Pith. sign in

REVIEW 4 major objections 6 minor 38 references

Towards a Foundation Model for Communication Systems

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A single transformer model estimates five 5G wireless channel features after masked pretraining on unlabeled data.

desk verdict The tokenization design is a genuine contribution, but the empirical evaluation is circular because the target labels come from the authors' own classical estimators and the inputs encode those same estimates. read the letter →

arxiv 2505.14603 v1 pith:LQ3JI7GF submitted 2025-05-20 cs.AI cs.LGeess.SP

Davide Buffelli , Sowmen Das , Yu-Wei Lin , Sattar Vakili , Chien-Yi Wang , Masoud Attarifar , Pritthijit Nath , Da-shan Shiu This is my paper · ORCID
classification cs.AIcs.LGeess.SP
keywords foundationmodelcommunicationsystemstransformermaskedpre-trainingchannelstateinformationMIMO-OFDMDopplerspreadestimationdelayprofile
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 tries to establish that one transformer-based model can act as a general-purpose learner for low-level wireless data, rather than a dedicated model per task. The authors design tokenization and normalization schemes that let a single network consume categorical, scalar, vector, matrix, and complex-valued features of varying sizes, then pre-train it by masking target features and reconstructing them. They report that the resulting model successfully estimates the transmission rank, the selected precoder, the Doppler spread, and the center and length of the delay profile from simulated 5G New Radio multiple-input multiple-output orthogonal frequency-division multiplexing (MIMO-OFDM) slots, in both forecasting and interpolation settings. If this holds, it points toward foundation models for communication systems that are trained on unlabeled channel data and then reused for several estimation tasks.

What carries the argument

The machinery is a transformer over tokens, where each communication feature is mapped into one or more tokens by a feature-specific tokenizer: scalars become Fourier-encoded coordinates, vectors are linearly projected, matrices are cut into patches that are linearly flattened, and categorical values get learned vectors. Complex-valued features are split into real and imaginary parts before tokenization. Every token carries a positional embedding shared by all features in the same slot, plus a learned feature embedding that tells the model which feature a token belongs to across slots. Pre-training randomly masks one of the five target features per slot and trains the model to reconstruct it through simple linear decoders, forcing the representations to capture cross-feature dependencies. The same slot-level tokens are then used for both forecasting and interpolation at evaluation.

What would settle it

Take the trained model and evaluate it on out-of-distribution slots—different channel types, SNR values outside [0,30] dB, or measured hardware captures—and compare the forecasting and interpolation MSEs to Table 5; a sharp rise for Doppler spread or delay-profile errors would show the model memorized the simulator's rectangular-profile assumptions rather than learning a general representation of wireless channels.

Watch

Extended reading notes

Core claim

The central claim is that masked pre-training on unlabeled channel data teaches a single transformer to invert the relationships among heterogeneous wireless features. Concretely, the paper shows that a transformer with per-feature tokenizers, shared slot-level positional embeddings, and feature embeddings can predict the next slot's value or a masked value of five target features—transmission rank, selected precoder, Doppler spread, and delay-profile center and length—with forecasting mean squared errors of 0.019 for delay-profile center, 0.021 for delay-profile length, 0.077 for Doppler spread, 0.129 for rank, and 0.101 for precoder on standardized scales, and slightly lower errors in interpolation. The rank and precoder, being discrete and matrix-valued, are the hardest; the scalar channel parameters are estimated most accurately. The paper interprets this as evidence that a single general-purpose model, not a set of task-specific networks, can carry out multiple CSI-acquisition subtasks.

Load-bearing premise

The load-bearing premise is that channels simulated with rectangular delay and Doppler profiles, a hand-set 3-sigma threshold for delay-profile detection, and predefined candidate sets for Doppler width and delay parameters faithfully represent real 5G deployment conditions; if the simulator differs from reality, the model has only learned to invert its own classical estimators on synthetic data.

