Pith. sign in

REVIEW 1 major objections 6 minor 45 references

Learning to Write on Dirty Paper

T0 review · 1 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read A neural encoder and decoder can learn dirty paper coding from data and outperform hand-designed precoders at low signal-to-noise ratios.

desk verdict A genuine proof-of-concept that learned encoders rediscover THP/lattice structures for dirty paper coding, but the baseline comparison is under-specified so the low-SNR gains are not yet controlled. read the letter →

arxiv 2507.17427 v1 pith:6ZHAVXYH submitted 2025-07-23 cs.IT math.IT

classification cs.ITmath.IT MSC 94A1594A40
keywords dirtypapercodingneuralencoder-decoderTomlinson-Harashimaprecodinglattice-basedsinusoidalactivationinterferencepre-cancellationsymbolerrorratelearnedmodulomapping
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

This paper claims that a neural network can perform dirty paper coding without being told the statistics of the interference, the channel, or the message. The authors train an encoder and decoder pair end-to-end on the symbol error rate, with a power penalty, and show that it learns periodic, modulo-like mappings that cancel known interference. These learned mappings reproduce the structure of Tomlinson-Harashima precoding and lattice precoding, yet they beat those classical schemes in the low-SNR regime and degrade gracefully when the test-time interference is much weaker than the training interference. If true, this is evidence that learning-based precoders can complement hand-designed dirty paper coding constructions in regimes where model assumptions break down.

What carries the argument

The load-bearing mechanism is the sinusoidal activation function in the hidden layers of both networks. Standard activations such as leaky ReLU bias networks toward smooth functions and make them underfit periodic behavior; sinusoidal activations let the encoder represent the modulo-like, many-to-one mappings that pre-cancel interference. The training objective is a cross-entropy loss on the decoded message plus a penalty $\lambda \|e_\theta(V,S)\|^2$ on transmit power, and sweeping $\lambda$ produces the SNR–SER trade-off that is compared with Tomlinson-Harashima precoding and lattice baselines.

What would settle it

Feed the trained encoder interference values spaced many periods beyond the range shown in the paper (for example, $s$ values far outside $\sigma^2_{s,\text{train}}=30$, such as $\pm 50$ or larger) and measure the symbol error rate. If the decision regions stop tiling or the symbol error rate jumps relative to the periodic extension of the learned mapping, the claimed periodicity and the robustness to unseen interference levels would be refuted.

Watch

Extended reading notes

Core claim

The central discovery is that a neural encoder and decoder, trained only on random samples of messages, interference, and noise, can rediscover the essence of dirty paper coding: instead of removing the interference, the encoder folds it into a periodic signal whose receiver-side decision regions tile the channel output space. In the one-dimensional BPSK case the learned encoder is a quasi-periodic triangular map whose slope and period adapt to the operating SNR; in the two-dimensional QPSK case the decoder's decision regions form a near-hexagonal tiling. This learned structure matches the modulo operation of Tomlinson-Harashima precoding at high SNR, where performance converges to that scheme, and outperforms Tomlinson-Harashima precoding and lattice benchmarks at low SNR, while generalizing across unseen interference variances. The paper presents this as the first interpretable proof-of-concept that a learning-based dirty paper coding scheme can recover known precoding structures and exceed them in some regimes.

Load-bearing premise

The argument that the learned encoder generalizes to any interference level assumes the encoder's mapping is truly periodic across the whole interference axis; if the network has only memorized a finite segment of the interference range, the claimed robustness beyond training would fail.

Editorial extensions

If this is right

  • At low SNR, the learned encoder's flexible triangular mapping gives lower symbol error rate than Tomlinson-Harashima precoding with scalar modulo operations.
  • As SNR grows, the learned encoder converges to Tomlinson-Harashima-like behavior, so the scheme does not sacrifice high-SNR performance.
  • Trained models remain effective when test interference variance is far below the training variance, approaching the interference-free AWGN bound as interference weakens.
  • For structured QPSK interference, the neural dirty paper coding scheme outperforms both Tomlinson-Harashima precoding and a fixed-ratio structured dirty paper coding baseline across the tested SNR range.
  • Sinusoidal activations are needed for the effect: with leaky ReLU activations the decision regions lose their structured tiling and the learned mapping degrades.

