp. 12–23, 2025

    work page 2025

  4. [4]

    An advanced defense mechanism for detecting false data injection attacks in cyber-physical systems through feature selection based-machine learning algorithms,

    Y . Gasmi, N. Bali, A. Karim, and S. Belagoune, “An advanced defense mechanism for detecting false data injection attacks in cyber-physical systems through feature selection based-machine learning algorithms,” in 2024 International Conference on Advances in Electrical and Com- munication Technologies (ICAECOT), 2024, pp. 1–5

    work page 2024

  5. [5]

    A new false data injection detection protocol based machine learning for p2p energy transaction between cevs,

    D. Said and M. Elloumi, “A new false data injection detection protocol based machine learning for p2p energy transaction between cevs,” in 2022 IEEE International Conference on Electrical Sciences and Technologies in Maghreb (CISTEM) , vol. 4, 2022, pp. 1–5

    work page 2022

  6. [6]

    Automated deep cnn-lstm architecture design for solar irradiance forecasting,

    S. M. J. Jalali, S. Ahmadian, A. Kavousi-Fard, A. Khosravi, and S. Naha- vandi, “Automated deep cnn-lstm architecture design for solar irradiance forecasting,” IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol. 52, no. 1, pp. 54–65, 2021

    work page 2021

  7. [7]

    A photovoltaic power forecasting model based on dendritic neuron networks with the aid of wavelet transform,

    T. Zhang, C. Lv, F. Ma, K. Zhao, H. Wang, and G. M. O’Hare, “A photovoltaic power forecasting model based on dendritic neuron networks with the aid of wavelet transform,” Neurocomputing, vol. 397, pp. 438–446, 2020

    work page 2020

  8. [8]

    A comparative study on short-term pv power forecasting using decomposition based optimized extreme learning machine algorithm,

    M. K. Behera and N. Nayak, “A comparative study on short-term pv power forecasting using decomposition based optimized extreme learning machine algorithm,” Engineering Science and Technology, an International Journal, vol. 23, no. 1, pp. 156–167, 2020

    work page 2020

Show all 50 references
  1. [9]

    Short-term photovoltaic power generation forecasting based on random forest feature selection and ceemd: A case study,

    D. Niu, K. Wang, L. Sun, J. Wu, and X. Xu, “Short-term photovoltaic power generation forecasting based on random forest feature selection and ceemd: A case study,” Applied soft computing , vol. 93, p. 106389, 2020

    2020

  2. [10]

    Q. Li, X. Zhang, T. Ma, C. Jiao, H. Wang, and W. Hu, “A multi-step ahead photovoltaic power prediction model based on similar day, en- hanced colliding bodies optimization, variational mode decomposition, and deep extreme learning machine,” Energy, vol. 224, p. 120094, 2021

    2021

  3. [11]

    Mixed- precision embedding using a cache,

    J. A. Yang, J. Huang, J. Park, P. T. P. Tang, and A. Tulloch, “Mixed- precision embedding using a cache,” arXiv preprint arXiv:2010.11305 , 2020

    2010 arXiv

  4. [12]

    Learnable embedding sizes for recommender systems,

    S. Liu, C. Gao, Y . Chen, D. Jin, and Y . Li, “Learnable embedding sizes for recommender systems,” in International Conference on Learning Representations , 2021. [Online]. Available: https: //openreview.net/forum?id=vQzcqQWIS0q

    2021

  5. [13]

    J. B. Schafer, D. Frankowski, J. Herlocker, and S. Sen, Collaborative Filtering Recommender Systems . Berlin, Heidelberg: Springer Berlin Heidelberg, 2007, pp. 291–324. [Online]. Available: https://doi.org/10. 1007/978-3-540-72079-9 9

    2007

  6. [14]

    A novel group recommendation model with two-stage deep learning,

    Z. Huang, Y . Liu, C. Zhan, C. Lin, W. Cai, and Y . Chen, “A novel group recommendation model with two-stage deep learning,” IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol. 52, no. 9, pp. 5853–5864, 2022

    2022

  7. [15]

    Deep residual networks for image recognition,

    P. Pateriya, P. Jain, P. Niveditha, V . Tiwari, and S. Vishwakarma, “Deep residual networks for image recognition,” International Journal of Innovative Research in Computer and Communication Engineering , vol. 11, pp. 10 742–10 747, 11 2023

    2023

  8. [16]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, “Attention is all you need,” in Proceedings of the 31st International Conference on Neural Information Processing Systems, ser. NIPS’17. Red Hook, NY , USA: Curran Associates ...

