Pith. sign in

REVIEW 3 major objections 5 minor 32 references

Fast Cross-Scenario Adaptation of CSI Models via Channel Conditional Parameter Generation

T0 review · 3 major / 5 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a CSI reconstruction model can be adapted to a new radio environment with a single forward pass (~3 seconds) by generating its lightweight LoRA adapter weights from compressed channel features, matching costly online

desk verdict A serious, well-engineered conditional LoRA generation pipeline for CSI models that plausibly delivers ~3 s adaptation with near-fine-tuning accuracy, but the 'single forward pass / training-free' framing hides a transductive batch-collection step and the tables have a numeric inconsistency. read the letter →

arxiv 2607.22637 v1 pith:4S4EZS6B submitted 2026-06-22 cs.AI cs.ITmath.IT

Xudong Zou , Siyu Wu , Zunlei Feng , Jie Song , Yuanyu Wan , Mingli Song , Jiacong Hu This is my paper
classification cs.AIcs.ITmath.IT
keywords CSIfeedbackchannelestimationMassiveMIMOparametergenerationLoRAdiffusionmodeldomaingeneralizationfastadaptation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that the usual online fine-tuning used to adapt massive-MIMO channel reconstruction models to new environments can be replaced by a conditional parameter generator. Once trained offline on a set of source scenarios, the generator takes the unlabeled channel features of a new scenario, compresses them, and produces the LoRA weights that should be plugged into the frozen decoder. The authors report that this single forward pass takes about 3 seconds and reaches reconstruction quality within roughly 0.3 dB of 45–61 minutes of online training on DeepMIMO and WAIR-D. If true, it would make plug-and-play deployment of CSI models practical at the pace of environmental change, without gradient updates at the terminal. The core insight is that the main obstacle to generating weights is not the high dimensionality of channel inputs, but the ambiguity of the weight space itself.

What carries the argument

The energy-driven canonicalization operator O: for LoRA matrices (A,B) with the same functional mapping BA, it sorts the rank components by ei = ||ai||2·||bi||2 and flips each row of A so its maximum-absolute-value entry is positive. This removes the permutation and sign symmetries that otherwise make the target weight space one-to-many, so the conditional diffusion backbone (a Diffusion Transformer with decoupled time/condition injection and a size-aware weighting mask) learns a deterministic parameter manifold instead of mode-averaging equivalent weights.

What would settle it

Measure the end-to-end time from a user equipment entering a new scenario with no buffered observations until the adapted CSI model is usable, including the time to acquire the channel-feature batch for the SVD condition; if this total is orders of magnitude above ~3 s, the cold-start deployment claim fails. Alternatively, run the generator with a single observation as condition and compare NMSE to the frozen base model.

Watch

Extended reading notes

Core claim

The central claim is that a frozen CSI decoder can be adapted to an unseen radio scenario by generating its LoRA adapter matrices from channel features, without any training or fine-tuning on the new scenario. The authors localize the adaptation bottleneck to the FFN layers of the Transformer decoder, then learn a diffusion model that maps a compact condition — obtained from SVD truncation plus a Perceiver Resampler — to canonicalized LoRA weights. The key novelty is the energy-based canonicalization operator: sorting rank components by energy and anchoring signs makes the target parameter manifold unique, so the diffusion model does not average over functionally equivalent but coordinate-wi

Load-bearing premise

At deployment time, the target scenario's channel features must already be available in a batch large enough to run the SVD-based condition encoder; without any target observations, CCPG has no condition and cannot generate weights, so the '3-second adaptation' excludes the time to collect its input.

Editorial extensions

If this is right

  • A single forward pass replaces 200 epochs of online gradient steps, so model adaptation can track scenario changes in near-real time.
  • Because the generated weights are plug-in LoRA matrices, one frozen base model can serve many users in different environments simultaneously, each with its own adapter.
  • The reported NMSE gap to online training is small (e.g., -26.96 vs -27.22 dB on DeepMIMO CF), so for many link budgets the training-free solution may be sufficient.
  • The 3-second adaptation time moves computational cost to offline generator training, which is compatible with terminal-side deployment constraints.

