Pith. sign in

REVIEW 3 major objections 5 minor 31 references

CLAP-S: Support Set Based Adaptation for Downstream Fiber-optic Acoustic Recognition

T0 review · 3 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read For fiber-optic acoustic recognition, the best CLAP adaptation combines a fine-tuned adapter with support-set retrieval on task-aligned embeddings.

desk verdict A useful incremental paper with real new datasets and a sensible retrieval-plus-adapter combination; the headline accuracy claim currently rests on undisclosed tuning of alpha and beta. read the letter →

arxiv 2501.09877 v1 pith:V4XWJ46F submitted 2025-01-16 eess.AS cs.LG

Jingchen Sun , Shaobo Han , Wataru Kohno , Changyou Chen This is my paper · ORCID
classification eess.AScs.LG
keywords fiber-opticacousticrecognitiondistributedsensingCLAPsupportsetadaptationfew-shotlearningdomainsoundeventclassificationadapterfine-tuning
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 best way to adapt a CLAP audio model to fiber-optic acoustic sensing is not fine-tuning alone and not retrieval alone, but a linear interpolation of a fine-tuned adapter with a support-set retrieval branch. This matters because fiber-optic distributed acoustic sensing is a low-label, high-domain-shift setting where standard microphone-pretrained representations transfer poorly. The authors build a fiber-optic version of ESC-50 and use a real gunshot-versus-firework dataset collected from telecom cables. Their results show the hybrid CLAP-S+ outperforms prompt tuning, adapter, Treff, and Tip-Adapter-F, and that the training-free CLAP-S outperforms Tip-Adapter. They also find that zero-shot text-aligned knowledge can hurt and that task-aligned embeddings give better retrieval.

What carries the argument

The machinery is the support set: the $NK$ labeled training audios are stored as key embeddings $F_{\text{train}}\in\mathbb{R}^{NK\times C}$ with one-hot label values $L_{\text{train}}$. A test query $u$ produces a retrieval distribution $p_{\text{support}}(y|x,u)=e^{-\beta(1-uF_{\text{train}}^T)}L_{\text{train}}^T$ via cross-attention, and a two-layer MLP adapter maps the text-aligned embedding $u_0$ to a task-aligned embedding $u_f$. CLAP-S+ sets both query and keys to $u_f$ and outputs $(1-\alpha)p_{\text{clap}}+\alpha p_{\text{support}}$, while CLAP-S uses $u_0$ with $\alpha=1$ and no training. This machinery carries the argument because the same labeled samples feed both the adapter and the retrieval memory, and because the ablation ties the gain to using task-aligned, not text-aligned, keys.

What would settle it

Using the released gunshot-firework dataset and the same train/validation/test splits, rerun CLAP-S+ with the retrieval branch disabled (alpha=0) and with support keys replaced by text-aligned embeddings; if either variant matches or beats the full CLAP-S+, or if the same-class minus different-class cosine similarity among fine-tuned test embeddings is not positive, the paper's claim that task-aligned support-set interpolation drives the gains is refuted.

Watch

Extended reading notes

Core claim

The paper's central claim is that CLAP can be adapted to fiber-optic distributed acoustic sensing by using the small labeled set twice: once to fine-tune a two-layer MLP adapter and once as a key-value support set for retrieval. The final prediction is a linear blend of the adapter's class distribution and the retrieval distribution. The authors report that when query and keys both use the task-aligned adapter embeddings, this blend, CLAP-S+, is the most accurate of the studied methods, averaging 78.8% accuracy across five datasets versus 76.5% for Tip-Adapter-F, the strongest baseline; with no training at all, the pure-retrieval CLAP-S reaches 71.6% and beats Tip-Adapter's 69.2%.

Load-bearing premise

The method assumes that after adapter fine-tuning, the CLAP audio encoder's embedding space keeps same-class fiber-optic sounds closer together than different-class sounds, so nearest-neighbor retrieval from the support set is a trustworthy signal.

Editorial extensions

If this is right

  • On the five datasets reported, CLAP-S+ raises average accuracy to 78.8%, the best among prompt tuning, adapter, Treff, and Tip-Adapter-F.
  • Without any training, CLAP-S reaches 71.6% average accuracy and beats Tip-Adapter (69.2%) while using no extra parameters and slightly less inference time.
  • Task-aligned adapter embeddings outperform text-aligned embeddings as support-set keys and queries in this domain.
  • Adding zero-shot CLAP predictions to either the support-set or adapter predictions can reduce accuracy, so zero-shot knowledge is not always beneficial under large domain shift.
  • Training one adapter jointly on multiple fiber-optic device domains outperforms training separate adapters per domain.