Editorial extensions

If this is right

  • A single pre-trained backbone can perform both next-slot forecasting and within-slot interpolation for five distinct channel features, so a bank of task-specific estimators is not needed for these CSI-acquisition subtasks.
  • The highest estimation accuracy is achieved for scalar channel parameters (Doppler width, delay-profile center and length), while the matrix-valued precoder and discrete rank are hardest, quantifying where a general model still lags.
  • Test loss decreases as model size and dataset size grow together, so the same masked-prediction objective exhibits scaling behavior that supports continued investment in larger models and more data.
  • Because the target features are estimated from the remaining features in the data, the pre-training objective requires no manual labels, only unlabeled slots.

Reading between the lines

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

  • A testable extension is to freeze the pre-trained backbone and attach new lightweight heads for other physical-layer tasks, such as direct channel estimation or CSI compression; the simple linear decoders used here suggest representation quality carries most of the predictive power.
  • The lack of out-of-distribution evaluation means the strongest check is to train on one set of channel conditions (one channel type, SNR range, or candidate set) and test on another, or on measured channel data; the paper's reported errors do not yet establish that transfer.
  • The tokenization is feature-agnostic, so the same architecture could absorb additional features beyond the five targets—for example, precoder codebooks of different ranks or different antenna configurations—by adding tokens and feature embeddings, which points toward a truly multimodal physical-layer model.
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. The paper proposes a transformer-based foundation model for communication systems, operating directly on heterogeneous low-level wireless data. It introduces tokenization schemes for scalar, vector, matrix, and categorical features, a feature-embedding mechanism to distinguish features across slots, and a masked-token pretraining objective. The model is evaluated on forecasting and interpolation of five features: transmission rank, selected precoder, Doppler spread, and delay profile center and length, using a simulated 5G-NR dataset generated with SIONNA. Scaling experiments with model size, dataset size, and compute are also reported.

Significance. If the empirical claims were supported, the paper would be a useful early step toward general-purpose models for wireless physical-layer data, and the tokenization and feature-embedding design could inform future work. The authors identify a genuine gap—most prior ML-for-communications work is task-specific—and propose a reasonable architecture. However, the current evaluation does not establish that the model estimates physical channel features: the target labels are the outputs of the paper's own classical estimators, and the same estimates are also present among the input features. In addition, there are no baseline comparisons, no genie-ground-truth evaluation, and no out-of-distribution or hardware validation. The scaling results also rest on an unexplained dataset-size inconsistency. The architectural contributions may be salvageable, but the reported evidence does not currently justify the central claim.

major comments (4)
  1. [§4, Table 1, §3.1.3–§3.1.6] The five target features are the outputs of the classical estimators described in Sections 3.1.3 to 3.1.6, and the same estimated quantities are also present among the model's input features, directly or through derived matrices. For instance, Rf,robust is constructed from mu-hat and ell-hat, and Rt,robust from w-hat. Thus, when a slot is masked during the interpolation evaluation, the model can recover the masked target by inverting the deterministic mapping from the remaining input features rather than by estimating a physical channel parameter. The paper defines genie values for center, length, and width but never uses them as evaluation targets. I request an evaluation against these genie values, or a masking scheme that also removes all input features derived from the target, and a re-statement of the claim accordingly.
  2. [§5.2, Table 5] The estimation results are reported without any baseline comparisons, including the classical estimators of Section 3. Because all features are normalized, the low MSE values cannot be interpreted in absolute terms; for the rank and precoder, even a trivial baseline may achieve low normalized error, and the paper gives no error rates. The central claim of successful multi-task estimation requires comparison with the Section 3 algorithms, or a reasonable classical baseline, on the same test set.
  3. [§5.1 vs §5.3, Figures 2–3] The scaling experiments in Section 5.3 state that the model was trained on 10^6, 5·10^6, and 10^7 training examples, but Section 5.1 states that the whole dataset consists of 1 million datapoints, split 80/10/10 into training, validation, and test sets. Unless additional data were generated separately for the scaling study, the stated dataset sizes are inconsistent with the total data budget. Please clarify the actual data generation and dataset sizes, because the scaling conclusions in Figures 2 and 3 depend on this.
  4. [§5.2, Table 5, Figures 2–3] All quantitative results are presented as point estimates without error bars, confidence intervals, or repeated runs over seeds or evaluation batches. Since the interpolation protocol randomly selects a slot to mask in each batch, the reported MSE is itself a random quantity. Please provide variance information or repeated-run statistics to support the comparisons across models and dataset sizes.
minor comments (6)
  1. [Table 4] The model configuration table appears to contain implausible values (e.g., 75 layers for the 100M model with token dimension 429); please verify the numbers and clarify how parameter counts are computed.
  2. [§5.2] The statement that the reported MSE values 'can be interpreted as percentage errors' is not correct: standardized features have arbitrary scale, so the MSE is not a percentage error.
  3. [§4, Figure 1] Please clarify how the feature embedding is concatenated with the token (e.g., along the token dimension) and how positional embeddings are combined with tokens before the concatenation, since the textual description and figure are ambiguous.
  4. [References] The arXiv identifier for Nikbakht et al. (TSPEC-LLM) appears to contain a typo: '2406.1768' should likely be '2406.01768'.
  5. [Figures 2 and 3] The axis labels 'compute used for data generation' and 'training compute' need explicit units (e.g., FLOPs, GPU-seconds, or number of slots generated) to make the scaling trends reproducible.
  6. [§3.2] Please state the total simulation budget implied by 0.5 seconds per slot and the number of slots in the dataset, and clarify how the 1 million datapoints were generated within that budget.