Reading between the lines

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

  • Beyond the paper: if the learned periodicity extends across the full interference axis, the same trained encoder could handle arbitrarily strong interference with no retraining, which would remove a main limitation of fixed-window hand-designed precoders.
  • Beyond the paper: a natural next test is applying the same architecture to higher-dimensional constellations, where the decoder's decision regions could reveal whether data-driven tiling discovers lattices denser than the rectangular ones in dimensions where optimal packings are unknown.
  • Beyond the paper: because training only needs samples of interference, channel output, and messages, the approach could in principle be adapted to channels with nonlinear distortion or non-Gaussian noise, where analytical dirty paper coding constructions do not exist.
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

1 major / 6 minor

Summary. The paper proposes a neural-network-based encoder and decoder for dirty paper coding over an AWGN channel with state known at the transmitter. The encoder maps the message and interference to a channel input, and the decoder infers the message from the received signal; both are trained end-to-end by minimizing cross-entropy plus an l2 power penalty (Eq. (10)). Experiments cover BPSK and QPSK messages with Gaussian and discrete QPSK interference. The authors report that sinusoidal activations lead the encoder to learn quasi-periodic, modulo-like mappings, that the proposed scheme outperforms THP and lattice-based baselines at low SNR, and that performance degrades gracefully when test interference variance is lower than the training variance.

Significance. The paper is a useful proof-of-concept: it shows that a generic learning objective (cross-entropy plus power penalty) can rediscover structured pre-cancellation mappings reminiscent of THP and lattice codes, and the sinusoidal-activation design choice is a plausible explanation. The code is publicly available, and the use of separate models per lambda with train/test mismatch evaluation is a reasonable experimental design. The main scientific claim—that the learned scheme outperforms classical THP/lattice precoding—is not yet fully supported because the baseline parameterization is underspecified. If the baseline comparison is made watertight, the paper would be a solid contribution to learning-based physical-layer design.

major comments (1)
  1. [Section IV-A, Eq. (8), Fig. 3] The comparison against THP and lattice-based baselines is not fully specified, so the reported low-SNR gains may be artifacts of non-competitive baselines. The paper does not state how the baseline transmit power P_X is set relative to the SNR axis, how the modulo base Delta in Eq. (3) or the lattice volume/scaling in Eq. (5) is chosen, or whether the scaling factor alpha in Eq. (5) is optimized at each SNR. Since the equivalent noise N' in Eq. (8) depends on alpha and on the modulo region, a fixed or arbitrary parameter choice directly changes the baseline SER. Please report the exact baseline parameterization (e.g., Delta = sqrt(12 P_X) with P_X matched to the neural model's measured power, alpha selected by MMSE) and, ideally, include a sensitivity analysis over these parameters.
minor comments (6)
  1. [Section IV-B, Figs. 3-4] The claim that the neural DPC 'generalizes well across a broad range of unseen test-time conditions' is only supported for sigma_{s,test}^2 <= sigma_{s,train}^2 = 30, i.e., for interference variances below the training variance. The learned encoder's periodicity is demonstrated only qualitatively, over a finite interval in Fig. 4; no evidence is provided that the mapping remains periodic for interference values far outside the training support. Please either evaluate at sigma_{s,test}^2 > 30 or qualify the claim as applying to lower-variance interference.
  2. [Section II-C and Fig. 3b] The label 'THP w/ QPSK input + Construction A' is misleading, since Construction A is a lattice construction rather than a variant of THP; consider renaming the baseline, e.g., 'lattice-based DPC (Construction A)'.
  3. [Section III] Training details such as learning rate, batch size, number of training samples, and the set of lambda values used for Fig. 3 are not reported; please provide them or point to the code for exact values.
  4. [Section IV-A] The AWGN lower-bound curves for BPSK/QPSK are not defined; state whether they correspond to the SER of uncoded BPSK/QPSK with a maximum-likelihood detector in the absence of interference.
  5. [Section IV-B, Figs. 2 and 4] The claims of 'hexagonal tiling' and 'quasi-periodic triangular structure' rest solely on visual inspection; a quantitative periodicity measure (e.g., autocorrelation or comparison with a lattice quantizer) would strengthen the interpretability argument.
  6. [Section II-B] The capacity expression is typeset as 'CDP C' in the manuscript; it should be C_DPC. There are also a few spacing/encoding artifacts in the author list and abstract.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the learned DPC mappings are trained from a generic cross-entropy plus power-penalty objective and are benchmarked against external THP/lattice baselines.