Reading between the lines

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

  • A natural extension the paper does not run is a systematic sweep of domain-shift severity; the larger gains on the outdoor real-world tasks suggest the retrieval branch matters most when CLAP's text-aligned space is most distorted.
  • Because alpha is selected on validation labels, a deployment version would need a label-free rule for alpha; one testable option is to set it from retrieval confidence or from the entropy of the support-set distribution.
  • If the mechanism transfers, the same adapter-plus-support-set recipe could be tried for other acoustic domains far from microphone pretraining, such as hydrophone or respiratory sound classification, where labels are scarce and language descriptions are weak.
  • The paper's negative result for zero-shot knowledge implies that removing the text branch entirely, or replacing class names with learned or noisy captions, could further improve adaptation; that is a direct testable consequence.
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. The paper proposes CLAP-S and CLAP-S+, two adaptation methods for using the CLAP audio-language model on fiber-optic acoustic recognition, a domain with significant distribution shift and limited labeled data. CLAP-S performs training-free classification by retrieving labels from a support set via cross-attention between test and training audio embeddings. CLAP-S+ combines a fine-tuned two-layer MLP adapter with support-set retrieval, interpolating the adapter's prediction and the retrieval prediction through a weighted sum. The authors evaluate on laboratory-recorded fiber-optic versions of ESC-50 and a real-world gunshot-firework dataset, reporting that CLAP-S+ achieves the highest average accuracy among the studied baselines (Prompt Tuning, Adapter, Treff, Tip-Adapter, Tip-Adapter-F), along with efficiency comparisons and ablations on the contributions of zero-shot knowledge and task-aligned embeddings.

Significance. If the empirical claims hold, the paper offers a simple, practical recipe for adapting a general-purpose audio foundation model to a sensor domain with a large domain gap: fine-tune a small adapter on the support set and use the task-aligned embeddings for both classification and support-set retrieval. The paper also introduces a new real-world DAS gunshot-firework dataset and releases code, which are valuable resources. The conceptual framing of 'implicit' versus 'explicit' knowledge is useful, and the ablation showing that zero-shot text-aligned knowledge can hurt in this domain is a noteworthy caveat for practitioners. However, the strength of the central 'highest accuracy' claim is currently limited by unreported hyperparameter selection, a suspiciously weak Prompt Tuning baseline, and the lack of statistical significance assessment for several small margins.

major comments (3)
  1. [Section II, Eq. (1)] The interpolation weight alpha and the kernel sharpness beta are described only as "tuned parameter" and "sharpness parameter" with no reported values, no statement of whether they were selected on the validation splits described in Table III, and no sensitivity analysis. This is load-bearing because the claimed advantages over Tip-Adapter-F are only +2.4 and +1.2 percentage points on FMO and FCO, so the choice of alpha and beta could change the ranking. Please report the per-dataset (or global) values used, specify the selection protocol, and provide a sensitivity study over alpha and beta.
  2. [Section III-B, Table II] The Prompt Tuning baseline achieves 4.0±4.0% and 5.0±5.2% on the real-world FMO and FCO tasks, which is below the 12.5% chance level for an 8-class problem. Such a result strongly suggests a misconfigured or undertrained baseline, not a faithful comparison. The paper's hypothesis that Prompt Tuning is ineffective on unseen tasks does not explain performance below chance. As presented, the baseline comparison is unfair, and the statement that CLAP-S+ "achieves the highest accuracy among all the baseline methods" is weakened. Please verify the Prompt Tuning implementation (e.g., prompt length, learning rate, epochs, or per-class text initialization) or re-run with a properly tuned configuration.
  3. [Section III-B, Table II] Several of the reported accuracy differences between CLAP-S+ and the strongest baselines are within one standard deviation of the reported runs; for example, on FM the result is 70.0±0.8 versus Adapter's 68.8±1.0 and Tip-Adapter-F's 68.6±1.0, and on FCO it is 92.0±1.7 versus 90.2±0.7. The paper does not report per-seed results, confidence intervals, or any paired significance test. The claim that CLAP-S+ is the best method across all five datasets therefore needs statistical support; please add significance tests or at least report the individual run outcomes so the variability can be assessed.
minor comments (5)
  1. [Throughout] The method is inconsistently referred to as both "CLAP-S+" and "CLAP-S-+"; please unify the notation.
  2. [Section I] There is a typo in "We focusing on how the pre-trained knowledge" — the verb form should be fixed.
  3. [Section III-B, Table II] The table header contains "Trainin-Free" — this should be corrected to "Training-Free."
  4. [Section II, Table I] The relation with Tip-Adapter and Treff would be clearer if the table indicated which embedding is used for the query versus keys in each method; the current notation "u0" and "uf" is not defined inside the table itself.
  5. [Section III-E, Table V] The claim that task-aligned embeddings are more effective for retrieval is only indirectly supported by final accuracy; reporting a direct retrieval metric (e.g., nearest-neighbor accuracy of the support-set keys) would strengthen this ablation.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the central accuracy claim is an empirical comparison against external baselines on held-out test splits, and the self-citations are not load-bearing.

