Pith. sign in

REVIEW 3 major objections 6 minor 83 references

Zero-Fi: Zero-Shot Wi-Fi-Based Human Activity Recognition via Contrastive Signal-Language Alignment

T0 review · 3 major / 6 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read A Wi-Fi model that aligns signals with text can recognize held-out activities, averaging 69.58% accuracy.

desk verdict Plausible and well-engineered zero-shot WiFi HAR result, but the unseen-class text descriptions may be contaminated by dataset visual aids, so the 69.58% number needs verification before I'd trust it. read the letter →

arxiv 2607.26381 v1 pith:KELKDPVJ submitted 2026-07-29 cs.CV cs.AI

Yitong Shen , Cheng Guo , Peiliang Wang , Jingzhe Zhang , Yi Sheng , Haopeng Zhang , Hongfei Xue , Yili Ren This is my paper
classification cs.CVcs.AI
keywords zero-shotlearningWi-Fisensinghumanactivityrecognitioncontrastivesignal-languagealignmentchannelstateinformationlargelanguagemodelsCLIP
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 claims that zero-shot Wi-Fi human activity recognition is achievable by learning a shared embedding space between CSI-derived signal features and natural-language descriptions of activities. On two large public datasets combined into 75 classes, with seven classes held out per trial, the proposed Zero-Fi system reaches a trimmed mean accuracy of 69.58% on unseen activities, far above the best baseline at 26.79%. The key move is to replace one-hot labels with rich, LLM-generated motion descriptions encoded by a frozen CLIP text encoder, then train the signal encoders with a bidirectional contrastive loss. If correct, Wi-Fi sensing can extend to new activity categories without collecting or labeling any Wi-Fi samples for them.

What carries the argument

The carrying mechanism is the shared embedding space built by contrastive learning. Three Transformer-based encoders turn CSI-derived Doppler-frequency-shift, phase, and amplitude representations into sequence embeddings; a signal embedding extractor fuses them; a domain discriminator with gradient reversal suppresses dataset- and environment-specific variation; a frozen CLIP text encoder embeds LLM-generated descriptions of each activity; and an InfoNCE-style bidirectional contrastive loss aligns signal and text embeddings. At inference, an unseen sample is classified by cosine similarity of its signal embedding to the fixed text embeddings of candidate activity descriptions.

What would settle it

Regenerate the held-out activity descriptions without using any of the datasets' images or videos, rerun the same seven-class holdout trials, and compare accuracy; if the trimmed mean drops well below 69.58% (near the simple-label baseline), the visual aids were supplying information about the specific test sessions, and the zero-shot claim is overstated. A second check: inspect whether the visual aids for a held-out class were recorded in the same session as the CSI test samples.

Watch

Extended reading notes

Core claim

The central claim is that zero-shot generalization for Wi-Fi activity recognition works when signal embeddings and text embeddings are aligned in a shared space, because both modalities describe the same underlying human motion. The paper reports that a system trained only on seen classes can classify held-out activity classes at 69.58% average accuracy, and that each component — three complementary signal representations (Doppler frequency shift, phase, amplitude), a domain discriminator that strips environment-specific information, and LLM-generated attribute-level descriptions — contributes to the result. The authors interpret this as evidence that unseen activities can be recognized as n

Load-bearing premise

The load-bearing premise is that the LLM-generated text descriptions of the held-out classes, produced with visual references from the original datasets, carry only class-level semantic knowledge and no instance-level information that could leak the Wi-Fi test samples they describe.

Editorial extensions

If this is right

  • Deployed Wi-Fi HAR systems could add new activity classes by only writing a text description, with no new labeled signal data.
  • The same contrastive recipe may transfer to other wireless sensing modalities such as mmWave radar or UWB, where the signal carries motion information.
  • The result suggests motion attributes shared across activities are the effective currency for generalization, not class names.
  • The domain discriminator indicates that removing environment-specific signal content improves semantic transfer, which could benefit cross-environment deployment.
  • Because simple activity labels yield 48.49% versus 69.58% for descriptions, careful text construction materially changes zero-shot performance.