Circularity Check

3 steps flagged · score 6.0 of 10

Three of five 'estimations' reduce to inverting the paper's own classical estimators encoded in the input features; genie values are never used as targets.

  1. self definitional [Section 3.1.3 (Delay Profile Estimation) and Table 1; evaluated in Section 5.2 Table 5]
    "Finally, we have ˆµ[i] = 1/(M fscNFFT) ((ˆnstart[i] + ˆnend[i])/2), ˆℓ[i] = 1/(M fscNFFT) (ˆnstart[i] − ˆnend[i] + 1), and the estimated correlation matrix ˆRf,robust with [ ˆRf,robust]m1,m2 = e−j2π ˆµ[i](m1−m2)M fsc sin(π ˆℓ[i](m1 − m2)M fsc)/(π ˆℓ[i](m1 − m2)M fsc)."

    Table 1 lists R̂f,robust as an input feature and μ̂/ℓ̂ as target features. The displayed formula defines R̂f,robust as a closed-form deterministic function of μ̂[i] and ℓ̂[i] for known M, f_sc, N_FFT. In the interpolation evaluation, only μ̂/ℓ̂ are masked at the chosen slot, while R̂f,robust at that slot remains in the input. The transformer can therefore recover the masked values by inverting this known map, rather than by estimating a physical delay profile. The paper computes genie μ[i] and ℓ[i] in the same subsection but never uses them as evaluation targets; Table 5 is explicitly in terms of the hats. The low MSE for center/length thus only shows that the model imitates the authors' own classical estimator on in-distribution synthetic data.

  2. fitted input called prediction [Section 3.1.4 (Doppler Spectrum Estimation) and Table 1; evaluated in Section 5.2 Table 5]
    "The corresponding estimated time correlation matrix is denoted by ˆRtime[i], which can be calculated from the entries of ˆCtime by rij = cij/√ciicjj. Then, the width w can be found by ˆw[i] = arg min w ∥Rt,robust(w) − ˆRtime[i]∥2 F, where [Rt,robust(w)]l1,l2 = sin(πw(l1 − l2)T)/(πw(l1 − l2)T)."

    Table 1 includes R̂time (and Ĉtime) as input features, while ŵ is one of the five masked target features. By the quoted procedure, ŵ is defined as the minimizer of a Frobenius-norm fit between the parametric sinc correlation matrix and the input R̂time, over the predefined candidate set introduced in Section 3.2. When ŵ is masked, R̂time is still present at that slot, so the target is the argmin of a known function of an available input. The model is therefore learning to invert the paper's grid-search/estimator rule rather than estimating a true Doppler spread; the genie width w[i] is computed in the same subsection but not used as the evaluation target.

1 more flagged steps
  1. self definitional [Table 1 and Section 4 (Pre-Training Procedure); evaluated in Section 5.2 Table 5]
    "Table 1 lists 'ˆW(R) Selected precoder of rankR' and 'ˆR Selected transmission rank' among the input features; Section 4 then says: 'we identify a subset of five target features from Table 1—namely, the transmission rank, selected precoder, Doppler spectrum, center and length of the delay profile.'"

