pith. machine review for the scientific record. sign in

arxiv: 2605.15032 · v1 · submitted 2026-05-14 · 📡 eess.SP · cs.LG

Recognition: 2 theorem links

· Lean Theorem

Multi-Block Attention for Efficient Channel Estimation in IRS-Assisted mmWave MIMO

Mehrdad Momen-Tayefeh , Mehrshad Momen-Tayefeh , Maryam Sabbaghian
Authors on Pith no claims yet

Pith reviewed 2026-05-15 03:11 UTC · model grok-4.3

classification 📡 eess.SP cs.LG
keywords IRS channel estimationmmWave MIMOdeep learningpilot overheadattention mechanismOFDMcascaded channel
0
0 comments X

The pith

A multi-block attention network recovers channel estimates from 87 percent fewer pilots in IRS-assisted mmWave MIMO systems.

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

The paper shows that selectively deactivating IRS elements during training cuts pilot overhead dramatically, provided a neural network compensates for the resulting loss of spatial features. It first establishes that DFT and Hadamard phase shifts are optimal for least-squares estimation, then introduces a two-stage architecture that uses convolutional attention to restore correlations and complex convolutions to suppress noise. The resulting Multi-Block Attention framework therefore trades a modest increase in computation for a large reduction in training symbols. A reader would care because pilot overhead grows with the number of IRS elements and quickly becomes the dominant cost in large deployments. Simulations report an 87 percent overhead reduction versus conventional least-squares estimation together with roughly 51 percent lower normalized mean-squared error at 10 dB SNR.

Core claim

The MBA architecture compensates for the feature loss induced by selective IRS deactivation through attention-guided spatial recovery in a Convolutional Attention Network followed by denoising in a Complex Multi-Convolutional Network, thereby achieving accurate cascaded channel estimates with far fewer pilots than least-squares methods while preserving low computational complexity across varied propagation environments.

What carries the argument

The Multi-Block Attention (MBA) two-stage network consisting of a Convolutional Attention Network (CAN) that restores spatial correlations and a Complex Multi-Convolutional Network (CMN) that performs noise suppression.

If this is right

  • Larger IRS arrays become practical because pilot count no longer scales linearly with the number of reflecting elements.
  • More transmission time slots remain available for data after the shortened training phase.
  • The same architecture works across different mmWave propagation scenarios without retuning.
  • Computational cost stays low enough for real-time operation on typical base-station hardware.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The attention-based recovery step might transfer to other passive array systems such as holographic surfaces if the spatial correlation structure is similar.
  • Hardware impairments like phase quantization errors could be folded into the training data to reduce the simulation-to-reality gap.
  • Extending the framework to time-varying channels would require only modest additions to the existing denoising stage.

Load-bearing premise

The statistics of the synthetic mmWave channels and noise used for training match those of real deployments closely enough for the network to generalize without retraining or fine-tuning.

What would settle it

A direct NMSE comparison between the trained MBA model and the LS estimator on measured channel traces collected from a physical IRS-assisted mmWave testbed at 10 dB SNR.

Figures

Figures reproduced from arXiv: 2605.15032 by Maryam Sabbaghian, Mehrdad Momen-Tayefeh, Mehrshad Momen-Tayefeh.

Figure 1
Figure 1. Figure 1: Schematic of IRS-assisted communication systems [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Comparison of IRS activation patterns, where yell [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Structure of the Attention Block (AB), combining a [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Overview of the proposed MBA framework. hierarchical design guarantees structural consistency and ro￾bustness to noise in the estimated channels. To optimize both networks, we adopt the MSE loss. The loss functions for CAN and CMN are defined as: LCAN = 1 2D X D i=1 [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Convergence of the proposed model in terms of NMSE o [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: MSE comparison between the LS estimator and the MBA [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: NMSE comparison of different methods for N [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Impact of the number of training signals on NMSE per [PITH_FULL_IMAGE:figures/full_fig_p011_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Generalization analysis of different models in UM [PITH_FULL_IMAGE:figures/full_fig_p011_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Effect of the number of BS antennas on NMSE perform [PITH_FULL_IMAGE:figures/full_fig_p012_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Impact of the number of elements in the IRS on NMSE [PITH_FULL_IMAGE:figures/full_fig_p012_11.png] view at source ↗
read the original abstract