Reading between the lines

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

  • If the method's premise holds, the canonicalization operator could generalize to any low-rank adapter family (not just LoRA) and to other physical-layer models where permutation and sign degeneracies obscure the weight manifold.
  • The reported ~3 s adaptation excludes the time to collect the batch of target-scenario channel features needed to form the SVD condition; with zero buffered observations, CCPG degenerates to the frozen base model. A fair deployment-time comparison should include that collection latency.
  • A testable extension: replace the fixed SVD truncation with a learned, differentiable compression and see whether the performance gap to online training narrows further while reducing sensitivity to the batch size.
  • The approach suggests a federated deployment model where each base station generates its own adapters from local channel statistics, never sharing raw CSI, since the condition encoder only needs unlabeled features.
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

3 major / 5 minor

Summary. This paper introduces CCPG, a conditional parameter-generation pipeline for adapting CSI feedback and channel estimation models to new radio scenarios. Given hidden channel features from a fixed encoder, a non-learnable truncated SVD plus a Perceiver Resampler yields a compact scene condition; an energy-based canonicalization aligns LoRA target weights; and a DiT-based diffusion model generates the LoRA weights for the decoder FFN layers. Experiments on DeepMIMO and WAIR-D report that with DDIM sampling CCPG adapts in approximately 3 seconds, with NMSE close to 200-epoch online fine-tuning, and leave-one-out ablations show each component contributes. The paper positions this as a training-free, plug-and-play alternative to costly online adaptation.

Significance. The reported reduction from roughly one hour of online fine-tuning to a few seconds of adaptation, with cross-domain NMSE close to the online reference, would be practically significant for deploying CSI models in dynamic environments. The paper's evaluation has genuine strengths: a held-out protocol over 10 test scenarios, comparisons with four parameter-generation baselines, leave-one-out ablations, and a public code link. The main caveats are the transductive requirement for target-scenario data at deployment and some inconsistencies in reported tables. If the deployment assumptions are made explicit and the numbers are reconciled, the method is a useful contribution to fast wireless model adaptation.

major comments (3)
  1. [III-B and IV-C, Eq. (7)] The problem is framed as adapting to a completely unseen scenario S_N using only z=F_en(s;Theta_en), but the diffusion condition c is computed from a truncated SVD of a batch of target-scenario hidden features (C_svd in Eq. (7)). Thus CCPG is transductive: before any weights are generated, a batch of unlabeled target samples must be available and processed through the frozen encoder, SVD, and Perceiver. The abstract's 'without training or fine-tuning on new-scenario data' is not the same as 'without using new-scenario data.' The paper should state the minimum batch size needed for a stable SVD, whether feature extraction/SVD/Perceiver time is included in the reported ~3 s, and how this aligns with the 'completely unseen' definition in Eq. (5). Section VI does not mention this deployment caveat.
  2. [Abstract and Section V-B] The abstract and contribution list claim adaptation 'through a single forward pass,' but Section V-B states that deployment uses DDIM sampling with 50 denoising steps. Each step requires a DiT forward pass, so the generation process is 50 iterative forward passes, plus the SVD/Perceiver condition computation. The efficiency claim should be restated as 'one diffusion sampling run' or the DDIM step count should be explicitly acknowledged; the reported 3 s budget should break down the condition-encoding and denoising components.
  3. [Table I vs Table III] The CCPG row reports different numbers for the same held-out test set. On WAIR-D CF, Table I gives an average NMSE of -20.59 dB, while Table III gives -21.13 dB for 'CCPG (Ours)'. The DeepMIMO CF individual scenario values also differ slightly (-27.44/-27.85 vs -27.46/-27.88) even though the average is the same. Both tables claim the Average is computed over all 10 held-out test scenarios; this inconsistency undermines the ablation comparison and reporting reliability.
minor comments (5)
  1. [Section V-B] The LoRA rank r is not specified anywhere in the main text, and the number of scenarios in DeepMIMO and WAIR-D used for the 10%/90% split is not stated. These are needed for reproducibility.
  2. [Section III-A / III-B] For the channel estimation task, the encoder is F_CE, but the formulation and condition encoder refer to z=F_en(s;Theta_en). The notation should be unified or the CE analog should be defined explicitly.
  3. [Tables I and III] The table headers are visually mangled in the text; scenario IDs and Average columns should be clearly separated for each dataset-task pair.
  4. [Section V-D] No variance or multiple-seed results are reported for the main table. Given that the generator is stochastic and baselines show high variability, mean and standard deviation over at least three seeds would strengthen the comparison.
  5. [Section I] The introduction says adaptation time is reduced 'exponentially,' but Table II shows a roughly constant reduction from ~1 hour to ~3 seconds. This is an overstatement; 'by more than an order of magnitude' would be accurate.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the generator is trained on source-scenario targets and evaluated on disjoint held-out scenarios; the transductive condition requirement is a deployment limitation, not a circular step.