full rationale

The paper's central claim is that CLAP-S+ achieves the highest accuracy among the studied baselines (Section III-B, Table II). This claim is supported by experiments on laboratory and real-world fiber-optic datasets with held-out test splits; the support set consists of training samples and their labels, and the test queries are not in the support set, so psupport = e^{-β(1-uF_train^T)}L_train^T is a standard few-shot retrieval predictor rather than a circular reuse of the test label. The interpolation in Eq. (1) combines the fine-tuned adapter's logits with the support-set retrieval; neither quantity is defined in terms of the test label being predicted. The paper does invoke self-citations: [18] (prompt-tuning adapter by one of the present authors), [27] (DAS gunshot-firework dataset by a co-author), and [29] (fiber-optic sensor hardware by overlapping authors). These are used as background, baseline, or dataset sources, and none of them supplies the load-bearing step that would force the reported superiority; the comparisons are against Tip-Adapter, Treff, Adapter, and Prompt Tuning on external benchmarks. The unreported values of α and β and the lack of a sensitivity analysis are an evaluation-protocol transparency concern, not a circularity: choosing hyperparameters on the test set would inflate accuracy, but that would be data leakage or overfitting, not an equation that reduces to its own input by construction. No uniqueness theorem, ansatz-by-citation, or renaming of a known result as a new derivation is present. The paper also explicitly scopes its domain conclusion ('This conclusion applies only to the specific domain considered'), which is an honest limitation and not a circular step. Therefore the paper receives a low circularity score.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The method relies on standard few-shot retrieval assumptions: CLAP embeddings are discriminative for the target domain, the support set represents the test distribution, and the exponential similarity kernel with tuned sharpness is appropriate. The only numbers directly fitted to the data are the interpolation weight α and kernel sharpness β, neither of which is reported with values or selection protocol. No new entities are introduced.

free parameters (2)
  • α (interpolation weight) = not reported
    Appears in Eq. (1); paper says it is a tuned parameter balancing implicit and explicit knowledge. Values and per-dataset or global tuning protocol are not given.
  • β (kernel sharpness) = not reported
    Appears in the support-set prediction formula; described as the sharpness parameter but its value and tuning are not reported.
assumptions (3)
  • domain assumption CLAP audio embeddings preserve class similarity in the fiber-optic acoustic domain despite domain shift.
    The whole retrieval mechanism depends on embedding distance corresponding to class similarity; invoked in Section II when computing u and Ftrain.
  • domain assumption The K-shot support set is representative of the test distribution for each class.
    Standard few-shot assumption; the method stores only the available labeled samples as keys, so retrieval quality is bounded by how representative they are. Used in all experiments.
  • ad hoc to paper The exponential kernel e^{-β(1-uF^T)} with tuned β is a suitable similarity measure for audio embeddings.
    Adopted from Tip-Adapter without independent justification; β is tuned on the validation set. Introduced in Section II.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CLAP-S: Support Set Based Adaptation for Downstream Fiber-optic Acoustic Recognition." pith.science (2026). https://pith.science/paper/V4XWJ46F

@misc{pith2026250109877,
  author       = {Pith},
  title        = {Pith review of: CLAP-S: Support Set Based Adaptation for Downstream Fiber-optic Acoustic Recognition},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/V4XWJ46F}},
  note         = {Machine review of arXiv:2501.09877}
}
read the original abstract