    The same table that defines the model's inputs also defines the targets, and the rank and precoder are definitionally linked: a selected precoder W^(R) lives in C^{N_T×R}. In the interpolation protocol only the chosen feature is masked at one slot, so when rank is the target the same-slot precoder Ŵ remains visible, and when precoder is the target the same-slot rank R̂ remains visible. Since the paper zero-pads matrix features and supplies padding masks, the unpadded dimension (hence rank) is in principle readable from the visible precoder, and the rank value restricts the codebook for Ŵ. Thus the reported rank/precoder 'estimations' are partly readoffs from same-slot features that are themselves estimator outputs, not independent predictions against physical ground truth.

full rationale

The central claim—'we empirically demonstrate that such a model can successfully estimate multiple features, including transmission rank, selected precoder, Doppler spread, and delay profile'—is substantially undermined by construction. Table 5 evaluates against the paper's own estimated quantities (μ̂, ℓ̂, ŵ, R̂, Ŵ), not against the genie values μ, ℓ, w that the simulator computes and that are described in Sections 3.1.3 and 3.1.4. For delay-profile parameters, the input feature R̂f,robust is defined, by the paper's own equation, as a closed-form function of μ̂ and ℓ̂; masking μ̂/ℓ̂ therefore leaves their values recoverable by inverting that function. For Doppler spread, ŵ is the argmin of a known fitting rule applied to the input R̂time, so masking ŵ still leaves the input that determines it. The rank/precoder pair is also mutually available as same-slot input features. The low MSEs in Table 5 therefore demonstrate that a transformer can learn the inverse of the paper's own classical estimation rules on in-distribution SIONNA data generated under rectangular delay/Doppler profiles and predefined candidate sets; they do not demonstrate estimation of physical channel conditions. No out-of-distribution or hardware validation is reported, so the transferability of the 'foundation model' claim cannot be checked. The paper's own statement that the target features 'can be estimated from the remaining ones, as discussed in Section 3' confirms the design but does not rescue the estimation claim. Self-citation plays no role in this circularity; the reduction is internal to the data-generation and evaluation pipeline.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central evaluation depends on hand-chosen thresholds and unreported candidate sets, and the simulation assumptions (rectangular delay/Doppler, Gaussian noise, SIONNA fidelity) are inherited without external validation. The model itself is a standard transformer with no new physical entities.

free parameters (4)
  • Delay-profile detection threshold = 3 * sigma_i^2
    In Section 3.1.3, only channel taps with power above 3 times the estimated noise variance are kept; this threshold is hand-chosen with no sensitivity analysis, and it directly determines the reported delay center and length targets.
  • Fourier encoding wavelength range = unspecified
    In Section 4, scalar features are encoded as [cos(2*pi*x/lambda_i), sin(2*pi*x/lambda_i)] with logarithmically spaced lambda_i chosen based on the scale and variability of the feature; the exact range is not reported, making the representation a tuned design choice.
  • Matrix patch-size heuristic = max 64 tokens, min 8x8 patch
    In Section 4, matrices are split into patches using a heuristic (at most 64 tokens, minimum patch 8x8); no ablation is provided, and the heuristic affects the model's ability to represent matrix features.
  • Candidate sets for Doppler width, delay center, and delay length = not listed
    In Section 3.2, the parameters w, mu, and length are selected from predefined candidate sets; the sets are not specified, so the target features are discretized in an unreported way.
assumptions (5)
  • domain assumption Channel matrix H is wide-sense stationary with zero mean and unit power per entry
    Stated in Section 3.1 as the channel model from which the simulator generates data; it is an idealization of real channels.
  • domain assumption Delay profile and Doppler spectrum are rectangular
    Sections 3.1.3 and 3.1.4 assume rectangular shapes for robust channel estimation, based on Li et al. (1998); this restricts the channel statistics the model is trained on.
  • domain assumption Noise is i.i.d. circularly-symmetric complex Gaussian with covariance C_n
    Stated in Section 3.1; standard assumption in MIMO-OFDM analysis.
  • domain assumption Same precoder is applied to all subcarriers and symbols within a slot
    Stated in Section 3.1; this simplifies the model and makes the precoder a per-slot target.
  • domain assumption SIONNA simulator faithfully reproduces 5G NR propagation
    The dataset is generated with SIONNA (Hoydis et al., 2022) under 3GPP TS 38.211 numerology; the paper provides no validation against measured channels.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards a Foundation Model for Communication Systems." pith.science (2026). https://pith.science/paper/LQ3JI7GF