Reading between the lines

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

  • A testable extension is to generate unseen-class descriptions without any visual aids; if accuracy drops sharply, the reported result is partly carried by visual leakage from the datasets rather than pure semantic alignment.
  • The current evaluation measures only across activity categories, not across subjects or environments; a stricter holdout protocol could reveal how much of the gain is domain-invariance versus class-transfer.
  • One could push the approach toward attribute-level or compositional outputs, predicting motion attributes from signals and comparing against attribute decompositions, which would make the mechanism more inspectable than a single cosine match.
  • The reliance on a frozen CLIP text encoder ties the system to the visual semantics CLIP learned; a text encoder trained on motion-centric corpora might change the zero-shot curve.
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 / 6 minor

Summary. The paper proposes Zero-Fi, a contrastive signal-language alignment framework for zero-shot Wi-Fi-based human activity recognition. The method extracts complementary signal representations (amplitude, phase, DFS) from CSI, processes them with transformer encoders, aligns the resulting embeddings to the text embeddings of LLM-generated activity descriptions using a frozen CLIP text encoder, and employs a gradient-reversal domain discriminator to reduce domain-specific variation. The authors evaluate on a combined set of 75 classes from WiDAR 3.0 and XRF55, holding out 7 randomly selected classes per trial over 6 trials, and report a trimmed mean accuracy of 69.58%, substantially outperforming baselines (Table 1). Ablations show contributions of each component. The central claim is that the model can recognize unseen activity classes without any Wi-Fi samples or model adaptation for those classes.

Significance. If the central empirical claim holds, this is a notable advance: it is among the first to demonstrate that contrastive signal-language alignment can scale to a wide vocabulary of 75 activities and achieve zero-shot classification of held-out classes with nearly 70% accuracy. The design choices are reasonable and mostly well-motivated: using a frozen CLIP text encoder avoids training a text model, the domain discriminator addresses cross-dataset variability, and the LLM-based description generator enriches class-level semantics. The paper also positions itself appropriately relative to prior Wi-Fi HAR work. However, the significance is conditional on resolving the provenance of the visual aids used to generate unseen-class descriptions, as they may compromise the zero-shot assumption. The absence of error bars and per-split results further limits the strength of the evidence.

major comments (3)
  1. [Results, Table 1] The zero-shot validity of the 69.58% result hinges on the independence of the unseen-class descriptions from the test Wi-Fi data. The paper states 'The visual aids are provided with the original datasets and are used only as class-level references for generating activity descriptions,' but provides no metadata, prompts, or generated descriptions to support this claim. In WiDAR 3.0 and XRF55, visual data are typically synchronized with RF acquisitions; if the visual reference for an unseen class is drawn from the same recording session or subject as the test CSI, the LLM may encode instance-level motion cues (e.g., a specific person's stride, direction, or body geometry) that can be matched during inference, turning the result into a measure of dataset-specific leakage rather than genuine zero-shot generalization. This directly undermines the central claim. The authors must demonstrate ei
  2. [Zero-Shot Evaluation Protocol and Results] The reported performance is a one-sample-per-tail trimmed mean over six randomly chosen splits, but no per-split accuracies, standard deviations, or confidence intervals are provided. With only 7 unseen classes per trial, the variance across splits is likely substantial; the relative ordering of methods (e.g., Zero-Fi at 69.58% vs. Wi-CLIP at 26.79%) may not be stable. The paper should report the full per-split results, the untrimmed mean, and a measure of dispersion (e.g., standard deviation or bootstrap CI). Without this, the magnitude and reliability of the claimed improvement cannot be assessed.
  3. [Conclusion] The evaluation protocol designates 7 unseen classes out of 75 and states that 'participants and environments are not treated as additional holdout factors.' This means that the held-out classes come from the same subjects and environments as the training classes. While this isolates cross-category generalization, it also means the model could exploit participant- or environment-specific cues that are shared between seen and unseen classes, potentially inflating accuracy. The claim is explicitly framed as category generalization, which is acceptable, but the paper should discuss the limitation and ideally include an additional experiment with class-disjoint plus subject-disjoint splits to show robustness. As written, the experiment does not test the 'robust generalization' wording in the conclusion.
minor comments (6)
  1. [Methodology, Eq. (1)] The circled terms 1, 2, 3, 4 are rendered as '1⁄bigcircle' etc., which is garbled. Use a cleaner notation or explicitly name the terms in text.
  2. [Model and Environment Settings] The sentence 'As described in Section ,' contains a missing cross-reference. Please fill in the actual section number.
  3. [Model and Environment Settings] The paper refers to 'ChatGPT 5.5' without specifying a release date or version. This is not a standard public model name as of writing; clarify the exact model identifier and API version used.
  4. [Model and Environment Settings] The computational environment is given as 'Linux 24.04' — likely a typo for Ubuntu 24.04. Also specify the PyTorch version more precisely.
  5. [Results, Figure 5] Figure 5 shows bars without error bars or significance markers. Add error bars or at least a note on the variability of the ablation numbers.
  6. [References] Several references have future years (e.g., Zhang et al. 2026, Li et al. 2025b, Wang et al. 2024b) — verify these are preprints or accepted papers and format accordingly. Also check that all cited works appear in the reference list.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: Zero-Fi's zero-shot accuracy is an empirically trained result, not a quantity forced by its inputs.