full rationale

No load-bearing circular step is present. The CCPG generator is trained on paired samples ⟨condition c_i, optimal LoRA weights W*_i⟩, where W*_i are obtained by the authors' own 200-epoch online fine-tuning (Eq. 3) on source scenarios. This creates a distillation-like relationship with the Online Training baseline, but the evaluation uses 10 held-out test scenarios whose labels do not enter generator fitting, and the reported NMSE is computed on those disjoint scenarios. The deployment-time condition c is computed from unlabeled target features via SVD and a Perceiver (Eq. 7), so the method is transductive and requires a batch of target-scenario features before weights can be generated; this is a limitation of the 'training-free / plug-and-play' framing, but it is not a definitional circularity because the generated weights are not fitted to target labels and Eq. (11) optimizes a parameter-space denoising objective on source domains only. Self-citations [16]–[19] appear only as background motivation about treating parameters as structured objects and are not load-bearing for the central derivation. The derivation chain (Eqs. 3–5, 7, 11) is self-contained: the generator maps a condition to LoRA residuals, and the held-out scenario evaluation provides independent evidence of cross-scenario generalization. Therefore the circularity score is 0.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

No new physical entities (particles, forces, channels, dimensions) are introduced; the energy-ordering operator, condition tokens, and size-aware masks are algorithmic constructs, not physical postulates. The free parameters are mostly hand-set architecture values plus the tuned ρ; the two unreported values (LoRA rank r, SNR threshold γ) are the largest reproducibility gaps.

free parameters (6)
  • SVD truncation rank L = 224
    Hand-selected in §V-B for the condition compressor; no sensitivity study is given, and L along with d=512 fixes the information budget available to the generator.
  • Condition latent tokens M / latent dim d = M=128, d=512
    Architecture choices for the Perceiver Resampler output (§V-B); they determine how much scene information the diffusion model sees.
  • Size-aware weighting ratio ρ = 0.3
    Selected by the scan in Fig. 6; ρ=0.3 vs 0.5 changes NMSE by 1.8–3.3 dB and the default gradient ratio is ~0.03, so this choice strongly shapes the reported gains (§V-E).
  • Truncated SNR threshold γ = not reported
    Appears in λ(t)=min(SNR(t),γ) in Eq. 11; the value is never stated, so the loss weighting cannot be exactly reproduced (§IV-E).
  • LoRA rank r = not reported
    The paper fixes the rank (Limitations) but never gives r, though the entire generated object (A∈R^{r×din}, B∈R^{dout×r}) has size r; required for reproduction (§IV-B).
  • Per-scenario oracle training epochs = 200
    Both the meta-dataset targets (Eq. 3) and the 'Online Training' reference use 200 epochs of LoRA fine-tuning; this binds the ceiling the generator is compared against (§V-C).
assumptions (6)
  • domain assumption FFN layers are the dominant cross-scenario bottleneck in Transformer CSI decoders.
    Shown only for TransNet (CSI feedback) on WAIR-D via freezing (Fig. 3, §IV-B), then assumed for the channel-estimation decoder and DeepMIMO experiments without analogous freezing evidence.
  • domain assumption Encoder-hidden features z are approximately low-rank and SVD truncation preserves scenario-semantic information.
    §IV-C motivates 'physical truncation' with the cluster/scattering argument; the truncation discards components and no analysis quantifies what is lost.
  • domain assumption A smooth learnable map exists from the condition latent c to canonicalized LoRA weights, generalizable to unseen conditions.
    Eq. 5 is declared intractable and replaced by a diffusion objective (§III-B); the extrapolation success is an empirical premise validated only by the reported tests.
  • standard math Diffusion/DDIM sampling and classifier-free guidance apply without modification to weight-token sequences.
    Eqs. 10–11 use the standard forward process, Min-SNR weighting [24], and CFG dropout; treated as background.
  • standard math Energy sorting plus pivot-sign flipping yields a unique canonical representative that preserves the functional manifold.
    Algorithm 1 defines a deterministic representative, but uniqueness depends on tie-breaking rules (index ordering, first largest-abs element); Fig. 5 gives only a qualitative 10-seed check with no quantitative statistics.
  • domain assumption UEs within one scenario share statistical channel structure, so a per-scenario condition token is meaningful.
    §IV-C claims UEs share scattering clusters; this justifies replacing instance-level conditions with one scenario-level code per environment.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Fast Cross-Scenario Adaptation of CSI Models via Channel Conditional Parameter Generation." pith.science (2026). https://pith.science/paper/4S4EZS6B

