Pith. sign in

REVIEW 1 major objections 52 references

LWM-CDE fine-tunes dataset embeddings from a wireless foundation model so that distances in the resulting space indicate how well models will transfer between datasets.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.3

2026-06-30 14:58 UTC pith:H42AG2GG

load-bearing objection LWM-CDE is a new contrastive-plus-geometry embedding method on a wireless foundation model that claims better dataset similarity for transfer tasks, but the abstract supplies no experimental controls or numbers to check the claim. the 1 major comments →

arxiv 2605.24077 v1 pith:H42AG2GG submitted 2026-05-22 eess.SP cs.LG

LWM-CDE: A Representation Space for Wireless Data Reasoning and Transferability

Sadjad Alikhani , Akshay Malhotra , Shahab Hamidi-Rad , Ahmed Alkhateeb This is my paper
classification eess.SP cs.LG
keywords wireless communicationtransfer learningdataset similaritycontrastive learningfoundation modelsrepresentation learningmodel transferabilitysignal processing
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Wireless machine learning models often fail to generalize across different locations and environments because signal data varies sharply. Existing ways to measure whether one dataset will transfer well to another are slow and unreliable. The paper builds LWM-CDE by taking embeddings from a pretrained foundation model and refining them with contrastive and geometry-shaping losses. The result is a manifold in which closer embeddings correspond to better transfer performance. Experiments show this metric matches real transfer outcomes more closely than prior methods and runs faster, enabling better choices about which data to use for training or augmentation.

Core claim

LWM-CDE fine-tunes the dataset embeddings of a pretrained wireless foundation model using a combination of contrastive and geometry-shaping losses, creating a structured manifold where distance reliably indicates transferability. On wireless benchmarks this space yields stronger correlation with empirical transfer performance than existing metrics while requiring less computation. The same space improves source dataset selection, label-aware augmentation, and budgeted pretraining decisions.

What carries the argument

The LWM-CDE manifold formed by applying contrastive learning plus geometry-shaping losses to dataset-level embeddings extracted from a pretrained wireless foundation model.

Load-bearing premise

Fine-tuning dataset embeddings with contrastive and geometry-shaping losses produces distances that reliably track actual model transfer performance across wireless environments.

What would settle it

A new wireless benchmark in which the rank correlation between LWM-CDE embedding distances and measured transfer accuracy falls below that of existing metrics such as Fréchet distance or simple feature averages.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Source dataset selection for transfer learning becomes more accurate and cheaper.
  • Label-aware data augmentation can be guided directly by proximity in the embedding space.
  • Budgeted pretraining decisions improve because the space ranks datasets by expected benefit.
  • Assessment of transferability between any pair of wireless datasets requires only embedding lookup rather than full model retraining.

Where Pith is reading between the lines

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

  • The same embedding space could be used to cluster deployment environments and flag when a new site is too dissimilar for safe transfer.
  • If the foundation model is replaced by one trained on broader signal data, the resulting manifold might extend beyond wireless to other sensor domains.
  • Periodic re-embedding of archived datasets would allow ongoing life-cycle management without repeated full evaluations.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

1 major / 0 minor

Summary. The manuscript introduces LWM-CDE (Contrastive learning of Dataset Embedding), a dataset similarity framework that fine-tunes embeddings from a pretrained wireless foundation model via a combination of contrastive and geometry-shaping losses. The resulting manifold is claimed to have distances that reliably indicate transferability. The paper asserts that extensive experiments on wireless benchmarks demonstrate stronger correlation with empirical transfer performance than existing metrics, greater computational efficiency, and utility for tasks including source dataset selection, label-aware augmentation, and budgeted pretraining.