full rationale

The paper's central claim is an empirical zero-shot accuracy (69.58%) produced by training a contrastive signal-text alignment on seen-class Wi-Fi samples and evaluating on held-out classes. The training objective (Eq. 4, InfoNCE) compares CSI-derived embeddings with frozen CLIP text embeddings of LLM-generated descriptions; unseen Wi-Fi samples are excluded from training, hyperparameter selection, receiver selection, and checkpoint selection by the stated protocol. No equation in the paper defines the predicted accuracy in terms of the fitted inputs, and no fitted parameter is renamed as a prediction. The LLM/CLIP text side is external and frozen, so the signal-text alignment is not optimized on unseen descriptions. Self-citations (e.g., Wi-Chat, Ren et al. 2025) appear only in related work and are not used to justify the framework's design or to rule out alternatives. The visual-aid provenance issue (unseen descriptions generated from original datasets' visual references) is a potential data-leakage/validity concern external to the derivation chain; since the paper asserts the aids are used only as class-level references and no reduction is shown, it is not a circular step under the required standard.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The central result is an empirical accuracy, so the ledger mostly records modeling and evaluation assumptions rather than fitted physical parameters. No new physical entities are introduced. The assumptions that carry the most weight are the cross-antenna phase-noise model, the semantic transferability of CLIP and LLM descriptions, and the independence of the visual aids used to write unseen-class descriptions.

free parameters (5)
  • Learnable log-temperature tau = learned, value not reported
    Used in contrastive similarity (Signal-Text Alignment) and in the domain discriminator's predicted activity distribution; standard CLIP temperature but tuned during training.
  • Gradient-reversal coefficient gamma_d = not reported
    Controls the adversarial pressure of the domain discriminator on the activity feature encoder; chosen by hand and affects the reported accuracy.
  • Contrastive loss weights a and b = a=b=1
    Set by default in Eq. 4; represents a design choice that could be tuned to change alignment balance.
  • Signal processing hyperparameters = not fully specified
    Temporal resampling to 1000 packets, STFT window/overlap, DFS frequency range, downsampling factor s, and patch sizes w x h are chosen manually and not all reported in the paper.
  • Evaluation design: 7 unseen classes, 6 trials, trimmed mean = 7 classes per trial, 6 trials, min/max removed
    Not a model parameter, but the reported 69.58% depends on this specific evaluation design; a different split size or more trials could change the number.
assumptions (6)
  • domain assumption Random phase offset and carrier frequency offset are identical across antennas and cancel under conjugate multiplication
    Used in Eq. 1 to justify the conjugate-multiplication denoising step; if the offsets differ across antennas, the cancellation is incomplete.
  • domain assumption Term 2 in Eq. 1 is negligible because a1*a2 is small
    The paper asserts that the product of attenuations from two antennas is small, so the term is discarded; no quantitative bound is provided.
  • domain assumption CLIP text embeddings provide semantically rich, visually grounded representations that align with Wi-Fi motion attributes
    Core premise of the cross-modal alignment; CLIP was trained on images and text, not Wi-Fi, and the transferability of its text space to radio signals is assumed.
  • domain assumption LLM-generated descriptions decompose activities into motion attributes shared across seen and unseen classes, and visual aids are legitimate class-level knowledge
    Premise of zero-shot transfer; the use of dataset visual aids for unseen classes is assumed not to leak test-instance information, which is not rigorously justified.
  • domain assumption The static CSI component is motion-invariant and removable by high-pass filtering
    Standard in Wi-Fi sensing, but treated as given in this pipeline; the static component is assumed to contain no useful activity information.
  • domain assumption Each dataset-environment configuration is a valid domain for adversarial domain-invariant training
    The domain discriminator treats dataset-environment pairs as domain labels; the paper does not hold out participants or environments, so domain invariance is only tested implicitly.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Zero-Fi: Zero-Shot Wi-Fi-Based Human Activity Recognition via Contrastive Signal-Language Alignment." pith.science (2026). https://pith.science/paper/KELKDPVJ