full rationale

The paper does not contain a self-definitional or fitted-input circularity. The neural encoder and decoder are trained by minimizing the generic objective L(θ, φ) = E[-log pφ(V|Y) + λ||eθ(V,S)||²] in Eq. (10), which contains no term that imitates THP, a lattice, or a modulo operation. The periodic and lattice-like structures shown in Figs. 2 and 4 are therefore empirical properties discovered during training, not quantities encoded into the loss or the network output by construction. The use of sinusoidal activations is disclosed as an explicit architectural choice, and the paper tests its effect by comparing with leaky ReLU activations; this is a controlled experiment rather than an assumption smuggled in through citation. The performance claims are evaluated against external baselines, including THP, Z² and Construction A lattices, and the Liu et al. scheme, and no neural model is fitted to those baseline outputs. Self-citations [25] and [27] are used only as motivation from the dual Wyner–Ziv setting, and none of the paper's DPC results depend on those prior results as premises. The low-SNR advantage over THP/lattice baselines may be sensitive to how those baselines were parameterized and calibrated, but that is an experimental-control concern, not a circular derivation. No specific reduction of a claimed result to its own inputs can be quoted; the correct finding is therefore no significant circularity.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central contribution is empirical, so the ledger is light on invented entities. The main hand-set parameters are lambda and the architecture, and the main unproved background assumptions are the classical DPC capacity result, the source-channel duality, and the approximation power of the chosen network family.

free parameters (3)
  • power-penalty coefficient lambda = values swept, e.g., 4, 5, 100
    Eq. (10) uses lambda to balance symbol error rate against transmit power; different lambda produce different SNR operating points on the reported curves. Chosen by hand, not inferred from data.
  • network hidden width and depth = 3 layers of 128 units
    Architecture hyperparameters chosen without ablation; the central results could depend on capacity.
  • training interference variance sigma_s,train^2 = 30
    Training interference power is fixed to 30; generalization to other values is tested, but the specific choice is arbitrary.
assumptions (4)
  • standard math Costa's capacity result for dirty paper coding with Gaussian state and noise
    Invoked in Sec. II-B as the theoretical benchmark; not proven in this paper.
  • domain assumption Duality between Wyner-Ziv source coding and dirty paper channel coding
    Cited from [29]-[31] in Sec. I to motivate transferring learned binning results from the source coding side; the paper relies on this equivalence without proof.
  • domain assumption Neural networks with sinusoidal activations can approximate the optimal encoder/decoder mappings for the one-shot DPC problem with adequate capacity
    The entire approach assumes the parameterized family contains a good enough solution; the paper provides no universality or convergence guarantee, only empirical evidence.
  • domain assumption Minimizing the cross-entropy loss in Eq. (10) with a soft power penalty is an adequate surrogate for minimizing SER under an average power constraint
    Standard practice in end-to-end learned communication, but the equivalence is not derived; the measured SNR is posterior to training, not a hard constraint.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning to Write on Dirty Paper." pith.science (2026). https://pith.science/paper/6ZHAVXYH

@misc{pith2026250717427,
  author       = {Pith},
  title        = {Pith review of: Learning to Write on Dirty Paper},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6ZHAVXYH}},
  note         = {Machine review of arXiv:2507.17427}
}
read the original abstract