    2017

  9. [17]

    Understanding capacity-driven scale-out neural recommendation infer- ence,

    M. Lui, Y . Yetim, Z. ¨Ozkan, S.-Y . Tsai, C.-J. Wu, and M. Hempstead, “Understanding capacity-driven scale-out neural recommendation infer- ence,” in 2021 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS) , 2021, pp. 162–171

    2021

  10. [18]

    Distributed hierarchical gpu parameter server for massive scale deep learning ads systems,

    W. Zhao, D. Xie, R. Jia, Y . Qian, R. Ding, M. Sun, and P. Li, “Distributed hierarchical gpu parameter server for massive scale deep learning ads systems,” arXiv preprint arXiv:2003.05622 , 2020

    2003 arXiv

  11. [19]

    Software-hardware co-design for fast and scalable training of deep learning recommendation models,

    D. Mudigere, Y . Hao, J. Huang, Z. Jia, A. Tulloch, S. Sridharan, X. Liu, M. Ozdal, J. Nie, J. Park, L. Luo, J. A. Yang, L. Gao, D. Ivchenko, A. Basant, Y . Hu, J. Yang, E. K. Ardestani, X. Wang, R. Komuravelli, C.-H. Chu, S. Yilmaz, H. Li, J. Qian, Z. Feng, Y . Ma, J. Yang, E...

    2022

  12. [20]

    Nvidia A100 GPU: Performance & innovation for gpu computing,

    J. Choquette and W. Gandhi, “Nvidia A100 GPU: Performance & innovation for gpu computing,” in 2020 IEEE Hot Chips 32 Symposium (HCS), 2020, pp. 1–43

    2020

  13. [21]

    Nvidia merlin hugectr,

    Nvidia, “Nvidia merlin hugectr,” {https://developer.nvidia.com/ nvidia-merlin/hugectr}

  14. [22]

    Post-training 4- bit quantization on embedding tables,

    H. Guan, A. Malevich, J. Yang, J. Park, and H. Yuen, “Post-training 4- bit quantization on embedding tables,” arXiv preprint arXiv:1911.02079, 2019

    1911 arXiv

  15. [23]

    TT-Rec: Tensor train compres- sion for deep learning recommendation models,

    C. Yin, B. Acun, X. Liu, and C.-J. Wu, “TT-Rec: Tensor train compres- sion for deep learning recommendation models,” 01 2021

    2021

  16. [24]

    Deep learning recommendation model for personalization and recommendation systems,

    M. Naumov, D. Mudigere, H. M. Shi, J. Huang, N. Sundaraman, J. Park, X. Wang, U. Gupta, C. Wu, A. G. Azzolini, D. Dzhulgakov et al. , “Deep learning recommendation model for personalization and recommendation systems,” arXiv preprint arXiv:1906.00091 , 2019

    1906 arXiv

  17. [25]

    Accelerating recommendation system training by leveraging popular choices,

    M. Adnan, Y . E. Maboud, D. Mahajan, and P. J. Nair, “Accelerating recommendation system training by leveraging popular choices,” Pro- ceedings of the VLDB Endowment , vol. 15, no. 1, pp. 127–140, 2021

    2021

  18. [26]

    Parallax: Sparsity-aware data parallel training of deep neural networks,

    S. Kim, G.-I. Yu, H. Park, S. Cho, E. Jeong, H. Ha, S. Lee, J. S. Jeong, and B.-G. Chun, “Parallax: Sparsity-aware data parallel training of deep neural networks,” in Proceedings of the Fourteenth EuroSys Conference 2019 , ser. EuroSys ’19. New York, NY , USA: Association for ...