@misc{pith2026260726381,
  author       = {Pith},
  title        = {Pith review of: Zero-Fi: Zero-Shot Wi-Fi-Based Human Activity Recognition via Contrastive Signal-Language Alignment},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KELKDPVJ}},
  note         = {Machine review of arXiv:2607.26381}
}
read the original abstract

Wi-Fi-based human activity recognition has advanced substantially, but most existing methods assume a closed set of activities and require labeled Wi-Fi samples for every target class, limiting their ability to recognize unseen activities. We present Zero-Fi, a contrastive signal-language alignment framework for zero-shot Wi-Fi-based human activity recognition. Zero-Fi learns unified representations from complementary Wi-Fi signal features and aligns them with the semantic representations of natural-language activity descriptions in a shared embedding space. This cross-modal alignment enables Zero-Fi to recognize new activity classes without requiring labeled Wi-Fi samples or model adaptation for those classes. Experiments on large-scale public benchmark datasets demonstrate effective zero-shot recognition of held-out activity classes, highlighting the potential of signal-language alignment to extend Wi-Fi sensing beyond predefined activity classes.

Figures

Figures reproduced from arXiv: 2607.26381 by the authors.

Figure 1
Figure 1. Wi-Fi sensing illustration. Preliminary Wi-Fi Sensing Basics Wi-Fi has evolved beyond its traditional role as a commu￾nication technology to become a promising sensing modal￾ity (Tan et al. 2022). As illustrated in [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Overview of Zero-Fi. Time (s) 0 1 2 3 4 5 F r e q u e n c y ( H z ) -50 0 50 Time (s) 0 1 2 3 4 5 P h a s e ( r a d ) -: -:/2 0 :/2 : Time (s) 0 1 2 3 4 5 A m p l i t u d e 0 20 40 60 80 100 [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 4
Figure 4. LLM-based activity describer. contrastive learning. Its text encoder therefore captures vi￾sually grounded semantic information related to objects, ac￾tions, and human motion, which can provide a useful su￾pervisory representation for activity-related signal features. Accordingly, we employ the frozen CLIP text encoder to encode the generated activity descriptions for subsequent contrastive alignment with the CSI re… view at source ↗
Figures from the paper (1 more)
Figure 5
Figure 5. Figure 5: Ablation study: impact of signal representations. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

83 extracted references · 5 linked inside Pith

  1. [1]

    Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies , volume=

    IndoTrack: Device-free indoor human tracking with commodity Wi-Fi , author=. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies , volume=. 2017 , publisher=

    2017

  2. [2]

    Proceedings of the 17th annual international conference on mobile systems, applications, and services , pages=

    Zero-effort cross-domain gesture recognition with Wi-Fi , author=. Proceedings of the 17th annual international conference on mobile systems, applications, and services , pages=

  3. [3]