Dirty paper coding (DPC) is a classical problem in information theory that considers communication in the presence of channel state known only at the transmitter. While the theoretical impact of DPC has been substantial, practical realizations of DPC, such as Tomlinson-Harashima precoding (THP) or lattice-based schemes, often rely on specific modeling assumptions about the input, state and channel. In this work, we explore whether modern learning-based approaches can offer a complementary path forward by revisiting the DPC problem. We propose a data-driven solution in which both the encoder and decoder are parameterized by neural networks. Our proposed model operates without prior knowledge of the state (also referred to as "interference"), channel or input statistics, and recovers nonlinear mappings that yield effective interference pre-cancellation. To the best of our knowledge, this is the first interpretable proof-of-concept demonstrating that learning-based DPC schemes can recover characteristic features of well-established solutions, such as THP and lattice-based precoding, and outperform them in several regimes.

Figures

Figures reproduced from arXiv: 2507.17427 by the authors.

Figure 1
Figure 1. The dirty paper coding problem. The transmitter maps the message V and known interference S to an input X, subject to an average power constraint. The receiver observes the channel output corrupted by additive noise N. inspired a broad line of research focused on translating DPC’s theoretical gains into practical multi-user systems [7]–[13], where DPC-based designs have been shown to significantly outperform their l… view at source ↗
Figure 2
Figure 2. Visualization of decision regions for the proposed learning-based decoder, where the message V is initially mapped to QPSK, with interference S ∼ N (0, 30) and channel noise N ∼ N (0, 1). The models in both panels are trained with λ = 5 in Eq. (10). The left figure shows the case where both the encoder and decoder use sinusoidal activation functions, resulting in a highly regular tiling reminiscent of the hexagonal … view at source ↗
Figure 3
Figure 3. Symbol error rate (SER) as a function of SNR, where the message V is initially mapped to a BPSK constellation (left) or QPSK constellation (right), and the interference S follows a Gaussian distribution. During training, the interference is drawn as S ∼ N (0, σ2 s,train = 30), and the channel noise is N ∼ N (0, 1). At test time, the same trained models are evaluated under varying interference power levels, with σ 2 … view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Visualization of encoder outputs as a function of the channel state realization, where the message V is initially mapped to a BPSK modulation (as in results provided in Fig. 3a). During training, the interference and channel noise are set as S ∼ N (0, 30) and N ∼ N (0,…
Figure 5
Figure 5. Figure 5: Symbol error rate (SER) as a function of signal-to-noise ratio (SNR), where both the message V and the interference S are drawn from a fixed QPSK constellation. During training, the interference power is selected from PS ∈ {1.5, 4.5}, and the channel noise is modeled a…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

45 extracted references · 43 canonical work pages

  1. [1]

    Coding for channel with random parameters,

    S. I. Gelfand and M. S. Pinsker, “Coding for channel with random parameters,” Probl. Pered. Inform. (Probl. Inf. Transm.) , vol. 9, no. 1, pp. 19–31, 1980

  2. [2]

    Writing on dirty paper (corresp.),

    M. Costa, “Writing on dirty paper (corresp.),” IEEE Transactions on Information Theory , vol. 29, no. 3, pp. 439–441, 1983

  3. [3]

    On the achievable throughput of a multiantenna Gaussian broadcast channel,

    G. Caire and S. Shamai, “On the achievable throughput of a multiantenna Gaussian broadcast channel,” IEEE Transactions on Information Theory , vol. 49, no. 7, pp. 1691–1706, 2003

  4. [4]

    Sum capacity of Gaussian vector broadcast channels,

    W. Yu and J. Cioffi, “Sum capacity of Gaussian vector broadcast channels,” IEEE Transactions on Information Theory , vol. 50, no. 9, pp. 1875–1892, 2004

  5. [5]

    Duality, achievable rates, and sum-rate capacity of Gaussian MIMO broadcast channels,