Contrastive Language-Audio Pretraining (CLAP) models have demonstrated unprecedented performance in various acoustic signal recognition tasks. Fiber-optic-based acoustic recognition is one of the most important downstream tasks and plays a significant role in environmental sensing. Adapting CLAP for fiber-optic acoustic recognition has become an active research area. As a non-conventional acoustic sensor, fiber-optic acoustic recognition presents a challenging, domain-specific, low-shot deployment environment with significant domain shifts due to unique frequency response and noise characteristics. To address these challenges, we propose a support-based adaptation method, CLAP-S, which linearly interpolates a CLAP Adapter with the Support Set, leveraging both implicit knowledge through fine-tuning and explicit knowledge retrieved from memory for cross-domain generalization. Experimental results show that our method delivers competitive performance on both laboratory-recorded fiber-optic ESC-50 datasets and a real-world fiber-optic gunshot-firework dataset. Our research also provides valuable insights for other downstream acoustic recognition tasks. The code and gunshot-firework dataset are available at https://github.com/Jingchensun/clap-s.

Figures

Figures reproduced from arXiv: 2501.09877 by the authors.

Figure 1
Figure 1. The pipeline of our proposed method. A test sample is sent to the frozen pre-trained audio encoder and fine-tuned adapter to obtain the embedding, which is then used to perform cross-attention with the keys from the support audio samples. The attention weights are further multiplied by the values of the support set to serve as Explicit Knowledge. The final prediction is obtained by Linear interpolation with the Expl… view at source ↗
Figure 2
Figure 2. The Few-Shot Adaptation Results. D. Efficiency Comparison We also compared the efficiency of our method with other baselines, as shown in Table IV. CLAP-S demonstrated the best efficiency, requiring no training or additional param￾eter storage, with the lowest training and inference time. However, this comes at the cost of sub-optimal performance. In contrast, our training-required version of CLAP-S+ achieves the hi… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references · 28 canonical work pages

  1. [1]

    Using global existing fiber networks for envi- ronmental sensing,

    Ezra Ip, Fabien Ravet, Hugo Martins, Ming-Fang Huang, Tatsuya Okamoto, Shaobo Han, Chaitnaya Narisetty, Jian Fang, Yue-Kai Huang, Milad Salemi, et al., “Using global existing fiber networks for envi- ronmental sensing,” Proceedings of the IEEE , vol. 110, no. 11, pp. 1853–1888, 2022. 1

    work page 2022

  2. [2]

    Distributed fibre optic sensors for pipeline protection,

    Femi Tanimola and David Hill, “Distributed fibre optic sensors for pipeline protection,” Journal of Natural Gas Science and Engineering , vol. 1, no. 4-5, pp. 134–143, 2009. 1

    work page 2009

  3. [3]

    Rail crack detection by analyzing the acoustic transmission process based on fiber distributed acoustic sensor,

    Cunzheng Fan, Fan Ai, Yijie Liu, Zhijie Xu, Geng Wu, Wei Zhang, Chen Liu, Zhijun Yan, Deming Liu, and Qizhen Sun, “Rail crack detection by analyzing the acoustic transmission process based on fiber distributed acoustic sensor,” in 2019 Optical Fiber Communications Conference and Exhibition (OFC) . IEEE, 2019, pp. 1–3. 1

    work page 2019

  4. [4]

    Drone detection and localization using enhanced fiber-optic acoustic sensor and distributed acoustic sensing technology,

    Jian Fang, Yaowen Li, Philip N Ji, and Ting Wang, “Drone detection and localization using enhanced fiber-optic acoustic sensor and distributed acoustic sensing technology,” Journal of Lightwave Technology, vol. 41, no. 3, pp. 822–831, 2022. 1

    work page 2022

  5. [5]

    Automatic fine-grained localization of utility pole landmarks on distributed acoustic sensing traces based on bilinear resnets,

    You Lu, Yue Tian, Shaobo Han, Eric Cosatto, Sarper Ozharar, and Yangmin Ding, “Automatic fine-grained localization of utility pole landmarks on distributed acoustic sensing traces based on bilinear resnets,” in ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2021, pp. 4675–4679. 1

    work page 2021

  6. [6]

    Utility pole localization by learning from ambient traces on distributed acoustic sensing,

    Zhuocheng Jiang, Yue Tian, Yangmin Ding, Sarper Ozharar, and Ting Wang, “Utility pole localization by learning from ambient traces on distributed acoustic sensing,” in ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2023, pp. 1–5. 1

    work page 2023

  7. [7]

    Distributed acoustic sensing for seismic activity monitoring,

    Mar ´ıa R Fern ´andez-Ruiz, Marcelo A Soto, Ethan F Williams, Sonia Martin-Lopez, Zhongwen Zhan, Miguel Gonzalez-Herraez, and Hugo F Martins, “Distributed acoustic sensing for seismic activity monitoring,” Apl Photonics, vol. 5, no. 3, 2020. 1

    work page 2020

  8. [8]