    ACM Computing Surveys (CSUR) , volume=

    WiFi sensing with channel state information: A survey , author=. ACM Computing Surveys (CSUR) , volume=. 2019 , publisher=

    2019

  4. [4]

    IEEE Internet of Things Journal , volume=

    Commodity WiFi sensing in ten years: Status, challenges, and opportunities , author=. IEEE Internet of Things Journal , volume=. 2022 , publisher=

    2022

  5. [5]

    The 25th Annual International Conference on Mobile Computing and Networking , pages=

    mD-Track: Leveraging multi-dimensionality for passive indoor Wi-Fi tracking , author=. The 25th Annual International Conference on Mobile Computing and Networking , pages=

  6. [6]

    Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies , volume=

    Xrf55: A radio frequency dataset for human indoor action analysis , author=. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies , volume=. 2024 , publisher=

    2024

  7. [7]

    International conference on machine learning , pages=

    Learning transferable visual models from natural language supervision , author=. International conference on machine learning , pages=. 2021 , organization=

    2021

  8. [8]

    Proceedings of the 22nd ACM conference on embedded networked sensor systems , pages=

    mmclip: Boosting mmwave-based zero-shot har via signal-text alignment , author=. Proceedings of the 22nd ACM conference on embedded networked sensor systems , pages=

Show all 83 references
  1. [9]

    Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies , volume=

    Rfboost: Understanding and boosting deep wifi sensing via physical data augmentation , author=. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies , volume=. 2024 , publisher=

    2024

  2. [10]

    arXiv preprint arXiv:2010.11929 , year=

    An image is worth 16x16 words: Transformers for image recognition at scale , author=. arXiv preprint arXiv:2010.11929 , year=

    2010 arXiv

  3. [11]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Two-stream convolution augmented transformer for human activity recognition , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  4. [12]

    IEEE Internet of Things Journal , volume=

    Diffusion-model-based contrastive learning for human activity recognition , author=. IEEE Internet of Things Journal , volume=. 2024 , publisher=

    2024

  5. [13]

    Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies , volume=

    FullBreathe: Full human respiration detection exploiting complementarity of CSI phase and amplitude of WiFi signals , author=. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies , volume=. 2018 , publisher=

    2018

  6. [14]

    IEEE Internet of Things Journal , volume=

    Passive human tracking with WiFi point clouds , author=. IEEE Internet of Things Journal , volume=. 2024 , publisher=

    2024

  7. [15]

    Proceedings of the 2017 CHI conference on human factors in computing systems , pages=

    Inferring motion direction using commodity Wi-Fi for interactive exergames , author=. Proceedings of the 2017 CHI conference on human factors in computing systems , pages=

    2017

  8. [16]

    Proceedings of the International Workshop on Environmental Sensing Systems for Smart Cities , pages=

    Wi-Chat: Large Language Model-powered Wi-Fi-based Human Activity Recognition , author=. Proceedings of the International Workshop on Environmental Sensing Systems for Smart Cities , pages=

  9. [17]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Wi-CBR: Salient-aware adaptive WiFi sensing for cross-domain behavior recognition , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  10. [18]

    ACM SIGCOMM computer communication review , volume=

    Tool release: Gathering 802.11 n traces with channel state information , author=. ACM SIGCOMM computer communication review , volume=. 2011 , publisher=

    2011

  11. [19]

    International conference on machine learning , pages=

    Learning sleep stages from radio signals: A conditional adversarial architecture , author=. International conference on machine learning , pages=. 2017 , organization=

    2017

  12. [20]

    arXiv preprint arXiv:2407.19893 , year=

    Leveraging foundation models for zero-shot IoT sensing , author=. arXiv preprint arXiv:2407.19893 , year=

    arXiv

  13. [21]

    Proceedings of the 23rd Annual International Conference on Mobile Systems, Applications and Services , pages=

    Data Can Speak for Itself: Quality-guided Utilization of Wireless Synthetic Data , author=. Proceedings of the 23rd Annual International Conference on Mobile Systems, Applications and Services , pages=

  14. [22]