    S. Vishwanath, N. Jindal, and A. Goldsmith, “Duality, achievable rates, and sum-rate capacity of Gaussian MIMO broadcast channels,” IEEE Transactions on Information Theory , vol. 49, no. 10, pp. 2658–2668, 2003

  6. [6]

    Sum capacity of the vector Gaussian broadcast channel and uplink–downlink duality,

    P. Viswanath and D. Tse, “Sum capacity of the vector Gaussian broadcast channel and uplink–downlink duality,” IEEE Transactions on Information Theory , vol. 49, no. 8, pp. 1912–1921, 2003

  7. [7]

    A comparison of time-sharing, DPC, and beamforming for MIMO broadcast channels with many users,

    M. Sharif and B. Hassibi, “A comparison of time-sharing, DPC, and beamforming for MIMO broadcast channels with many users,” IEEE Transactions on Communications , vol. 55, no. 1, pp. 11–15, 2007

  8. [8]

    DPC rates and multiplexing gains for MIMO broadcast systems with multi-dimensional space-time modulation,

    M. Kassouf and H. Leib, “DPC rates and multiplexing gains for MIMO broadcast systems with multi-dimensional space-time modulation,” IEEE Transactions on Communications , vol. 59, no. 3, pp. 639–647, 2011

Show all 45 references
  1. [9]

    Uniform channel decomposition for MIMO communications,

    Y . Jiang, J. Li, and W. Hager, “Uniform channel decomposition for MIMO communications,” IEEE Transactions on Signal Processing , vol. 53, no. 11, pp. 4283–4294, 2005

  2. [10]

    Tunable channel decomposition for MIMO communications using channel state information,

    Y . Jiang, W. Hager, and J. Li, “Tunable channel decomposition for MIMO communications using channel state information,” IEEE Trans- actions on Signal Processing , vol. 54, no. 11, pp. 4405–4418, 2006

  3. [11]

    Block-diagonal geometric mean decomposition (BD-GMD) for multiuser MIMO broadcast channels,

    S. Lin, W. W. L. Ho, and Y .-c. Liang, “Block-diagonal geometric mean decomposition (BD-GMD) for multiuser MIMO broadcast channels,” 2006 IEEE 17th International Symposium on Personal, Indoor and Mobile Radio Communications , pp. 1–5, 2006

  4. [12]

    MIMO transceiver design via majorization theory,

    D. P. Palomar and Y . Jiang, “MIMO transceiver design via majorization theory,” F oundations and Trends in Communications and Information Theory, vol. 3, no. 4-5, pp. 331–551, 2007

  5. [13]

    Revisiting dirty paper coding to hybrid precoding for massive MIMO downlink broadcast channel,

    X. Su and Y . Jiang, “Revisiting dirty paper coding to hybrid precoding for massive MIMO downlink broadcast channel,” IEEE Transactions on Communications, vol. 71, no. 7, pp. 4079–4090, 2023

  6. [14]

    New automatic equaliser employing modulo arithmetic,

    M. Tomlinson, “New automatic equaliser employing modulo arithmetic,” Electronics Letters, vol. 7, pp. 138–139, 1971

  7. [15]

    Matched-transmission technique for channels with intersymbol interference,

    H. Harashima and H. Miyakawa, “Matched-transmission technique for channels with intersymbol interference,” IEEE Transactions on Commu- nications, vol. 20, no. 4, pp. 774–780, 1972

  8. [16]

    Achievable rates for Tomlinson-Harashima precoding,

    R. Wesel and J. Cioffi, “Achievable rates for Tomlinson-Harashima precoding,” IEEE Transactions on Information Theory , vol. 44, no. 2, pp. 824–831, 1998

  9. [17]

    Tomlinson- Harashima precoding in space-time transmission for low-rate backward channel,

    R. Fischer, C. Windpassinger, A. Lampe, and J. Huber, “Tomlinson- Harashima precoding in space-time transmission for low-rate backward channel,” 2002 International Zurich Seminar on Broadband Communi- cations Access - Transmission - Networking , pp. 7–7, 2002