Intelligent Reflecting Surfaces (IRSs) are a promising technology for enhancing the spectral and energy efficiency of millimeter-wave (mmWave) multiple-input multiple-output (MIMO) systems. In these systems, accurate channel estimation remains challenging due to the passive nature of IRS elements and the high pilot overhead in large-scale deployments. This paper presents a deep learning-based Multi-Block Attention (MBA) framework for efficient cascaded channel estimation in IRS-assisted mmWave MIMO systems that utilize orthogonal frequency division multiplexing (OFDM). First, we show the optimality of the discrete Fourier transform (DFT) and Hadamard matrices as phase configurations for least squares (LS) estimation. To reduce training overhead, we selectively deactivate IRS elements and compensate for induced feature loss using a two-stage architecture: (i) a Convolutional Attention Network (CAN) for spatial correlation recovery and (ii) a Complex Multi-Convolutional Network (CMN) for noise suppression. The MBA architecture mitigates error propagation through attention-guided feature refinement and denoising. Simulation results indicate that the MBA method reduces pilot overhead by up to 87% compared to the LS estimator. Additionally, at signal-to-noise ratios of 10 dB, our proposed method achieves approximately 51% lower normalized mean squared error (NMSE) than leading methods. It also maintains low computational complexity and adapts effectively to various propagation environments.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 0 minor

Summary. The paper presents a deep learning-based Multi-Block Attention (MBA) framework for efficient cascaded channel estimation in IRS-assisted mmWave MIMO-OFDM systems. It first shows the optimality of DFT and Hadamard matrices as phase configurations for LS estimation. To reduce training overhead, selective deactivation of IRS elements is used, with a two-stage architecture: Convolutional Attention Network (CAN) for spatial correlation recovery and Complex Multi-Convolutional Network (CMN) for noise suppression. The MBA mitigates error propagation through attention-guided feature refinement. Simulation results show up to 87% pilot overhead reduction compared to LS and approximately 51% lower NMSE at 10 dB SNR than leading methods.

Significance. If the simulation results hold under the stated models, the work offers a concrete reduction in pilot overhead (up to 87%) while improving NMSE by ~51% at moderate SNR, which would be valuable for scaling IRS deployments in mmWave MIMO. The attention-based compensation for deactivation-induced feature loss is a targeted architectural contribution, and the low-complexity claim strengthens its potential utility. The simulation-only nature and dependence on synthetic cascaded channels limit immediate practical significance without additional validation.

major comments (1)
  1. [§5, Table II] §5 (Simulation Results), Table II: the 51% NMSE reduction at 10 dB SNR versus leading methods is reported without specifying the exact baseline algorithms, the number of Monte Carlo trials, or error bars; this makes it impossible to judge whether the gain is statistically robust or sensitive to random seeds in the synthetic data generation.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive comment on the simulation results section. We agree that additional details are needed for reproducibility and to demonstrate statistical robustness. We will revise the manuscript to address this point fully.

read point-by-point responses

  1. Referee: [§5, Table II] §5 (Simulation Results), Table II: the 51% NMSE reduction at 10 dB SNR versus leading methods is reported without specifying the exact baseline algorithms, the number of Monte Carlo trials, or error bars; this makes it impossible to judge whether the gain is statistically robust or sensitive to random seeds in the synthetic data generation.

    Authors: We agree that the current presentation lacks sufficient detail for assessing statistical significance. In the revised manuscript, we will explicitly name the baseline algorithms (the leading deep-learning methods referenced in Section V), state that 1000 independent Monte Carlo trials were performed per SNR point, and add error bars (standard deviation) to Table II and the associated NMSE curves. These changes will allow readers to evaluate robustness directly. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper proposes a neural network architecture (MBA with CAN and CMN blocks) trained on synthetic mmWave cascaded channels to estimate IRS-assisted MIMO-OFDM channels under reduced pilot overhead. The optimality claim for DFT/Hadamard phase configurations is presented as a standard LS analysis result, not a self-referential definition. All performance numbers (87% overhead reduction, 51% NMSE improvement) are empirical simulation outputs from a trained model evaluated against independent baselines; they are not algebraically forced by the inputs or by any self-citation chain. The derivation chain remains self-contained with externally falsifiable simulation benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no explicit free parameters, axioms, or invented entities; the work rests on standard wireless channel models and DL training assumptions.