@misc{pith2026260722637,
  author       = {Pith},
  title        = {Pith review of: Fast Cross-Scenario Adaptation of CSI Models via Channel Conditional Parameter Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4S4EZS6B}},
  note         = {Machine review of arXiv:2607.22637}
}
read the original abstract

Deep learning has shown strong potential for massive multiple-input multiple-output (Massive MIMO) physical-layer tasks, including channel state information (CSI) feedback and channel estimation. However, environmental heterogeneity can severely degrade CSI models in unseen scenarios, while conventional adaptation requires target-domain data and substantial computation. This paper proposes Channel Conditional Parameter Generation (CCPG), an end-to-end pipeline for rapid deployment of CSI models in dynamic wireless environments. CCPG identifies scene-sensitive adaptation bottlenecks through component-freezing experiments and generates only lightweight LoRA weights instead of full model parameters. It compresses high-dimensional channel features into compact latent conditions using cascaded SVD and a Perceiver Resampler. An energy-based canonicalization mechanism mitigates permutation and sign ambiguities in LoRA weights, while a diffusion-based generator incorporates structural information and an asymmetric size-aware loss for topology-aware parameter generation. Experiments on DeepMIMO and WAIR-D for CSI feedback and channel estimation show that CCPG adapts to new scenarios in about 3 seconds with a single forward pass, without target-scenario training or fine-tuning, and achieves cross-domain recovery performance comparable to costly online adaptation. These results demonstrate that CCPG enables efficient deployment of CSI models in large-scale dynamic wireless scenarios for intelligent 6G communications.

Figures

Figures reproduced from arXiv: 2607.22637 by the authors.

Figure 1
Figure 1. System model of CSI feedback and the motivation for cross-scenario [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Overall workflow of the proposed CCPG framework. During offline training, CCPG constructs paired condition-parameter samples from multiple [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. NMSE performance comparison under different component-freezing [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (2 more)
Figure 5
Figure 5. Figure 5: Visualization of the parameter manifold before and after the energy [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Performance curve of average NMSE with respect to the asymmetric [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

32 extracted references · 2 linked inside Pith

  1. [1]

    A vision of 6g wireless systems: Applications, trends, technologies, and open research problems,

    W. Saad, M. Bennis, and M. Chen, “A vision of 6g wireless systems: Applications, trends, technologies, and open research problems,”IEEE Network, vol. 34, no. 3, pp. 134–142, 2020

    2020

  2. [2]

    Power of deep learning for channel estimation and signal detection in OFDM systems,

    H. Ye, G. Y . Li, and B.-H. F. Juang, “Power of deep learning for channel estimation and signal detection in OFDM systems,”IEEE Wireless Communications Letters, vol. 7, no. 1, pp. 114–117, 2018

    2018

  3. [3]

    Robust and Efficient CSI Compression Using CLLWCsiNet in Noisy MIMO Environments ,

    F. Ansari and M. Mohassel Feghhi, “ Robust and Efficient CSI Compression Using CLLWCsiNet in Noisy MIMO Environments ,” IEEE Transactions on Mobile Computing, vol. 25, no. 04, pp. 4936– 4947, Apr. 2026. [Online]. Available: https://doi.ieeecomputersociety. org/10.1109/TMC.2025.3627109

    arXiv 2026

  4. [4]

    SwiftChannel: Algorithm-Hardware Co-Design for Deep Learning-Based 5G Channel Estimation ,

    S. Lyu, Y . She, D. Duan, T. Ni, Y . H. Chan, C. Luo, R. C. C. Cheung, and W. Xu, “ SwiftChannel: Algorithm-Hardware Co-Design for Deep Learning-Based 5G Channel Estimation ,”IEEE Transactions on Mobile Computing, vol. 25, no. 06, pp. 8268–8284, Jun. 2026. [Online]. Available: https://doi.ieeecomputersociety.org/10.1109/TMC. 2025.3647644

    arXiv 2026

  5. [5]

    Enabling feedback-free mimo transmission for fd-ran: A data-driven approach,

    J. Liu, J. Chen, Z. Liu, and H. Zhou, “Enabling feedback-free mimo transmission for fd-ran: A data-driven approach,”IEEE Transactions on Mobile Computing, vol. 24, no. 3, pp. 2437–2454, 2025

    2025

  6. [6]

    Leveraging generative artificial intelligence for uplink feedback-free transmission in 6g fd-ran,