    Long- term monitoring and analysis of brood x cicada activity by distributed fiber optic sensing technology,

    Sarper Ozharar, Jessica L Ware, Yue Tian, and Yangmin Ding, “Long- term monitoring and analysis of brood x cicada activity by distributed fiber optic sensing technology,” Journal of Insect Science , vol. 23, no. 6, pp. 3, 2023. 1

    work page 2023

Show all 31 references
  1. [9]

    Eavesdropping at the speed of light: Distributed acoustic sensing of baleen whales in the arctic,

    L ´ea Bouffaut, Kittinat Taweesintananon, Hannah J Kriesell, Robin A Rørstadbotnen, John R Potter, Martin Landrø, St ˚ale E Johansen, Jan K Brenne, Aksel Haukanes, Olaf Schjelderup, et al., “Eavesdropping at the speed of light: Distributed acoustic sensing of baleen whales in ...

    2022

  2. [10]

    Distributed optical fiber hydrophone based on ϕ-otdr and its field test,

    Bin Lu, Bingyan Wu, Jinfeng Gu, Junqi Yang, Kan Gao, Zhaoyong Wang, Lei Ye, Qing Ye, Ronghui Qu, Xiaobao Chen, et al., “Distributed optical fiber hydrophone based on ϕ-otdr and its field test,” Optics express, vol. 29, no. 3, pp. 3147–3162, 2021. 1

    2021

  3. [11]

    Characteristics and explanations of interference fading of a ϕ- otdr with a multi-frequency source,

    Jun Zhou, Zhengqing Pan, Qing Ye, Haiwen Cai, Ronghui Qu, and Zujie Fang, “Characteristics and explanations of interference fading of a ϕ- otdr with a multi-frequency source,” Journal of Lightwave Technology , vol. 31, no. 17, pp. 2947–2954, 2013. 1

    2013

  4. [12]

    Low-rank constrained multichannel signal denoising considering channel-dependent sensitivity inspired by self-supervised learning for optical fiber sensing,

    Noriyuki Tonami, Wataru Kohno, Sakiko Mishima, Yumi Arai, Reishi Kondo, and Tomoyuki Hino, “Low-rank constrained multichannel signal denoising considering channel-dependent sensitivity inspired by self-supervised learning for optical fiber sensing,” in ICASSP 2024- 2024 IEEE I...

    2024

  5. [13]

    Clap learning audio concepts from natural language supervision,

    Benjamin Elizalde, Soham Deshmukh, Mahmoud Al Ismail, and Huam- ing Wang, “Clap learning audio concepts from natural language supervision,” in ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2023, pp. 1–5. 1

    2023

  6. [14]

    Natural language supervision for general-purpose audio representations,

    Benjamin Elizalde, Soham Deshmukh, and Huaming Wang, “Natural language supervision for general-purpose audio representations,” in ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2024, pp. 336–340. 1

    2024

  7. [15]

    Learning to prompt for vision-language models,

    Kaiyang Zhou, Jingkang Yang, Chen Change Loy, and Ziwei Liu, “Learning to prompt for vision-language models,” International Journal of Computer Vision , vol. 130, no. 9, pp. 2337–2348, 2022. 1, 2, 3

    2022

  8. [16]

    Audio-free prompt tuning for language-audio models,

    Yiming Li, Xiangdong Wang, and Hong Liu, “Audio-free prompt tuning for language-audio models,” in ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2024, pp. 491–495. 1, 2

    2024

  9. [17]

    Palm: Few-shot prompt learning for audio language models,

    Asif Hanif, Maha Tufail Agro, Mohammad Areeb Qazi, and Hanan Aldarmaki, “Palm: Few-shot prompt learning for audio language models,” arXiv preprint arXiv:2409.19806 , 2024. 1

    2024 arXiv

  10. [18]

    Prompt tuning based adapter for vision-language model adaption,

    Jingchen Sun, Jiayu Qin, Zihao Lin, and Changyou Chen, “Prompt tuning based adapter for vision-language model adaption,” arXiv preprint arXiv:2303.15234, 2023. 1

    2023 arXiv

  11. [19]

    Clip-adapter: Better vision-language models with feature adapters,

    Peng Gao, Shijie Geng, Renrui Zhang, Teli Ma, Rongyao Fang, Yongfeng Zhang, Hongsheng Li, and Yu Qiao, “Clip-adapter: Better vision-language models with feature adapters,” International Journal of Computer Vision, vol. 132, no. 2, pp. 581–595, 2024. 1, 2, 3