@misc{pith2026250514603,
  author       = {Pith},
  title        = {Pith review of: Towards a Foundation Model for Communication Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LQ3JI7GF}},
  note         = {Machine review of arXiv:2505.14603}
}
read the original abstract

Artificial Intelligence (AI) has demonstrated unprecedented performance across various domains, and its application to communication systems is an active area of research. While current methods focus on task-specific solutions, the broader trend in AI is shifting toward large general models capable of supporting multiple applications. In this work, we take a step toward a foundation model for communication data--a transformer-based, multi-modal model designed to operate directly on communication data. We propose methodologies to address key challenges, including tokenization, positional embedding, multimodality, variable feature sizes, and normalization. Furthermore, we empirically demonstrate that such a model can successfully estimate multiple features, including transmission rank, selected precoder, Doppler spread, and delay profile.

Figures

Figures reproduced from arXiv: 2505.14603 by the authors.

Figure 1
Figure 1. Scheme of our foundation model for communications data and the pre-training procedure. Tokens for each feature are first added to a positional embedding; features at the same input slot receive the same positional embedding (indicated by the color of the arrows). A feature embedding is then concatenated (&); the tokens for the same feature at different slots receive the same feature embedding. The transformer proces… view at source ↗
Figure 2
Figure 2. Scaling behaviour vs. data generation compute. Each curve shows the test loss for a model of a given size (5M, 30M, or 100M parameters) as a function of compute used for data genera￾tion. Among the features, Doppler spectrum width (wˆ) and delay profile parameters (µˆ, ˆℓ), which are all scalar-valued, are estimated with the highest accuracy. In contrast, transmis￾sion rank (Rˆ) and especially the selected precoder … view at source ↗
Figure 3
Figure 3. Scaling behaviour vs. training compute. Each curve shows the test loss for models of a given size (5M, 30M, 100M) as a function of compute used for training. 6. Conclusion In this work, we laid the groundwork for a foundation model for communication systems. We discussed and addressed the unique challenges involved in designing such a model, developed a simulation system to generate training data, and demonstrated t… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 17 canonical work pages

  1. [1]

    NR; Physical Channels and Modulation, 3GPP Standard TS 38.211, 2018

    3GPP. NR; Physical Channels and Modulation, 3GPP Standard TS 38.211, 2018

    work page 2018

  2. [2]

    Abbaspourazad, S. et al. Large-scale training of foundation models for wearable biosignals. In International Conference on Learning Representations, 2024. URL https://openreview.net/forum?id=pC3WJHf51j

    work page 2024

  3. [3]

    Alayrac, J.-B. et al. Flamingo: a visual language model for few-shot learning. In Advances in Neural Information Processing Systems, 2022. URL https://openreview.net/forum?id=EbMuimAbPbs

    work page 2022

  4. [4]

    Bodnar, C. et al. A foundation model for the earth system. arXiv, 2024. URL https://arxiv.org/abs/2405.13063. arxiv:2405.13063

    arXiv 2024

  5. [5]

    Bommasani, R. et al. On the opportunities and risks of foundation models. arxiv, 2021. URL https://arxiv.org/abs/2108.07258. arxiv:2108.07258

    arXiv 2021

  6. [6]

    D., Piovesan, N., and Maatouk, A

    Bornea, A.-L., Ayed, F., Domenico, A. D., Piovesan, N., and Maatouk, A. Telco-rag: Navigating the challenges of retrieval-augmented language models for telecommunications. arxiv, 2024. URL https://arxiv.org/abs/2404.15939. arxiv:2404.15939

    arXiv 2024

  7. [7]

    Chen, Z. et al. Expanding performance boundaries of open-source multimodal models with model, data, and test-time scaling. arxiv, 2025. URL https://arxiv.org/abs/2412.05271. arxiv:2412.05271

    arXiv 2025

  8. [8]