    H. Zhou, Y . Xu, T. Zhang, X. Zhang, J. Chen, and X. Shen, “Leveraging generative artificial intelligence for uplink feedback-free transmission in 6g fd-ran,”IEEE Transactions on Mobile Computing, vol. 25, no. 5, pp. 6801–6817, 2026

    2026

  7. [7]

    Deep learning for massive mimo csi feedback,

    C.-K. Wen, W.-T. Shih, and S. Jin, “Deep learning for massive mimo csi feedback,”IEEE Wireless Communications Letters, vol. 7, no. 5, pp. 748–751, 2018

    2018

  8. [8]

    Model- driven deep learning for physical layer communications,

    H. He, S. Jin, C.-K. Wen, F. Gao, G. Y . Li, and Z. Xu, “Model- driven deep learning for physical layer communications,”IEEE Wireless Communications, vol. 26, no. 5, pp. 77–83, 2019

    2019

Show all 32 references
  1. [9]

    Dda-net: A discrepancy- based domain adaptation network for csi feedback transferability,

    Y . Feng, C. Ye, R. Li, H. Pan, and D. Korpi, “Dda-net: A discrepancy- based domain adaptation network for csi feedback transferability,” in ICC 2023 - IEEE International Conference on Communications, 2023, pp. 4157–4162

    2023

  2. [10]

    Model-agnostic meta-learning for fast adaptation of deep networks,

    C. Finn, P. Abbeel, and S. Levine, “Model-agnostic meta-learning for fast adaptation of deep networks,” inProceedings of the 34th International Conference on Machine Learning. PMLR, 2017, pp. 1126–1135

    2017

  3. [11]

    Enhanced meta-transfer learning assisted csi feedback in massive mimo systems,

    H. Zhao, W. Liu, W. Xia, Y . Shen, and H. Zhu, “Enhanced meta-transfer learning assisted csi feedback in massive mimo systems,”IEEE Wireless Communications Letters, vol. 14, no. 2, pp. 499–503, 2025

    2025

  4. [12]

    Transnet: Full attention network for csi feedback in fdd massive mimo system,

    Y . Cui, A. Guo, and C. Song, “Transnet: Full attention network for csi feedback in fdd massive mimo system,”IEEE Wireless Communications Letters, vol. 11, no. 5, pp. 903–907, 2022

    2022

  5. [13]

    Hrnet: Deep massive mimo csi feedback with hypernetworks,

    X. Cen, C.-T. Lam, B. Ng, and S.-K. Im, “Hrnet: Deep massive mimo csi feedback with hypernetworks,” in2024 9th International Conference on Signal and Image Processing (ICSIP), 2024, pp. 500–504

    2024

  6. [14]

    Hypernetwork-aided channel estimation for integrated data and energy transfer,

    Y . Lei, Y . Liu, J. Hu, and K. Yang, “Hypernetwork-aided channel estimation for integrated data and energy transfer,”IEEE Transactions on Vehicular Technology, vol. 74, no. 2, pp. 3401–3406, 2025

    2025

  7. [15]

    A hypernetwork based framework for non-stationary channel prediction,

    G. Liu, Z. Hu, L. Wang, H. Zhang, J. Xue, and M. Matthaiou, “A hypernetwork based framework for non-stationary channel prediction,” IEEE Transactions on Vehicular Technology, vol. 73, no. 6, pp. 8338– 8351, 2024

    2024

  8. [16]

    Model lego: Creating models like disassembling and assembling building blocks,

    J. Hu, J. Gao, J. Ye, Y . Gao, X. Wang, Z. Feng, and M. Song, “Model lego: Creating models like disassembling and assembling building blocks,” inAdvances in Neural Information Processing Systems, A. Globerson, L. Mackey, D. Belgrave, A. Fan, U. Paquet, J. Tomczak, and C. Zhang...

    2024

  9. [17]

    Transformer doctor: Diagnosing and treating vision transformers,

    J. Hu, H. Chen, K. Chen, Y . Gao, J. Ye, X. Wang, M. Song, and Z. Feng, “Transformer doctor: Diagnosing and treating vision transformers,” in Advances in Neural Information Processing Systems, A. Globerson, L. Mackey, D. Belgrave, A. Fan, U. Paquet, J. Tomczak, and C. Zhang, E...

    2024

  10. [18]

    Vision mamba mender,

    J. Hu, A. Cao, Z. Feng, S. Zhang, Y . Wang, L. Jia, and M. Song, “Vision mamba mender,” inAdvances in Neural Information Processing Systems, A. Globerson, L. Mackey, D. Belgrave, A. Fan, U. Paquet, J. Tomczak, and C. Zhang, Eds., vol. 37. Curran Associates, Inc., 2024, pp. 51 ...