pith-pipeline@v0.9.0 · 5559 in / 960 out tokens · 47626 ms · 2026-05-15T03:11:35.261610+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

44 extracted references · 44 canonical work pages · 2 internal anchors

  1. [1]

    Intellige nt reflecting surface-aided wireless communications: A tutorial,

    Q. Wu, S. Zhang, B. Zheng, C. Y ou, and R. Zhang, “Intellige nt reflecting surface-aided wireless communications: A tutorial,” IEEE Transactions on Communications , vol. 69, no. 5, pp. 3313–3351, 2021

    work page 2021

  2. [2]

    Toward smart wireless communications via intellig ent reflecting surfaces: A contemporary survey,

    S. Gong, X. Lu, D. T. Hoang, D. Niyato, L. Shu, D. I. Kim, and Y .-C. Liang, “Toward smart wireless communications via intellig ent reflecting surfaces: A contemporary survey,” IEEE Communications Surveys & Tutorials, vol. 22, no. 4, pp. 2283–2314, 2020

    work page 2020

  3. [3]

    Reconfigur able in- telligent surfaces vs. relaying: Differences, similariti es, and performance comparison,

    M. Di Renzo, K. Ntontin, J. Song, F. H. Danufane, X. Qian, F . Lazarakis, J. De Rosny, D.-T. Phan-Huy, O. Simeone, R. Zhang, M. Debbah, G. Lerosey, M. Fink, S. Tretyakov, and S. Shamai, “Reconfigur able in- telligent surfaces vs. relaying: Differences, similariti es, and performance comparison,” IEEE Open Journal of the Communications Society , vol. 1, pp....

    work page 2020

  4. [4]

    Intelligent reflecting surface enhan ced wireless network via joint active and passive beamforming,

    Q. Wu and R. Zhang, “Intelligent reflecting surface enhan ced wireless network via joint active and passive beamforming,” IEEE Transactions on Wireless Communications , vol. 18, no. 11, pp. 5394–5409, 2019

    work page 2019

  5. [5]

    Reconfigurable intelligent surfaces for wireless communica- tions: Principles, challenges, and opportunities,

    M. A. ElMossallamy, H. Zhang, L. Song, K. G. Seddik, Z. Han , and G. Y . Li, “Reconfigurable intelligent surfaces for wireless communica- tions: Principles, challenges, and opportunities,” IEEE Transactions on Cognitive Communications and Networking , vol. 6, no. 3, pp. 990–1002, 2020

    work page 2020

  6. [6]

    Intelligent reflecting surface-e nhanced OFDM: Channel estimation and reflection optimization,

    B. Zheng and R. Zhang, “Intelligent reflecting surface-e nhanced OFDM: Channel estimation and reflection optimization,” IEEE Wireless Commu- nications Letters , vol. 9, no. 4, pp. 518–522, 2020

    work page 2020

  7. [7]

    Channel estimation and low- complexity beamforming design for passive intelligent surface assist ed MISO wireless energy transfer,

    D. Mishra and H. Johansson, “Channel estimation and low- complexity beamforming design for passive intelligent surface assist ed MISO wireless energy transfer,” in ICASSP 2019 - 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICA SSP), Brighton, UK, May 2019, pp. 4659–4663

    work page 2019

  8. [8]

    An optimal channel estim ation scheme for intelligent reflecting surfaces based on a minimum varia nce unbiased estimator,

    T. L. Jensen and E. D. Carvalho, “An optimal channel estim ation scheme for intelligent reflecting surfaces based on a minimum varia nce unbiased estimator,” in ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , Barcelona, Spain, May 2020, pp. 5000–5004

    work page 2020

  9. [9]

    Intelligent re flecting surface meets OFDM: Protocol design and rate maximization,

    Y . Y ang, B. Zheng, S. Zhang, and R. Zhang, “Intelligent re flecting surface meets OFDM: Protocol design and rate maximization, ” IEEE Transactions on Communications , vol. 68, no. 7, pp. 4522–4535, 2020

    work page 2020

  10. [10]

    Intelligent reflecting s urface assisted multi-user OFDMA: Channel estimation and training design,

    B. Zheng, C. Y ou, and R. Zhang, “Intelligent reflecting s urface assisted multi-user OFDMA: Channel estimation and training design, ” IEEE Transactions on Wireless Communications , vol. 19, no. 12, pp. 8315– 8329, 2020

    work page 2020

  11. [11]