    Dai, W. et al. Nvlm: Open frontier-class multimodal LLMs . arxiv, 2024. URL https://arxiv.org/abs/2409.11402. arxiv:2409.11402

    arXiv 2024

Show all 38 references
  1. [9]

    A decoder-only foundation model for time-series forecasting

    Das, A., Kong, W., Sen, R., and Zhou, Y. A decoder-only foundation model for time-series forecasting. arxiv, 2024. URL https://arxiv.org/abs/2310.10688. arxiv:2310.10688

    2024 arXiv

  2. [10]

    BERT : Pre-training of deep bidirectional transformers for language understanding

    Devlin, J., Chang, M.-W., Lee, K., and Toutanova, K. BERT : Pre-training of deep bidirectional transformers for language understanding. In Conference of the North A merican Chapter of the Association for Computational Linguistics: Human Language Technologies , pp.\ 4171--4186....

    2019 doi

  3. [11]

    Dosovitskiyet, A. et al. An image is worth 16x16 words: Transformers for image recognition at scale. In International Conference on Learning Representations, 2021. URL https://openreview.net/forum?id=YicbFdNTTy

    2021

  4. [12]

    Ekambaram, V. et al. Tiny time mixers ( TTM s): Fast pre-trained models for enhanced zero/few-shot forecasting of multivariate time series. In Conference on Neural Information Processing Systems, 2024. URL https://openreview.net/forum?id=3O5YCEWETq

    2024

  5. [13]

    Moment: A family of open time-series foundation models

    Goswami, M., Szafer, K., Choudhry, A., Cai, Y., Li, S., and Dubrawski, A. Moment: A family of open time-series foundation models. In International Conference on Machine Learning, 2024

    2024

  6. [14]

    Grattafiori, A. et al. The Llama 3 herd of models. arxiv, 2024. URL https://arxiv.org/abs/2407.21783. arxiv:2407.21783

    2024 arXiv

  7. [15]

    Hamdan, M. Q. et al. Recent advances in machine learning for network automation in the o-ran. Sensors, 23 0 (21), 2023. ISSN 1424-8220. doi:10.3390/s23218792. URL https://www.mdpi.com/1424-8220/23/21/8792

    2023 doi

  8. [16]

    Hoydis, J. et al. Sionna: An open-source library for next-generation physical layer research. arxiv, 2022. URL https://arxiv.org/abs/2203.11854. arxiv:2203.118548

    2022 arXiv

  9. [17]

    Lewis, P. et al. Retrieval-augmented generation for knowledge-intensive nlp tasks. In Proceedings of the 34th International Conference on Neural Information Processing Systems, NIPS '20, Red Hook, NY, USA, 2020. Curran Associates Inc. ISBN 9781713829546

    2020

  10. [18]

    Robust channel estimation for ofdm systems with rapid dispersive fading channels

    Li, Y., Cimini, L., and Sollenberger, N. Robust channel estimation for ofdm systems with rapid dispersive fading channels. IEEE Transactions on Communications, 46 0 (7): 0 902--915, 1998. doi:10.1109/26.701317

    1998 doi

  11. [19]

    Deep learning methods in communication systems: A review

    Liao, F., Wei, S., and Zou, S. Deep learning methods in communication systems: A review. Journal of Physics: Conference Series, 1617 0 (1): 0 012024, aug 2020. doi:10.1088/1742-6596/1617/1/012024. URL https://dx.doi.org/10.1088/1742-6596/1617/1/012024

    2020 doi

  12. [20]

    Liu, X. et al. Moirai-moe: Empowering time series foundation models with sparse mixture of experts. arxiv, 2024. URL https://arxiv.org/abs/2410.10469. arxiv:2410.10469

    2024 arXiv

  13. [21]

    Liu, Y. et al. Sundial: A family of highly capable time series foundation models. arxiv, 2025. URL https://arxiv.org/abs/2502.00816. arxiv:2502.00816

    2025 arXiv

  14. [22]

    Narayanswamy, G. et al. Scaling wearable foundation models. In International Conference on Learning Representations, 2025. URL https://openreview.net/forum?id=yb4QE6b22f

    2025

  15. [23]