Significance. If the central claims hold after verification of the experimental controls and loss formulations, the work could supply a more reliable and efficient tool for assessing data similarity and model transferability in wireless ML, where location-specific signal structure and data scarcity are persistent issues. This would support more principled decisions in model deployment and lifecycle management.

major comments (1)
  1. [Abstract] Abstract: the claim that 'extensive experiments on wireless benchmarks show that LWM-CDE achieves stronger correlation with empirical transfer performance than existing metrics while being more computationally efficient' is presented without any description of baselines, the correlation coefficient or statistical test used, data splits, number of trials, or controls for post-hoc selection. This absence makes it impossible to assess whether the reported correlation is load-bearing or artifactual.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their review. We address the single major comment below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that 'extensive experiments on wireless benchmarks show that LWM-CDE achieves stronger correlation with empirical transfer performance than existing metrics while being more computationally efficient' is presented without any description of baselines, the correlation coefficient or statistical test used, data splits, number of trials, or controls for post-hoc selection. This absence makes it impossible to assess whether the reported correlation is load-bearing or artifactual.

    Authors: The abstract is a concise summary of the paper's claims, consistent with standard practice in the field. All requested experimental details—specific baselines, the correlation coefficient employed, statistical tests, data splits, number of trials, and controls against post-hoc selection—are fully specified in Sections 4 and 5 of the manuscript, together with the corresponding tables, figures, and statistical reporting. Readers can therefore evaluate the strength of the reported correlations directly from the body of the paper. revision: no

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The abstract defines LWM-CDE explicitly as a fine-tuning procedure that applies contrastive and geometry-shaping losses to dataset embeddings of a pretrained model, then reports an empirical finding that the resulting distances correlate more strongly with transfer performance than prior metrics. This is a standard construction-plus-validation pattern with no equations shown that would make the reported correlation reduce by definition to the fitted parameters themselves, no self-citation invoked as a uniqueness theorem, and no renaming of known results presented as a derivation. The central claim therefore remains an independent empirical statement rather than a self-referential identity.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The method rests on the existence and utility of a pretrained wireless foundation model plus the effectiveness of the chosen contrastive and geometry-shaping losses; these are domain assumptions not independently verified in the abstract.

axioms (1)
  • domain assumption A pretrained wireless foundation model exists whose feature space is suitable for dataset-level embeddings.
    The entire LWM-CDE pipeline is built on top of this model as stated in the abstract.
invented entities (1)
  • LWM-CDE representation space no independent evidence
    purpose: Structured manifold in which distance indicates transferability
    New space produced by the fine-tuning procedure; no independent evidence supplied in the abstract.

pith-pipeline@v0.9.1-grok · 5740 in / 1341 out tokens · 53972 ms · 2026-06-30T14:58:11.865243+00:00 · methodology

0 comments
read the original abstract

Machine learning deployments in real-world wireless communication tasks face significant generalization challenges due to location and environment-specific signal structure, high diversity in data across different deployments, and limited availability of real-world data. Current approaches for assessing data similarity between training and inference (deployment) distributions, as well as evaluating model transferability, suffer from high computational costs and inconsistent performance, leaving critical model deployment and model life cycle management decisions without a principled foundation. To address this, we introduce a dataset similarity framework built upon the feature space of a pretrained wireless foundation model. Our method, LWM-CDE (Contrastive learning of Dataset Embedding), fine-tunes the dataset embeddings of the foundation model using a combination of contrastive and geometry-shaping losses, creating a structured manifold where distance reliably indicates transferability. Extensive experiments on wireless benchmarks show that LWM-CDE achieves stronger correlation with empirical transfer performance than existing metrics while being more computationally efficient. The learned representation space supports more effective and data-efficient decision-making for tasks like source dataset selection, label-aware augmentation, and budgeted pretraining, demonstrating its broader utility across different wireless communication applications.

Figures

Figures reproduced from arXiv: 2605.24077 by Ahmed Alkhateeb, Akshay Malhotra, Sadjad Alikhani, Shahab Hamidi-Rad.