    Intelligent reflecting surface-assisted multi-user MIS O commu- nication: Channel estimation and beamforming design,

    H. Alwazani, A. Kammoun, A. Chaaban, M. Debbah, M.-S. Al ouini et al. , “Intelligent reflecting surface-assisted multi-user MIS O commu- nication: Channel estimation and beamforming design,” IEEE Open Journal of the Communications Society , vol. 1, pp. 661–680, 2020

    work page 2020

  12. [12]

    Channel estimation for RIS- aided mmWave MIMO systems via atomic norm minimization,

    J. He, H. Wymeersch, and M. Juntti, “Channel estimation for RIS- aided mmWave MIMO systems via atomic norm minimization,” IEEE Transactions on Wireless Communications , vol. 20, no. 9, pp. 5786– 5797, 2021

    work page 2021

  13. [13]

    Sparse channe l estimation for IRS-assisted millimeter wave MIMO OFDM systems,

    M. Liu, T. Lin, Y . Zhu, and Y .-J. A. Zhang, “Sparse channe l estimation for IRS-assisted millimeter wave MIMO OFDM systems,” IEEE Trans- actions on Communications , vol. 72, no. 10, pp. 6553–6568, 2024

    work page 2024

  14. [14]

    Channel est imation for reconfigurable intelligent surface aided multi-user mm Wave MIMO systems,

    J. Chen, Y .-C. Liang, H. V . Cheng, and W. Y u, “Channel est imation for reconfigurable intelligent surface aided multi-user mm Wave MIMO systems,” IEEE Transactions on Wireless Communications , vol. 22, no. 10, pp. 6853–6869, 2023

    work page 2023

  15. [15]

    Inductiv e matrix completion and root-MUSIC-based channel estimation for in telligent re- flecting surface (IRS)-aided hybrid MIMO systems,

    K. F. Masood, J. Tong, J. Xi, J. Y uan, and Y . Y u, “Inductiv e matrix completion and root-MUSIC-based channel estimation for in telligent re- flecting surface (IRS)-aided hybrid MIMO systems,” IEEE Transactions on Wireless Communications , vol. 22, no. 11, pp. 7917–7931, 2023

    work page 2023

  16. [16]

    Efficient multi-user chan nel estimation for RIS-aided mmWave systems using shared channel subspace ,

    H. Chung, S. Hong, and S. Kim, “Efficient multi-user chan nel estimation for RIS-aided mmWave systems using shared channel subspace ,” IEEE Transactions on Wireless Communications , vol. 23, no. 8, pp. 8512– 8527, 2024

    work page 2024

  17. [17]

    Efficient channel e stimation for millimeter-wave RIS-assisted SIMO-OFDM systems with redu ced com- plexity and pilot overhead,

    F.-S. Tseng, H. Chung, and P .-H. Wu, “Efficient channel e stimation for millimeter-wave RIS-assisted SIMO-OFDM systems with redu ced com- plexity and pilot overhead,” IEEE Transactions on V ehicular Technology, vol. 73, no. 10, pp. 14 809–14 821, 2024

    work page 2024

  18. [18]

    A low-rank sparse te nsor recovery-based channel estimation scheme for RIS-assiste d mmWave OFDM systems,

    Z. Huang, C. Liu, Y . Song, and Y . Fu, “A low-rank sparse te nsor recovery-based channel estimation scheme for RIS-assiste d mmWave OFDM systems,” IEEE Transactions on V ehicular Technology , pp. 1– 16, 2025

    work page 2025

  19. [19]

    Y . C. Eldar, A. Goldsmith, D. G¨ und¨ uz, and H. V . Poor,Machine learning and wireless communications . Cambridge University Press, 2022

    work page 2022

  20. [20]

    Art ificial neural networks-based machine learning for wireless networks: A t utorial,

    M. Chen, U. Challita, W. Saad, C. Yin, and M. Debbah, “Art ificial neural networks-based machine learning for wireless networks: A t utorial,” IEEE Communications Surveys & Tutorials , vol. 21, no. 4, pp. 3039– 3071, 2019

    work page 2019

  21. [21]

    An introduction to deep learni ng for the physical layer,

    T. O’Shea and J. Hoydis, “An introduction to deep learni ng for the physical layer,” IEEE Transactions on Cognitive Communications and Networking, vol. 3, no. 4, pp. 563–575, 2017

    work page 2017

  22. [22]