    International Conference on Artificial Intelligence of Things and Systems , pages=

    Wi-CLIP: Toward Zero-Shot Air Gesture Recognition Based on RF-Text Foundation Model , author=. International Conference on Artificial Intelligence of Things and Systems , pages=. 2025 , organization=

    2025

  15. [23]

    Expert Systems , volume=

    Human activity recognition: A comprehensive review , author=. Expert Systems , volume=. 2024 , publisher=

    2024

  16. [24]

    IEEE Reviews in Biomedical Engineering , volume=

    Contactless WiFi sensing and monitoring for future healthcare-emerging trends, challenges, and opportunities , author=. IEEE Reviews in Biomedical Engineering , volume=. 2022 , publisher=

    2022

  17. [25]

    2013 Proceedings IEEE INFOCOM , pages=

    Rejecting the attack: Source authentication for wi-fi management frames using csi information , author=. 2013 Proceedings IEEE INFOCOM , pages=. 2013 , organization=

    2013

  18. [26]

    Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies , volume=

    IF-ConvTransformer: A framework for human activity recognition using IMU fusion and ConvTransformer , author=. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies , volume=. 2022 , publisher=

    2022

  19. [27]

    Proceedings of the IEEE conference on Computer Vision and Pattern Recognition , pages=

    A closer look at spatiotemporal convolutions for action recognition , author=. Proceedings of the IEEE conference on Computer Vision and Pattern Recognition , pages=

  20. [28]

    Sensors , volume=

    Deep convolutional and lstm recurrent neural networks for multimodal wearable activity recognition , author=. Sensors , volume=. 2016 , publisher=

    2016

  21. [29]

    Proceedings of the 3rd ACM Workshop on Millimeter-wave Networks and Sensing Systems , pages=

    Radhar: Human activity recognition from point clouds generated through a millimeter-wave radar , author=. Proceedings of the 3rd ACM Workshop on Millimeter-wave Networks and Sensing Systems , pages=

  22. [30]

    IEEE Internet of Things Journal , volume=

    Temporal convolutional networks for multiperson activity recognition using a 2-d lidar , author=. IEEE Internet of Things Journal , volume=. 2020 , publisher=

    2020

  23. [31]

    IEEE access , volume=

    SmartWall: Novel RFID-enabled ambient human activity recognition using machine learning for unobtrusive health monitoring , author=. IEEE access , volume=. 2019 , publisher=

    2019

  24. [32]

    Proceedings of the 2015 ACM conference on special interest group on data communication , pages=

    Spotfi: Decimeter level localization using wifi , author=. Proceedings of the 2015 ACM conference on special interest group on data communication , pages=

    2015

  25. [33]

    Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies , volume=

    Liquid level sensing using commodity wifi in a smart home environment , author=. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies , volume=. 2020 , publisher=

    2020

  26. [34]

    Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies , volume=

    GoPose: 3D human pose estimation using WiFi , author=. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies , volume=. 2022 , publisher=

    2022

  27. [35]

    32nd USENIX Security Symposium (USENIX Security 23) , pages=

    Person re-identification in 3d space: A \ WiFi \ vision-based approach , author=. 32nd USENIX Security Symposium (USENIX Security 23) , pages=

  28. [36]

    IEEE journal of biomedical and health informatics , volume=

    Wi-breath: A WiFi-based contactless and real-time respiration monitoring scheme for remote healthcare , author=. IEEE journal of biomedical and health informatics , volume=. 2022 , publisher=

    2022

  29. [37]

    arXiv preprint arXiv:2510.24936 , year=

    IBIS: A Hybrid Inception-BiLSTM and SVM Ensemble for Robust Doppler-based Human Activity Recognition , author=. arXiv preprint arXiv:2510.24936 , year=

    arXiv

  30. [38]

    arXiv preprint arXiv:2103.11943 , year=

    BERT: a review of applications in natural language processing and understanding , author=. arXiv preprint arXiv:2103.11943 , year=

    arXiv

  31. [39]

    Proceedings of the 19th ACM Conference on Embedded Networked Sensor Systems , pages=