Figure 1
Figure 1. Figure 1: LWM-based dataset distancing for dataset-level decisions. From a dataset library, we compute (top) a transfer-performance matrix via source→target evaluation and (bottom) a distance matrix by embedding channel samples with a pretrained LWM and comparing datasets with label-aware distances. When distances correlate with transfer, the learned geometry becomes actionable and supports source selection, distanc… view at source ↗
Figure 2
Figure 2. Figure 2: Pretraining subset selection under a dataset budget [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: True upper bound (UB) vs. transfer-matrix oracle for task [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Correlation between dataset distances and transfer perfor￾mance as a function of dataset size. Pearson (top) and Spearman (bottom) for beam prediction, LoS/NLoS, and CSI compression as samples per dataset grow. between the SW teacher and the centroid student, improving rank consistency but not by itself guaranteeing a calibrated linear relation. Consistency and correlation are variance reducers. Consistenc… view at source ↗
Figure 5
Figure 5. Figure 5: Effect of task difficulty on Euclidean and sliced Wasserstein [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: t-SNE visualization of dataset embeddings, colored by dataset [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Distance computation time using different number of samples [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗

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

52 extracted references · 52 canonical work pages

  1. [1]

    Deep learning for channel estimation: Interpretation, performance, and comparison,

    Q. Hu, F. Gao, H. Zhang, S. Jin, and G. Y . Li, “Deep learning for channel estimation: Interpretation, performance, and comparison,”IEEE Transactions on Wireless Communications, vol. 20, no. 4, pp. 2398–2412, 2021

    work page 2021

  2. [2]

    Deep learning for wireless physical layer: Opportunities and challenges,

    T. Wang, C.-K. Wen, H. Wang, F. Gao, T. Jiang, and S. Jin, “Deep learning for wireless physical layer: Opportunities and challenges,”China Communications, vol. 14, no. 11, pp. 92–111, 2017

    work page 2017

  3. [3]

    TR 38.843: Study on artificial intelligence (AI)/machine learning (ML) for NR air interface,

    “TR 38.843: Study on artificial intelligence (AI)/machine learning (ML) for NR air interface,” 3rd Generation Partnership Project (3GPP), Technical Report 3GPP TR 38.843, 2024, rel-18

    work page 2024

  4. [4]

    Transfer learning for wireless networks: A comprehensive survey,

    C. T. Nguyen, N. Van Huynh, N. H. Chu, Y . M. Saputra, D. T. Hoang, D. N. Nguyen, Q.-V . Pham, D. Niyato, E. Dutkiewicz, and W.-J. Hwang, “Transfer learning for wireless networks: A comprehensive survey,” Proceedings of the IEEE, vol. 110, no. 8, pp. 1073–1115, 2022

    work page 2022

  5. [5]

    Representation similarity analysis for efficient task taxonomy & transfer learning,

    K. Dwivedi and G. Roig, “Representation similarity analysis for efficient task taxonomy & transfer learning,” 2019

    work page 2019

  6. [6]

    Overcoming data scarcity in biomedical imaging with a foundational multi-task model,

    R. Sch ¨afer, T. Nicke, H. H ¨ofener, A. Lange, D. Merhof, F. Feuerhake, V . Schulz, J. Lotz, and F. Kiessling, “Overcoming data scarcity in biomedical imaging with a foundational multi-task model,” 2023

    work page 2023

  7. [7]

    GLISTER: Generalization based data subset selection for efficient and robust learning,

    K. Killamsetty, D. Sivasubramanian, G. Ramakrishnan, and R. Iyer, “GLISTER: Generalization based data subset selection for efficient and robust learning,” 2021. 13

    work page 2021

  8. [8]

    Don’t stop pretraining: Adapt language models to domains and tasks,

    S. Gururangan, A. Marasovi ´c, S. Swayamdipta, K. Lo, I. Beltagy, D. Downey, and N. A. Smith, “Don’t stop pretraining: Adapt language models to domains and tasks,” 2020

    work page 2020

  9. [9]

    MATES: Model-aware data selection for efficient pretraining with data influence models,

    Z. Yu, S. Das, and C. Xiong, “MATES: Model-aware data selection for efficient pretraining with data influence models,” 2024

    work page 2024

  10. [10]