    2024

  11. [19]

    Parameter manifold purification,

    J. Hu, J. Wu, S. Zhang, S. Liu, H. Zhang, M. Song, and Z. Feng, “Parameter manifold purification,” inProceedings of the International Conference on Machine Learning, 2026, accepted to ICML 2026. [Online]. Available: https://parameter-purification.github.io/

    2026

  12. [20]

    The approximation of one matrix by another of lower rank,

    C. Eckart and G. Young, “The approximation of one matrix by another of lower rank,”Psychometrika, vol. 1, no. 3, pp. 211–218, 1936

    1936

  13. [21]

    Perceiver: General perception with iterative attention,

    A. Jaegle, F. Gimeno, A. Brock, O. Vinyals, A. Zisserman, and J. Carreira, “Perceiver: General perception with iterative attention,” in Proceedings of the 38th International Conference on Machine Learning, ser. Proceedings of Machine Learning Research, vol. 139. PMLR, 2021, pp...

    2021

  14. [22]

    The role of permutation invariance in linear mode connectivity of neural networks,

    R. Entezari, H. Sedghi, O. Saukh, and B. Neyshabur, “The role of permutation invariance in linear mode connectivity of neural networks,” inInternational Conference on Learning Representations, 2022

    2022

  15. [23]

    Git re-basin: Merging models modulo permutation symmetries,

    S. K. Ainsworth, J. Hayase, and S. Srinivasa, “Git re-basin: Merging models modulo permutation symmetries,” inInternational Conference on Learning Representations, 2023

    2023

  16. [24]

    Efficient diffusion training via min-snr weighting strategy,

    T. Hang, S. Gu, C. Li, J. Bao, D. Chen, H. Hu, X. Geng, and B. Guo, “Efficient diffusion training via min-snr weighting strategy,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 7407–7451

    2023

  17. [25]

    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,”arXiv preprint arXiv:1902.06435, 2019. [Online]. Available: https://arxiv.org/abs/1902. 06435

    1902 arXiv

  18. [26]

    W AIR-D: Wireless AI research dataset,

    Y . Huangfu, J. Wang, S. Dai, R. Li, J. Wang, C. Huang, and Z. Zhang, “W AIR-D: Wireless AI research dataset,”arXiv preprint arXiv:2212.02159, 2022. [Online]. Available: https://arxiv.org/abs/2212. 02159

    2022 arXiv

  19. [27]

    Denoising diffusion implicit models,

    J. Song, C. Meng, and S. Ermon, “Denoising diffusion implicit models,” inInternational Conference on Learning Representations, 2021

    2021

  20. [28]

    Weight diffusion for future: Learn to generalize in non-stationary environments,

    M. Xie, S. Li, B. Xie, C. H. Liu, J. Liang, Z. Sun, K. Feng, and C. Zhu, “Weight diffusion for future: Learn to generalize in non-stationary environments,” inAdvances in Neural Information Processing Systems (NeurIPS), 2024

    2024

  21. [29]

    LoRAGen: Structure-aware weight space learning for LoRA generation,

    H. Huang, J. Ding, M. Liao, X. Wang, J. Ban, Y . Yuan, H. Wan, and Y . Li, “LoRAGen: Structure-aware weight space learning for LoRA generation,” inInternational Conference on Learning Representations,

  22. [30]

    Diffusion-based neural network weights generation,

    B. Soro, B. Andreis, H. Lee, W. Jeong, S. Chong, F. Hutter, and S. J. Hwang, “Diffusion-based neural network weights generation,” in International Conference on Learning Representations, Y . Yue, A. Garg, N. Peng, F. Sha, and R. Yu, Eds., vol. 2025, 2025, pp. 99 644–99 675. [O...

    2025

  23. [31]

    Metadiff: Meta-learning with conditional diffusion for few-shot learning,

    B. Zhang, C. Luo, D. Yu, X. Li, H. Lin, Y . Ye, and B. Zhang, “Metadiff: Meta-learning with conditional diffusion for few-shot learning,”Proceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 15, pp. 16 687–16 695, Mar. 2024. [Online]. Available: https://o...

    2024

  24. [2026]

    Available: https://openreview.net/forum?id=mrafO7aTYj

    [Online]. Available: https://openreview.net/forum?id=mrafO7aTYj

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.