    2019

  19. [27]

    Scaling distributed machine learning with the parameter server,

    M. Li, D. G. Andersen, J. W. Park, A. J. Smola, A. Ahmed, V . Josifovski, J. Long, E. J. Shekita, and B.-Y . Su, “Scaling distributed machine learning with the parameter server,” in 11th USENIX Symposium on Operating Systems Design and Implementation (OSDI 14) . Broomfield, CO...

    2014

  20. [28]

    Scalefreectr: Mixcache-based distributed training system for ctr models with huge embedding table,

    H. Guo, W. Guo, Y . Gao, R. Tang, X. He, and W. Liu, “Scalefreectr: Mixcache-based distributed training system for ctr models with huge embedding table,” in Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval, ser. SI...

    2021

  21. [29]

    Centaur: a chiplet-based, hybrid sparse-dense accelerator for personalized recommendations,

    R. Hwang, T. Kim, Y . Kwon, and M. Rhu, “Centaur: a chiplet-based, hybrid sparse-dense accelerator for personalized recommendations,” in Proceedings of the ACM/IEEE 47th Annual International Symposium on Computer Architecture, ser. ISCA ’20. IEEE Press, 2020, p. 968–981. [Onli...

    2020

  22. [30]

    Aibox: Ctr prediction model training on a single node,

    W. Zhao, J. Zhang, D. Xie, Y . Qian, R. Jia, and P. Li, “Aibox: Ctr prediction model training on a single node,” in Proceedings of the 28th ACM International Conference on Information and Knowledge Management, ser. CIKM ’19. New York, NY , USA: Association for Computing Machin...

    2019

  23. [31]

    Unisolar: An open dataset of photovoltaic solar energy generation in a large multi-campus university setting,

    S. Wimalaratne, D. Haputhanthri, S. Kahawala, G. Gamage, D. Ala- hakoon, and A. Jennings, “Unisolar: An open dataset of photovoltaic solar energy generation in a large multi-campus university setting,” in 2022 15th International Conference on Human System Interaction (HSI), 20...

    2022

  24. [32]

    Compressing recurrent neural network with tensor train,

    A. Tjandra, S. Sakti, and S. Nakamura, “Compressing recurrent neural network with tensor train,” in 2017 International Joint Conference on Neural Networks (IJCNN) , 2017, pp. 4451–4458

    2017

  25. [33]

    Hardware-enabled efficient data processing with tensor- train decomposition,

    Z. Qu, L. Deng, B. Wang, H. Chen, J. Lin, L. Liang, G. Li, Z. Zhang, and Y . Xie, “Hardware-enabled efficient data processing with tensor- train decomposition,” IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems , vol. 41, no. 2, pp. 372–385, 2022

    2022

  26. [34]

    Tensor-train decomposition,

    I. V . Oseledets, “Tensor-train decomposition,” SIAM Journal on Scientific Computing , vol. 33, no. 5, pp. 2295–2317, 2011. [Online]. Available: https://epubs.siam.org/doi/abs/10.1137/090752286

    2011 doi

  27. [35]

    Adtt: a highly efficient distributed tensor-train decomposition method for iiot big data,

    X. Wang, L. T. Yang, Y . Wang, L. Ren, and M. J. Deen, “Adtt: a highly efficient distributed tensor-train decomposition method for iiot big data,” IEEE Transactions on Industrial Informatics , vol. 17, no. 3, pp. 1573– 1582, 2020

    2020

  28. [36]

    Tensor train decomposition on tensorflow (t3f),

    A. Novikov, P. Izmailov, V . Khrulkov, M. Figurnov, and I. Oseledets, “Tensor train decomposition on tensorflow (t3f),” Journal of Machine Learning Research, vol. 21, no. 30, pp. 1–7, 2020

    2020

  29. [37]

    Tensorized embedding layers,

    O. Hrinchuk, V . Khrulkov, L. Mirvakhabova, E. Orlova, and I. Oseledets, “Tensorized embedding layers,” in Findings of the Association for Computational Linguistics: EMNLP 2020 , T. Cohn, Y . He, and Y . Liu, Eds. Online: Association for Computational Linguistics, Nov. 2020, p...