    UMAP: Uniform manifold approximation and projection for dimension reduction,

    L. McInnes, J. Healy, and J. Melville, “UMAP: Uniform manifold approximation and projection for dimension reduction,” 2020

    work page 2020

  11. [11]

    Wireless dataset similarity: Measuring distances in supervised and unsupervised machine learning,

    J. Morais, S. Alikhani, A. Malhotra, S. Hamidi-Rad, and A. Alkhateeb, “Wireless dataset similarity: Measuring distances in supervised and unsupervised machine learning,” 2026

    work page 2026

  12. [12]

    Understanding the transferability of representations via task-relatedness,

    A. Mehra, Y . Zhang, and J. Hamm, “Understanding the transferability of representations via task-relatedness,” inAdvances in Neural Information Processing Systems (NeurIPS), 2024, arXiv:2307.00823

    work page arXiv 2024

  13. [13]

    The geometry of multilingual language model representations,

    T. A. Chang, Z. Tu, and B. K. Bergen, “The geometry of multilingual language model representations,”arXiv preprint arXiv:2205.10964, 2023

    work page arXiv 2023

  14. [14]

    Correlations between multilingual language model geometry and crosslin- gual transfer performance,

    C. Shah, Y . Chandak, A. M. Mane, B. Bergen, and T. A. Chang, “Correlations between multilingual language model geometry and crosslin- gual transfer performance,” inProceedings of the Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024). ELRA and ICCL, 2024, pp. 4059– 4066

    work page 2024

  15. [15]

    Fantastic generalization measures and where to find them,

    Y . Jiang, B. Neyshabur, H. Mobahi, D. Krishnan, and S. Bengio, “Fantastic generalization measures and where to find them,” 2019

    work page 2019

  16. [16]

    Large wirele ss model (LWM): A foundation model for wireless channels,

    S. Alikhani, G. Charan, and A. Alkhateeb, “Large Wireless Model: A foundation model for wireless channels,”arXiv preprint arXiv:2411.08872, 2024

    work page arXiv 2024

  17. [17]

    LWM: A pre-trained wireless foundation model for universal feature extraction,

    ——, “LWM: A pre-trained wireless foundation model for universal feature extraction,” in2025 IEEE International Conference on Machine Learning for Communication and Networking (ICMLCN), 2025, pp. 1–6

    work page 2025

  18. [18]

    Beyond separability: Analyzing the linear transferability of contrastive representations to related subpopulations,

    J. Z. HaoChen, C. Wei, A. Kumar, and T. Ma, “Beyond separability: Analyzing the linear transferability of contrastive representations to related subpopulations,” 2022

    work page 2022

  19. [19]

    A theory of learning from different domains,

    S. Ben-David, J. Blitzer, K. Crammer, A. Kulesza, F. Pereira, and J. W. Vaughan, “A theory of learning from different domains,”Machine Learning, vol. 79, no. 1–2, pp. 151–175, 2010

    work page 2010

  20. [20]

    Unsupervised domain adaptation by backpropagation,

    Y . Ganin and V . Lempitsky, “Unsupervised domain adaptation by backpropagation,” inProceedings of the 32nd International Conference on Machine Learning (ICML), 2015, pp. 1180–1189

    work page 2015

  21. [21]

    A kernel method for the two-sample problem,

    A. Gretton, K. Borgwardt, M. J. Rasch, B. Scholkopf, and A. J. Smola, “A kernel method for the two-sample problem,” 2008

    work page 2008

  22. [22]

    Energy statistics: A class of statistics based on distances,

    G. J. Sz ´ekely and M. L. Rizzo, “Energy statistics: A class of statistics based on distances,”Journal of Statistical Planning and Inference, vol. 143, no. 8, pp. 1249–1272, 2013

    work page 2013

  23. [23]

    Similarity of neural network representations revisited,

    S. Kornblith, M. Norouzi, H. Lee, and G. Hinton, “Similarity of neural network representations revisited,” 2019

    work page 2019

  24. [24]

    Computational optimal transport,

    G. Peyr ´e and M. Cuturi, “Computational optimal transport,” 2020

    work page 2020

  25. [25]