    Deep learning-ba sed time-varying channel estimation for RIS assisted communic ation,

    M. Xu, S. Zhang, J. Ma, and O. A. Dobre, “Deep learning-ba sed time-varying channel estimation for RIS assisted communic ation,” IEEE Communications Letters , vol. 26, no. 1, pp. 94–98, 2022

    work page 2022

  23. [23]

    Deep denoising neural network assisted compressive channel est imation for mmWave intelligent reflecting surfaces,

    S. Liu, Z. Gao, J. Zhang, M. Di Renzo, and M.-S. Alouini, “ Deep denoising neural network assisted compressive channel est imation for mmWave intelligent reflecting surfaces,” IEEE Transactions on V ehicu- lar Technology, vol. 69, no. 8, pp. 9223–9228, 2020

    work page 2020

  24. [24]

    Deep lea rning- based channel extrapolation for hybrid RIS-aided mmWave sy stems with low-resolution ADCs,

    T. Gong, S. Zhang, F. Gao, Z. Li, and M. Di Renzo, “Deep lea rning- based channel extrapolation for hybrid RIS-aided mmWave sy stems with low-resolution ADCs,” IEEE Transactions on Wireless Communications, vol. 23, no. 10, pp. 14 408–14 420, 2024

    work page 2024

  25. [25]

    Deep le arning optimized sparse antenna activation for reconfigurable int elligent sur- face assisted communication,

    S. Zhang, S. Zhang, F. Gao, J. Ma, and O. A. Dobre, “Deep le arning optimized sparse antenna activation for reconfigurable int elligent sur- face assisted communication,” IEEE Transactions on Communications , vol. 69, no. 10, pp. 6691–6705, 2021

    work page 2021

  26. [26]

    Channel estimation for inte lligent reflecting surface assisted massive MIMO systems—a deep lea rning approach,

    Z. Mao, X. Liu, and M. Peng, “Channel estimation for inte lligent reflecting surface assisted massive MIMO systems—a deep lea rning approach,” IEEE Communications Letters , vol. 26, no. 4, pp. 798–802, 2022

    work page 2022

  27. [27]

    RIS- aided mmWave MIMO channel estimation using deep learning an d compressive sensing,

    A. Abdallah, A. Celik, M. M. Mansour, and A. M. Eltawil, “ RIS- aided mmWave MIMO channel estimation using deep learning an d compressive sensing,” IEEE Transactions on Wireless Communications , vol. 22, no. 5, pp. 3503–3521, 2023

    work page 2023

  28. [28]

    Deep learning for s uper- resolution channel estimation in reconfigurable intellige nt surface aided systems,

    W. Shen, Z. Qin, and A. Nallanathan, “Deep learning for s uper- resolution channel estimation in reconfigurable intellige nt surface aided systems,” IEEE Transactions on Communications , vol. 71, no. 3, pp. 1491–1503, 2023

    work page 2023

  29. [29]

    Hybrid d riven learning for channel estimation in intelligent reflecting s urface aided millimeter wave communication,

    S. Zheng, S. Wu, C. Jiang, W. Zhang, and X. Jing, “Hybrid d riven learning for channel estimation in intelligent reflecting s urface aided millimeter wave communication,” IEEE Transactions on Wireless Com- munications, vol. 23, no. 6, pp. 5801–5815, 2024

    work page 2024

  30. [30]

    GA N- based channel estimation for IRS-aided communication syst ems,

    M. Haider, I. Ahmed, A. Rubaai, C. Pu, and D. B. Rawat, “GA N- based channel estimation for IRS-aided communication syst ems,” IEEE Transactions on V ehicular Technology, vol. 73, no. 4, pp. 6012–6017, 2024

    work page 2024

  31. [31]

    Channel estimation for intelligent reflecting surface aided wireless communications using con ditional GAN,

    M. Y e, H. Zhang, and J.-B. Wang, “Channel estimation for intelligent reflecting surface aided wireless communications using con ditional GAN,” IEEE Communications Letters , vol. 26, no. 10, pp. 2340–2344, 2022

    work page 2022

  32. [32]

    Join t direct and indirect channel estimation for RIS-assisted millimeter- wave systems based on array signal processing,