    Onefi: One-shot recognition for unseen gesture via cots wifi , author=. Proceedings of the 19th ACM Conference on Embedded Networked Sensor Systems , pages=

  32. [40]

    IEEE Internet of Things Journal , volume=

    AutoFi: Toward automatic Wi-Fi human sensing via geometric self-supervised learning , author=. IEEE Internet of Things Journal , volume=. 2022 , publisher=

    2022

  33. [41]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Data-free generalized zero-shot learning , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  34. [42]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Zeromamba: Exploring visual state space model for zero-shot learning , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  35. [43]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Robust test-time adaptation for zero-shot prompt tuning , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  36. [44]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Softclip: Softer cross-modal alignment makes clip stronger , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  37. [45]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Vilt-clip: Video and language tuning clip with multimodal prompt learning and scenario-guided optimization , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  38. [46]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Tagclip: A local-to-global framework to enhance open-vocabulary multi-label classification of clip without training , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  39. [47]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Calip: Zero-shot enhancement of clip with parameter-free attention , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  40. [48]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Exploring clip for assessing the look and feel of images , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  41. [49]

    IEEE transactions on pattern analysis and machine intelligence , volume=

    Meta-learning in neural networks: A survey , author=. IEEE transactions on pattern analysis and machine intelligence , volume=. 2021 , publisher=

    2021

  42. [50]

    11th USENIX Symposium on Networked Systems Design and Implementation (NSDI 14) , pages=

    Bringing gesture recognition to all devices , author=. 11th USENIX Symposium on Networked Systems Design and Implementation (NSDI 14) , pages=

  43. [51]

    Sensors , volume=

    A novel human activity recognition and prediction in smart home based on interaction , author=. Sensors , volume=. 2019 , publisher=

    2019

  44. [52]

    IEEE Aerospace and Electronic Systems Magazine , volume=

    High precision human detection and tracking using millimeter-wave radars , author=. IEEE Aerospace and Electronic Systems Magazine , volume=. 2021 , publisher=

    2021

  45. [53]

    Sensors , volume=

    A 2D-Lidar-equipped unmanned robot-based approach for indoor human activity detection , author=. Sensors , volume=. 2023 , publisher=

    2023

  46. [54]

    Sensors , volume=

    A comprehensive survey of vision-based human action recognition methods , author=. Sensors , volume=. 2019 , publisher=

    2019

  47. [55]

    IEEE Transactions on Image Processing , volume=

    Deep image-to-video adaptation and fusion networks for action recognition , author=. IEEE Transactions on Image Processing , volume=. 2019 , publisher=

    2019

  48. [56]

    Proceedings of the 21st annual international conference on mobile computing and networking , pages=

    Understanding and modeling of wifi signal based human activity recognition , author=. Proceedings of the 21st annual international conference on mobile computing and networking , pages=

  49. [57]

    Proceedings of the ACM SIGCOMM 2013 conference on SIGCOMM , pages=

    See through walls with WiFi! , author=. Proceedings of the ACM SIGCOMM 2013 conference on SIGCOMM , pages=

    2013

  50. [58]

    Proceedings of the IEEE international conference on computer vision , pages=

    Youtube2text: Recognizing and describing arbitrary activities using semantic hierarchies and zero-shot recognition , author=. Proceedings of the IEEE international conference on computer vision , pages=

  51. [59]

    IEEE transactions on cybernetics , volume=

    Complementary attributes: A new clue to zero-shot learning , author=. IEEE transactions on cybernetics , volume=. 2019 , publisher=

    2019

  52. [60]

    IEEE Transactions on Mobile Computing , volume=

    Human activity recognition across scenes and categories based on CSI , author=. IEEE Transactions on Mobile Computing , volume=. 2020 , publisher=

    2020

  53. [61]

    Proceedings of the 2016 ACM international joint conference on pervasive and ubiquitous computing , pages=

    Dynamic-MUSIC: Accurate device-free indoor localization , author=. Proceedings of the 2016 ACM international joint conference on pervasive and ubiquitous computing , pages=