    Generalized sliced wasserstein distances,

    S. Kolouri, K. Nadjahi, U. Simsekli, R. Badeau, and G. K. Rohde, “Generalized sliced wasserstein distances,” 2019

    work page 2019

  26. [26]

    Sliced and Radon Wasserstein barycenters of measures,

    N. Bonneel, J. Rabin, G. Peyr ´e, and H. Pfister, “Sliced and Radon Wasserstein barycenters of measures,”Journal of Mathematical Imaging and Vision, vol. 51, no. 1, pp. 22–45, 2015

    work page 2015

  27. [27]

    Wasserstein task embedding for measuring task similarities,

    Y . Liu, Z. Zhang, Y . Li, and P.-Y . Chen, “Wasserstein task embedding for measuring task similarities,”Neural Networks, 2025, arXiv:2410.11456

    work page arXiv 2025

  28. [28]

    A dataset similarity evaluation framework for wireless communications and sensing,

    J. Morais, S. Alikhani, A. Malhotra, S. Hamidi-Rad, and A. Alkhateeb, “A dataset similarity evaluation framework for wireless communications and sensing,” in2024 58th Asilomar Conference on Signals, Systems, and Computers, 2024, pp. 1144–1149

    work page 2024

  29. [29]

    Task2Vec: Task embedding for meta-learning,

    A. Achille, M. Lam, R. Tewari, A. Ravichandran, S. Maji, C. C. Fowlkes, S. Soatto, and P. Perona, “Task2Vec: Task embedding for meta-learning,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2019, pp. 6430–6439

    work page 2019

  30. [30]

    Taskonomy: Disentangling task transfer learning,

    A. R. Zamir, A. Sax, W. Shen, L. J. Guibas, J. Malik, and S. Savarese, “Taskonomy: Disentangling task transfer learning,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018, pp. 3712–3722

    work page 2018

  31. [31]

    SimCSE: Simple contrastive learning of sentence embeddings,

    T. Gao, X. Yao, and D. Chen, “SimCSE: Simple contrastive learning of sentence embeddings,” 2022

    work page 2022

  32. [32]

    RankCSE: Unsupervised sentence representations learning via learning to rank,

    J. Liu, J. Liu, Q. Wang, J. Wang, W. Wu, Y . Xian, D. Zhao, K. Chen, and R. Yan, “RankCSE: Unsupervised sentence representations learning via learning to rank,” 2023

    work page 2023

  33. [33]

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

    J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “BERT: Pre-training of deep bidirectional transformers for language understanding,” 2019

    work page 2019

  34. [34]

    A simple framework for contrastive learning of visual representations,

    T. Chen, S. Kornblith, M. Norouzi, and G. Hinton, “A simple framework for contrastive learning of visual representations,” inProceedings of the 37th International Conference on Machine Learning (ICML), 2020, pp. 1597–1607

    work page 2020

  35. [35]

    Momentum contrast for unsupervised visual representation learning,

    K. He, H. Fan, Y . Wu, S. Xie, and R. Girshick, “Momentum contrast for unsupervised visual representation learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020, pp. 9729–9738

    work page 2020

  36. [36]

    Unsupervised learning of visual features by contrasting cluster assign- ments,

    M. Caron, I. Misra, J. Mairal, P. Goyal, P. Bojanowski, and A. Joulin, “Unsupervised learning of visual features by contrasting cluster assign- ments,” inAdvances in Neural Information Processing Systems (NeurIPS), 2020

    work page 2020

  37. [37]

    Contrastive represen- tation learning: A framework and review,

    P. H. Le-Khac, G. Healy, and A. F. Smeaton, “Contrastive represen- tation learning: A framework and review,”IEEE Access, vol. 8, p. 193907–193934, 2020

    work page 2020

  38. [38]

    Language models improve when pretraining data matches target tasks,

    S. Mizrahi, R. Aharoni, and O. Abend, “Language models improve when pretraining data matches target tasks,”arXiv preprint arXiv:2307.08723, 2023

    work page arXiv 2023

  39. [39]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” inAdvances in Neural Information Processing Systems (NeurIPS), 2017

    work page 2017

  40. [40]