    S. Noh, K. Seo, Y . Sung, D. J. Love, J. Lee, and H. Y u, “Join t direct and indirect channel estimation for RIS-assisted millimeter- wave systems based on array signal processing,” IEEE Transactions on Wireless Communications, vol. 22, no. 11, pp. 8378–8391, 2023

    work page 2023

  33. [33]

    C. M. Bishop and N. M. Nasrabadi, Pattern recognition and machine learning. Springer, 2006, vol. 4, no. 4

    work page 2006

  34. [34]

    TR 138 901 v17.0.0 (2022-04): 5G; study on channe l model for frequencies from 0.5 to 100 GHz (3GPP TR 38.901 version 17 .0.0 release 17),

    ETSI, “TR 138 901 v17.0.0 (2022-04): 5G; study on channe l model for frequencies from 0.5 to 100 GHz (3GPP TR 38.901 version 17 .0.0 release 17),” ETSI, Technical Report, 2022. 14

    work page 2022

  35. [35]

    Deep CNN-based channel estimation for mmWave massive MIMO syste ms,

    P . Dong, H. Zhang, G. Y . Li, I. S. Gaspar, and N. NaderiAli zadeh, “Deep CNN-based channel estimation for mmWave massive MIMO syste ms,” IEEE Journal of Selected Topics in Signal Processing , vol. 13, no. 5, pp. 989–1000, 2019

    work page 2019

  36. [36]

    An attenti on-aided deep learning framework for massive MIMO channel estimatio n,

    J. Gao, M. Hu, C. Zhong, G. Y . Li, and Z. Zhang, “An attenti on-aided deep learning framework for massive MIMO channel estimatio n,” IEEE Transactions on Wireless Communications , vol. 21, no. 3, pp. 1823– 1835, 2022

    work page 2022

  37. [37]

    Attention is all you need,

    A. V aswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jone s, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” Advances in neural information processing systems , vol. 30, 2017

    work page 2017

  38. [38]

    Deep Learning is Robust to Massive Label Noise

    D. Rolnick, A. V eit, S. Belongie, and N. Shavit, “Deep le arning is robust to massive label noise,” 2018. [Online]. Availabl e: https://arxiv.org/abs/1705.10694

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  39. [39]

    Ax- ial attention in multidimensional transformers,

    J. Ho, N. Kalchbrenner, D. Weissenborn, and T. Salimans , “Ax- ial attention in multidimensional transformers,” arXiv preprint arXiv:1912.12180, 2019

    work page arXiv 1912

  40. [40]

    The dynamical manin–mumford problem for plane polynomial automorphisms,

    R. Dujardin and C. Favre, “The dynamical manin–mumford problem for plane polynomial automorphisms,” Journal of the European Mathemat- ical Society , vol. 19, no. 11, pp. 3421–3465, 2017

    work page 2017

  41. [41]

    Extracting and composing robust features with denoising autoencoders

    P . Vincent, H. Larochelle, Y . Bengio, and P .-A. Manzago l, “Extracting and composing robust features with denoising autoencoders ,” in Proceedings of the 25th International Conference on Machin e Learning, ser. ICML ’08. New Y ork, NY , USA: Association for Computing Machinery, July 2008, p. 1096–1103. [Online]. Av ailable: https://doi.org/10.1145/1390...

    work page doi:10.1145/1390156.1390294 2008

  42. [42]

    Beyond a gaussian denoiser: Residual learning of deep CNN for image denoising ,

    K. Zhang, W. Zuo, Y . Chen, D. Meng, and L. Zhang, “Beyond a gaussian denoiser: Residual learning of deep CNN for image denoising ,” IEEE Transactions on Image Processing , vol. 26, no. 7, pp. 3142–3155, 2017

    work page 2017

  43. [43]

    Convolutional neural networks at cons trained time cost,

    K. He and J. Sun, “Convolutional neural networks at cons trained time cost,” in 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , Boston, MA, USA, June 2015, pp. 5353–5360

    work page 2015

  44. [44]

    Adam: A Method for Stochastic Optimization

    D. P . Kingma and J. Ba, “Adam: A method for stochastic optimization,” CoRR, vol. abs/1412.6980, 2014. [Online]. Available: https://api.semanticscholar.org/CorpusID:6628106

    work page internal anchor Pith review Pith/arXiv arXiv 2014