  10. [18]

    Multi-branch Tomlinson- Harashima precoding design for MU-MIMO systems: Theory and al- gorithms,

    K. Zu, R. C. de Lamare, and M. Haardt, “Multi-branch Tomlinson- Harashima precoding design for MU-MIMO systems: Theory and al- gorithms,” IEEE Transactions on Communications , vol. 62, no. 3, pp. 939–951, 2014

  11. [19]

    Tomlinson-Harashima precoding for broadcast channels with uncertainty,

    M. B. Shenouda and T. N. Davidson, “Tomlinson-Harashima precoding for broadcast channels with uncertainty,” IEEE Journal on Selected Areas in Communications , vol. 25, no. 7, pp. 1380–1389, 2007

  12. [20]

    A multi-user precoding scheme achieving crosstalk cancellation with application to DSL systems,

    G. Ginis and J. Cioffi, “A multi-user precoding scheme achieving crosstalk cancellation with application to DSL systems,” The Thirty- F ourth Asilomar Conference on Signals, Systems and Computers , pp. 1627–1631 vol.2, 2000

  13. [21]

    Multiuser diversity for a dirty paper approach,

    Z. Tu and R. Blum, “Multiuser diversity for a dirty paper approach,” IEEE Communications Letters , vol. 7, no. 8, pp. 370–372, 2003

  14. [22]

    The intersymbol interference channel: lower bounds on capacity and channel precoding loss,

    S. Shamai and R. Laroia, “The intersymbol interference channel: lower bounds on capacity and channel precoding loss,” IEEE Transactions on Information Theory , vol. 42, no. 5, pp. 1388–1404, 1996

  15. [23]

    A close-to-capacity dirty paper coding scheme,

    U. Erez and S. ten Brink, “A close-to-capacity dirty paper coding scheme,” IEEE Transactions on Information Theory , vol. 51, no. 10, pp. 3417–3432, 2005

  16. [24]

    Neural distributed source coding,

    J. Whang, A. Nagle, A. Acharya, H. Kim, and A. G. Dimakis, “Neural distributed source coding,” IEEE Journal on Selected Areas in Informa- tion Theory , vol. 5, pp. 493–508, 2024

  17. [25]

    Neural distributed compressor discovers binning,

    E. Özyılkan, J. Ballé, and E. Erkip, “Neural distributed compressor discovers binning,” IEEE Journal on Selected Areas in Information Theory, vol. 5, pp. 246–260, 2024

  18. [26]

    Distributed compression in the era of machine learning: A review of recent advances,

    E. Özyılkan and E. Erkip, “Distributed compression in the era of machine learning: A review of recent advances,” 2024 58th Annual Conference on Information Sciences and Systems (CISS) , pp. 1–6, 2024

  19. [27]

    Learned Wyner–Ziv compressors recover binning,

    E. Özyılkan, J. Ballé, and E. Erkip, “Learned Wyner–Ziv compressors recover binning,” 2023 IEEE International Symposium on Information Theory (ISIT) , pp. 701–706, 2023

  20. [28]

    The rate–distortion function for source coding with side information at the decoder,

    A. Wyner and J. Ziv, “The rate–distortion function for source coding with side information at the decoder,” IEEE Transactions on Information Theory, vol. 22, no. 1, pp. 1 – 10, 1976

  21. [29]

    Duality between source coding and channel coding and its extension to the side information case,

    S. Pradhan, J. Chou, and K. Ramchandran, “Duality between source coding and channel coding and its extension to the side information case,” IEEE Transactions on Information Theory , vol. 49, no. 5, pp. 1181–1203, 2003

  22. [30]

    Fixed binning schemes: an operational dual- ity between channel and source coding problems with side information,

    H. Wang and P. Viswanath, “Fixed binning schemes: an operational dual- ity between channel and source coding problems with side information,” 2004 IEEE International Symposium on Information Theory (ISIT) , pp. 197–, 2004

  23. [31]