    2024

  12. [20]

    Tip-adapter: Training- free clip-adapter for better vision-language modeling,

    Renrui Zhang, Rongyao Fang, Wei Zhang, Peng Gao, Kunchang Li, Jifeng Dai, Yu Qiao, and Hongsheng Li, “Tip-adapter: Training- free clip-adapter for better vision-language modeling,” arXiv preprint arXiv:2111.03930, 2021. 1, 2, 3

    2021 arXiv

  13. [21]

    Large-scale contrastive language- audio pretraining with feature fusion and keyword-to-caption augmen- tation,

    Yusong Wu, Ke Chen, Tianyu Zhang, Yuchen Hui, Taylor Berg- Kirkpatrick, and Shlomo Dubnov, “Large-scale contrastive language- audio pretraining with feature fusion and keyword-to-caption augmen- tation,” in ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech an...

    2023

  14. [22]

    An adapter based pre-training for efficient and scalable self-supervised speech represen- tation learning,

    Samuel Kessler, Bethan Thomas, and Salah Karout, “An adapter based pre-training for efficient and scalable self-supervised speech represen- tation learning,” in ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2022, pp....

    2022

  15. [23]

    Generalization through memorization: Nearest neighbor language models,

    Urvashi Khandelwal, Omer Levy, Dan Jurafsky, Luke Zettlemoyer, and Mike Lewis, “Generalization through memorization: Nearest neighbor language models,” arXiv preprint arXiv:1911.00172 , 2019. 1

    1911 arXiv

  16. [24]

    Resmem: Learn what you can and memorize the rest,

    Zitong Yang, Michal Lukasik, Vaishnavh Nagarajan, Zonglin Li, Ankit Rawat, Manzil Zaheer, Aditya K Menon, and Sanjiv Kumar, “Resmem: Learn what you can and memorize the rest,” Advances in Neural Information Processing Systems , vol. 36, 2024. 1

    2024

  17. [25]

    Adapting language-audio models as few-shot audio learners,

    Jinhua Liang, Xubo Liu, Haohe Liu, Huy Phan, Emmanouil Benetos, Mark D Plumbley, and Wenwu Wang, “Adapting language-audio models as few-shot audio learners,” arXiv preprint arXiv:2305.17719 , 2023. 1, 2, 3, 4

    2023 arXiv

  18. [26]

    ESC: Dataset for Environmental Sound Classification,

    Karol J. Piczak, “ESC: Dataset for Environmental Sound Classification,” in Proceedings of the 23rd Annual ACM Conference on Multimedia . pp. 1015–1018, ACM Press. 1, 3

  19. [27]

    Deep learning-based intrusion detection and impulsive event classification for distributed acoustic sensing across telecom networks,

    Shaobo Han, Ming-Fang Huang, Tingfeng Li, Jian Fang, Zhuocheng Jiang, and Ting Wang, “Deep learning-based intrusion detection and impulsive event classification for distributed acoustic sensing across telecom networks,” Journal of Lightwave Technology , 2024. 1, 3

    2024

  20. [28]

    Attention is all you need,

    A Vaswani, “Attention is all you need,” Advances in Neural Information Processing Systems, 2017. 2

    2017

  21. [29]

    Text-guided device-realistic sound generation for fiber- based sound event classification,

    Wataru Kohno, Noriyuki Tonami, Jian Fang, Shaobo Han, Jingchen Sun, and Ting Wang, “Text-guided device-realistic sound generation for fiber- based sound event classification,” IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , 2025. 3

    2025

  22. [30]

    Distributed fiber-optic sensor as an acoustic communication receiver array,

    Wataru Kohno, Jian Fang, Shuji Murakami, Giovanni Milione, and Ting Wang, “Distributed fiber-optic sensor as an acoustic communication receiver array,” 28th International Conference on Optical Fiber Sensors, p. Tu3.33, 2023. 3

    2023

  23. [31]

    Towards open respiratory acoustic foundation models: Pretraining and benchmarking,

    Yuwei Zhang, Tong Xia, Jing Han, Yu Wu, Georgios Rizos, Yang Liu, Mohammed Mosuily, Jagmohan Chauhan, and Cecilia Mascolo, “Towards open respiratory acoustic foundation models: Pretraining and benchmarking,” in The Thirty-eight Conference on Neural Information Processing Syste...

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.