    Distilling the knowledge in a neural network,

    G. Hinton, O. Vinyals, and J. Dean, “Distilling the knowledge in a neural network,” 2015

    work page 2015

  41. [41]

    Adam: A method for stochastic optimization,

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” inInternational Conference on Learning Representations (ICLR), 2015

    work page 2015

  42. [42]

    Deep learning for mmWave beam and blockage prediction using Sub-6 GHz channels,

    M. Alrabeiah and A. Alkhateeb, “Deep learning for mmWave beam and blockage prediction using Sub-6 GHz channels,”IEEE Transactions on Communications, vol. 68, no. 9, pp. 5504–5518, 2020

    work page 2020

  43. [43]

    DeepMIMO: A generic deep learning dataset for millimeter wave and massive MIMO applications,

    A. Alkhateeb, “DeepMIMO: A generic deep learning dataset for millimeter wave and massive MIMO applications,” 2019

    work page 2019

  44. [44]

    Fast and eager k-medoids clustering: O(k) runtime improvement of the PAM, CLARA, and CLARANS algorithms,

    E. Schubert and P. J. Rousseeuw, “Fast and eager k-medoids clustering: O(k) runtime improvement of the PAM, CLARA, and CLARANS algorithms,” inInformation Systems, vol. 101, 2021, p. 101804

    work page 2021

  45. [45]

    Channel charting: Locating users within the radio environment using channel state information,

    C. Studer, S. Medjkouh, E. G ¨on¨ultas ¸, T. Goldstein, and O. Tirkkonen, “Channel charting: Locating users within the radio environment using channel state information,”IEEE Access, vol. 6, pp. 47 682–47 698, 2018

    work page 2018

  46. [46]

    Triplet-based wireless channel charting: Architecture and experiments,

    P. Ferrand, A. Decurninge, and M. Guillaud, “Triplet-based wireless channel charting: Architecture and experiments,”IEEE Journal on Selected Areas in Communications, vol. 39, no. 8, pp. 2361–2373, 2021

    work page 2021

  47. [47]

    LWM- Temporal: Sparse spatio-temporal attention for wireless channel repre- sentation learning,

    S. Alikhani, A. Malhotra, S. Hamidi-Rad, and A. Alkhateeb, “LWM- Temporal: Sparse spatio-temporal attention for wireless channel repre- sentation learning,” 2026

    work page 2026

  48. [48]

    WiFo: Wireless foundation model for channel prediction,

    B. Liu, S. Gao, X. Liu, X. Cheng, and L. Yang, “WiFo: Wireless foundation model for channel prediction,”Science China Information Sciences, 2025

    work page 2025

  49. [49]

    WirelessGPT: A generative pre-trained multi-task learning framework for wireless communication,

    T. Yang, P. Zhang, M. Zheng, Y . Shi, L. Jing, J. Huang, and N. Li, “WirelessGPT: A generative pre-trained multi-task learning framework for wireless communication,”IEEE Network, 2025

    work page 2025

  50. [50]

    LWM-Spectro: A foundation model for wireless baseband signal spectrograms,

    N. Kim, S. Alikhani, and A. Alkhateeb, “LWM-Spectro: A foundation model for wireless baseband signal spectrograms,” 2026

    work page 2026

  51. [51]

    The algorithmic foundations of differential privacy,

    C. Dwork and A. Roth, “The algorithmic foundations of differential privacy,”Foundations and Trends in Theoretical Computer Science, vol. 9, no. 3–4, pp. 211–407, 2014

    work page 2014

  52. [52]

    Deep learning with differential privacy,

    M. Abadi, A. Chu, I. Goodfellow, H. B. McMahan, I. Mironov, K. Talwar, and L. Zhang, “Deep learning with differential privacy,” inProceedings of the 2016 ACM SIGSAC Conference on Computer and Communications Security (CCS), 2016, pp. 308–318

    work page 2016