    2016

  54. [62]

    Proceedings of the 21st Annual International Conference on Mobile Computing and Networking , pages=

    Tonetrack: Leveraging frequency-agile radios for time-based indoor wireless localization , author=. Proceedings of the 21st Annual International Conference on Mobile Computing and Networking , pages=

  55. [63]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Latent independent excitation for generalizable sensor-based cross-person activity recognition , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  56. [64]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Frame order matters: A temporal sequence-aware model for few-shot action recognition , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  57. [65]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Person-in-wifi 3d: End-to-end multi-person 3d pose estimation with wi-fi , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  58. [66]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Multitarget device-free localization via cross-domain Wi-Fi RSS training data and attentional prior fusion , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  59. [67]

    IEEE Network , volume=

    Target-oriented wifi sensing for respiratory healthcare: from indiscriminate perception to in-area sensing , author=. IEEE Network , volume=. 2024 , publisher=

    2024

  60. [68]

    ACM Transactions on Sensor Networks , volume=

    A contactless authentication system based on WiFi CSI , author=. ACM Transactions on Sensor Networks , volume=. 2023 , publisher=

    2023

  61. [69]

    Proceedings of the AAAI Conference on Artificial intelligence , volume=

    Consense: Continually sensing human activity with wifi via growing and picking , author=. Proceedings of the AAAI Conference on Artificial intelligence , volume=

  62. [70]

    ACM Transactions on Computing for Healthcare , volume=

    WiLife: Long-term daily status monitoring and habit mining of the elderly leveraging ubiquitous Wi-fi signals , author=. ACM Transactions on Computing for Healthcare , volume=. 2025 , publisher=

    2025

  63. [71]

    Proceedings of the 22nd ACM Conference on Embedded Networked Sensor Systems , pages=

    Wi2DMeasure: WiFi-based 2D object size measurement , author=. Proceedings of the 22nd ACM Conference on Embedded Networked Sensor Systems , pages=

  64. [72]

    arXiv preprint arXiv:1412.6980 , year=

    Adam: A method for stochastic optimization , author=. arXiv preprint arXiv:1412.6980 , year=

    arXiv

  65. [73]

    Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education

    Clancey, William J. Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education. Proceedings of the Eighth International Joint Conference on Artificial Intelligence (IJCAI-83)

  66. [74]

    Classification Problem Solving

    Clancey, William J. Classification Problem Solving. Proceedings of the Fourth National Conference on Artificial Intelligence

  67. [75]

    , title =

    Robinson, Arthur L. , title =. 1980 , doi =. https://science.sciencemag.org/content/208/4447/1019.full.pdf , journal =

    1980

  68. [76]

    New Ways to Make Microcircuits Smaller---Duplicate Entry

    Robinson, Arthur L. New Ways to Make Microcircuits Smaller---Duplicate Entry. Science

  69. [77]

    Clancey and Glenn Rennels , abstract =

    Diane Warner Hasling and William J. Clancey and Glenn Rennels , abstract =. Strategic explanations for a diagnostic consultation system , journal =. 1984 , issn =. doi:https://doi.org/10.1016/S0020-7373(84)80003-6 , url =

    1984 doi

  70. [78]

    and Rennels, Glenn R

    Hasling, Diane Warner and Clancey, William J. and Rennels, Glenn R. and Test, Thomas. Strategic Explanations in Consultation---Duplicate. The International Journal of Man-Machine Studies

  71. [79]

    Poligon: A System for Parallel Problem Solving

    Rice, James. Poligon: A System for Parallel Problem Solving

  72. [80]

    Transfer of Rule-Based Expertise through a Tutorial Dialogue

    Clancey, William J. Transfer of Rule-Based Expertise through a Tutorial Dialogue

  73. [81]

    The Engineering of Qualitative Models

    Clancey, William J. The Engineering of Qualitative Models

  74. [82]

    2023 , eprint=

    Attention Is All You Need , author=. 2023 , eprint=

    2023

  75. [83]

    Pluto: The 'Other' Red Planet

    NASA. Pluto: The 'Other' Red Planet

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.