    Joint Wyner–Ziv/dirty-paper coding by modulo-lattice modulation,

    Y . Kochman and R. Zamir, “Joint Wyner–Ziv/dirty-paper coding by modulo-lattice modulation,” IEEE Transactions on Information Theory , vol. 55, no. 11, pp. 4878–4889, 2009

  24. [32]

    Channel coding in the presence of side information,

    G. Keshet, Y . Steinberg, and N. Merhav, “Channel coding in the presence of side information,” F oundations and Trends in Communications and Information Theory , vol. 4, no. 6, pp. 445–586, 2008

  25. [33]

    On the spectral bias of neural networks,

    N. Rahaman, A. Baratin, D. Arpit, F. Draxler, M. Lin, F. Hamprecht, Y . Bengio, and A. Courville, “On the spectral bias of neural networks,” Proceedings of the 36th International Conference on Machine Learning , vol. 97, pp. 5301–5310, 2019

  26. [34]

    Frequency bias in neural networks for input of non-uniform density,

    R. Basri, M. Galun, A. Geifman, D. Jacobs, Y . Kasten, and S. Kritchman, “Frequency bias in neural networks for input of non-uniform density,” in Proceedings of the 37th International Conference on Machine Learning , 2020

  27. [35]

    Neural networks fail to learn periodic functions and how to fix it,

    L. Ziyin, T. Hartwig, and M. Ueda, “Neural networks fail to learn periodic functions and how to fix it,” in Advances in Neural Information Processing Systems (NeurIPS) , 2020

  28. [36]

    The Gaussian watermarking game,

    A. Cohen and A. Lapidoth, “The Gaussian watermarking game,” IEEE Transactions on Information Theory , vol. 48, no. 6, pp. 1639–1667, 2002

  29. [37]

    Precoding in multiantenna and multiuser communications,

    C. Windpassinger, R. Fischer, T. Vencel, and J. Huber, “Precoding in multiantenna and multiuser communications,” IEEE Transactions on Wireless Communications, vol. 3, no. 4, pp. 1305–1316, 2004

  30. [38]

    Nested linear/lattice codes for structured multiterminal binning,

    R. Zamir, S. Shamai, and U. Erez, “Nested linear/lattice codes for structured multiterminal binning,” IEEE Transactions on Information Theory, vol. 48, no. 6, pp. 1250–1276, 2002

  31. [39]

    Zamir, B

    R. Zamir, B. Nazer, Y . Kochman, and I. Bistritz, Lattice Coding for Signals and Networks: A Structured Coding Approach to Quantization, Modulation and Multiuser Information Theory . Cambridge University Press, 2014

  32. [40]

    JAX: Composable transformations of Python+NumPy programs,

    J. Bradbury, R. Frostig, P. Hawkins, M. J. Johnson, C. Leary, D. Maclaurin, G. Necula, A. Paszke, J. VanderPlas, S. Wanderman-Milne, and Q. Zhang, “JAX: Composable transformations of Python+NumPy programs,” 2018. [Online]. Available: http://github.com/google/jax

  33. [41]

    Adam: A method for stochastic optimization,

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” in International Conference on Learning Representations , 2015

  34. [42]

    J. H. Conway and N. J. A. Sloane, Sphere Packings, Lattices and Groups. New York: Springer-Verlag, 1988

  35. [43]

    Structured dirty paper coding with known interference structure at receiver,

    B. Liu, H. Liu, and S. Roy, “Structured dirty paper coding with known interference structure at receiver,”The Thirty-Ninth Asilomar Conference on Signals, Systems and Computers , pp. 643–647, 2005

  36. [44]

    The sphere packing problem in dimension 24,

    H. Cohn, A. Kumar, S. D. Miller, D. Radchenko, and M. Viazovska, “The sphere packing problem in dimension 24,” Annals of Mathematics , vol. 185, no. 3, pp. 1017–1033, 2017

  37. [45]

    A. E. Gamal and Y .-H. Kim, Network Information Theory . USA: Cambridge University Press, 2012

Pith tools

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