    Tspec-llm: An open-source dataset for llm understanding of 3gpp specifications

    Nikbakht, R., Benzaghta, M., and Geraci, G. Tspec-llm: An open-source dataset for llm understanding of 3gpp specifications. arxiv, 2024. URL https://arxiv.org/abs/2406.01768. arxiv:2406.1768

    2024 arXiv

  16. [24]

    Openai o1 system card, 2024

    OpenAI. Openai o1 system card, 2024. https://openai.com/index/openai-o1-system-card/

    2024

  17. [25]

    D., and Ayed, F

    Piovesan, N., Domenico, A. D., and Ayed, F. Telecom language models: Must they be large? arxiv, 2024. URL https://arxiv.org/abs/2403.04666. arxiv:2403.04666

    2024 arXiv

  18. [26]

    Rasul, K. et al. Lag-Llama : Towards foundation models for probabilistic time series forecasting. arxiv, 2024. arxiv:2310.08278

    2024 arXiv

  19. [27]

    Saad, W. et al. Artificial general intelligence (agi)-native wireless systems: A journey beyond 6g. arxiv, 2024. URL https://arxiv.org/abs/2405.02336. arxiv:2405.02336

    2024 arXiv

  20. [28]

    Shao, J. et al. Wirelessllm: Empowering large language models towards wireless intelligence. Journal of Communications and Information Networks, 9 0 (2): 0 99--112, 2024. doi:10.23919/JCIN.2024.10582827

    2024

  21. [29]

    Time-moe: Billion-scale time series foundation models with mixture of experts

    Shi, X., Wang, S., Nie, Y., Li, D., Ye, Z., Wen, Q., and Jin, M. Time-moe: Billion-scale time series foundation models with mixture of experts. In International Conference on Learning Representations, 2025. URL https://openreview.net/forum?id=e1wDDFmlVu

    2025

  22. [30]

    Machine learning in communication systems and networks

    Sun, Y., Lee, H., and Simpson, O. Machine learning in communication systems and networks. Sensors, 24 0 (6), 2024. ISSN 1424-8220. doi:10.3390/s24061925. URL https://www.mdpi.com/1424-8220/24/6/1925

    2024 doi

  23. [31]

    Team, G. et al. Gemini: A family of highly capable multimodal models. arxiv, 2024. URL https://arxiv.org/abs/2312.11805. arxiv:2312.11805

    2024 arXiv

  24. [32]

    Emergence of deep learning as a potential solution for detection, recovery and de-noising of signals in communication systems

    Thakkar, K., Goyal, A., and Bhattacharyya, B. Emergence of deep learning as a potential solution for detection, recovery and de-noising of signals in communication systems. International Journal of Intelligent Networks, 1: 0 119--127, 2020. ISSN 2666-6030. doi:https://doi.org/...

    2020 doi

  25. [33]

    and Viswanath, P

    Tse, D. and Viswanath, P. Fundamentals of wireless communication. Cambridge University Press, USA, 2005. ISBN 0521845270

    2005

  26. [34]

    and Li, G

    Wang, S. and Li, G. Y. Machine learning in communications: A road to intelligent transmission and processing. arxiv, 2024. URL https://arxiv.org/abs/2407.11595. arxiv:2407.11595

    2024 arXiv

  27. [35]

    Wang, Y. et al. Transformer-empowered 6g intelligent networks: From massive mimo processing to semantic communication. Wireless Communications, 30 0 (6): 0 127–135, December 2023. ISSN 1536-1284. doi:10.1109/MWC.008.2200157. URL https://doi.org/10.1109/MWC.008.2200157

    2023 doi

  28. [36]

    Role of deep learning in wireless communications

    Yu, W., Sohrabi, F., and Jiang, T. Role of deep learning in wireless communications. IEEE BITS the Information Theory Magazine, 2 0 (2): 0 56--72, 2022. doi:10.1109/MBITS.2022.3212978

    2022

  29. [37]

    and Šišul, G

    Žeger, I. and Šišul, G. Introduction to deep learning possibilities in communication systems. In International Symposium ELMAR, pp.\ 21--24, 2021. doi:10.1109/ELMAR52657.2021.9550825

    2021

  30. [38]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.