    2020

  30. [38]

    Space: Locality-aware pro- cessing in heterogeneous memory for personalized recommendations,

    H. Kal, S. Lee, G. Ko, and W. W. Ro, “Space: Locality-aware pro- cessing in heterogeneous memory for personalized recommendations,” in 2021 ACM/IEEE 48th Annual International Symposium on Computer Architecture (ISCA), 2021, pp. 679–691

    2021

  31. [39]

    Deeprecsys: A system for optimizing end-to-end at-scale neural recommendation inference,

    U. Gupta, S. Hsia, V . Saraph, X. Wang, B. Reagen, G.-Y . Wei, H.- H. S. Lee, D. Brooks, and C.-J. Wu, “Deeprecsys: A system for optimizing end-to-end at-scale neural recommendation inference,” in 2020 ACM/IEEE 47th Annual International Symposium on Computer Architecture (ISCA...

    2020

  32. [40]

    Rabbit order: Just-in-time parallel reordering for fast graph analysis,

    J. Arai, H. Shiokawa, T. Yamamuro, M. Onizuka, and S. Iwamura, “Rabbit order: Just-in-time parallel reordering for fast graph analysis,” in 2016 IEEE International Parallel and Distributed Processing Sympo- sium (IPDPS), 2016, pp. 22–31

    2016

  33. [41]

    A novel modularity- based discrete state transition algorithm for community detection in networks,

    X. Zhou, K. Yang, Y . Xie, C. Yang, and T. Huang, “A novel modularity- based discrete state transition algorithm for community detection in networks,” Neurocomputing, vol. 334, pp. 89–99, 2019

    2019

  34. [42]

    Fast algorithm for modularity-based graph clustering,

    H. Shiokawa, Y . Fujiwara, and M. Onizuka, “Fast algorithm for modularity-based graph clustering,” in Proceedings of the Twenty- Seventh AAAI Conference on Artificial Intelligence , ser. AAAI’13. Bellevue, Washington: AAAI Press, 2013, p. 1170–1176

    2013

  35. [43]

    Click-through rate prediction,

    W. C. Steve Wang, “Click-through rate prediction,” {https://kaggle.com/ competitions/avazu-ctr-prediction}, 2014

    2014

  36. [44]

    Terabyte click logs,

    Criteo, “Terabyte click logs,” {https://labs.criteo.com/2013/12/ download-terabyte-click-logs-2/ }, 2013

    2013

  37. [45]

    Display advertising challenge,

    O. C. Jean-Baptiste Tien, joycenv, “Display advertising challenge,” {https://kaggle.com/competitions/criteo-display-ad-challenge }, 2014

    2014

  38. [46]

    Matpower-a matlab power system simulation package: User

    R. Zimmerman, C. Murillo-Sanchez, and D. Gan, “Matpower-a matlab power system simulation package: User”s manual,” 12 1997

    1997

  39. [47]

    Intrusion detection of cyber physical energy system based on multivariate ensemble classification,

    Y . Li, W. Xue, T. Wu, H. Wang, B. Zhou, S. Aziz, and Y . He, “Intrusion detection of cyber physical energy system based on multivariate ensemble classification,” Energy, vol. 218, p. 119505, 2021. [Online]. Available: https://www.sciencedirect.com/ science/article/pii/S036054...

    2021

  40. [48]

    Torchrec,

    Meta, “Torchrec,” {github.com/pytorch/torchrec}, 2022

    2022

  41. [49]

    Nvtabular,

    N. Merlin, “Nvtabular,” {https://developer.nvidia.com/nvidia-merlin/ nvtabular}

  42. [50]

    Nvidia, “cublas,” {developer.nvidia.com/cublas}

Pith tools

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