pith. sign in

arxiv: 2605.19031 · v1 · pith:7KU43LGSnew · submitted 2026-05-18 · 💻 cs.AI · eess.SP

KAN-MLP-Mixer: A comprehensive investigation of the usage of Kolmogorov-Arnold Networks (KANs) for improving IMU-based Human Activity Recognition

Mengxi Liu , Sizhen Bian , Vitor Fortes , Francisco Calatrava Nicolas , Daniel Gei{\ss}ler , Maximilian Kiefer-Emmanouilidis , Bo Zhou , Paul Lukowicz This is my paper

Pith reviewed 2026-05-20 10:32 UTC · model grok-4.3

classification 💻 cs.AI eess.SP
keywords Kolmogorov-Arnold NetworksHuman Activity RecognitionIMU sensorsHybrid neural networksWearable computingActivity classificationSensor data processing
0
0 comments X

The pith

A hybrid architecture places KAN modules only at input embedding and final classification while using MLPs for feature mixing to improve IMU-based human activity recognition.

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

The paper shows that Kolmogorov-Arnold Networks learn precise functions well on clean low-dimensional data but lose accuracy on noisy real-world IMU signals, whereas MLPs tolerate noise better yet lack the same precision. To solve this, the authors test multiple placements of KAN components inside standard HAR networks and identify one effective combination: a KAN layer for initial embedding, conventional MLP layers for intermediate mixing, and a specialized LarctanKAN head for activity classification. Across eight public datasets this hybrid delivers a 5.33 percent relative gain in average macro F1 score over a pure-MLP baseline and also raises performance when inserted into other published HAR models. The result matters for wearable sensing because it gives a concrete recipe for combining the strengths of both network types without replacing every component.

Core claim

Replacing every MLP component with KANs degrades both accuracy and efficiency on noisy IMU data, but a targeted hybrid that applies KANs only for input embedding and for a final LarctanKAN classification head while retaining MLP layers for feature mixing produces more accurate and robust activity recognition models.

What carries the argument

The KAN-MLP hybrid architecture that uses a KAN-based input embedding layer, MLP layers for intermediate feature mixing, and a specialized LarctanKAN module for final activity classification.

If this is right

  • The hybrid strategy raises macro F1 scores on every one of the eight tested public HAR datasets relative to pure-MLP baselines.
  • Inserting the same KAN placement pattern into other published state-of-the-art HAR architectures produces consistent performance gains.
  • KAN modules deliver their precision advantage only when noise-tolerant MLP layers handle the bulk of the intermediate representation work.
  • Real-world wearable sensing benefits from this selective use of KANs rather than wholesale replacement of existing MLP pipelines.

Where Pith is reading between the lines

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

  • The same placement rule could be tested on other noisy time-series tasks such as fall detection or gesture recognition from wrist sensors.
  • Edge-device implementations might further benefit if the LarctanKAN head is replaced by a lighter conventional layer while keeping the input KAN.
  • Collecting a small calibration set from a new user or sensor type and fine-tuning only the KAN modules could preserve the reported gains without full retraining.

Load-bearing premise

The specific placement of KAN modules at the input and classification stages stays optimal and does not overfit when the model encounters new IMU datasets or different sensor noise patterns.

What would settle it

Training the hybrid model on the eight datasets and then evaluating it on an independent ninth public HAR dataset collected with different sensors or noise levels to check whether the 5.33 percent average F1 improvement still appears.

Figures

Figures reproduced from arXiv: 2605.19031 by Bo Zhou, Daniel Gei{\ss}ler, Francisco Calatrava Nicolas, Maximilian Kiefer-Emmanouilidis, Mengxi Liu, Paul Lukowicz, Sizhen Bian, Vitor Fortes.

Figure 1
Figure 1. Figure 1: Comparison of model predictions on synthetic functions representing typical characteristics of sensor data. The step function [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The proposed hybrid network architecture KAN-MLP-Mixer based on an empirical study. It consists of three modules: KAN [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Average performance improvement compared to the MLPHAR baseline across eight datasets using the hybrid model with [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Performance comparison for KAN-MLP-Mixer and MLPHAR models on five datasets under three sensor configurations (single [PITH_FULL_IMAGE:figures/full_fig_p015_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Performance comparison for KAN-MLP-Mixer and MLPHAR models under three window size configurations. Numerical [PITH_FULL_IMAGE:figures/full_fig_p016_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: The extending hybrid design across diverse neural backbones, the original models only have the pure convolutional layers on [PITH_FULL_IMAGE:figures/full_fig_p017_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Number of parameters in different components of KAN-MLP-Mixer and MLPHAR models.(The [PITH_FULL_IMAGE:figures/full_fig_p018_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Parameter-efficiency when scaling models comparing between KAN-MLP-Mixer and MLPHAR across eight benchmark [PITH_FULL_IMAGE:figures/full_fig_p019_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Computational efficiency comparison between KAN-MLP-Mixer and MLPHAR across eight benchmark datasets. For KAN [PITH_FULL_IMAGE:figures/full_fig_p020_9.png] view at source ↗
read the original abstract

Kolmogorov-Arnold Networks (KANs) have demonstrated an exceptional ability to learn complex functions on clean, low-dimensional data but struggle to maintain performance on noisy and imperfect real-world datasets. In contrast, conventional multi-layer perceptrons (MLPs) are far more tolerant to noise and computationally efficient. Replacing all MLP components with KANs in HAR models often degrades accuracy and computation efficiency, highlighting an open challenge: how to combine KANs' precision with MLPs' noise robustness and efficiency. To address this, we systematically explore various placements of KAN modules within deep HAR networks and propose a hybrid architecture that strategically synergizes the strengths of both paradigms, which uses a KAN-based input embedding layer, retains MLP layers for intermediate feature mixing, and introduces a specialized LarctanKAN module for final activity classification. Across eight public HAR datasets, the hybrid KAN-MLP model achieves an average macro F1 score relative improvement of 5.33\% compared pure-MLP model, significantly outperforming standalone KAN and MLP baselines. Furthermore, integrating this hybrid strategy into other state-of-the-art HAR architectures consistently boosts their performance. Our findings demonstrate that a carefully orchestrated combination of KAN, MLP, or other conventional neural components yields more robust and accurate HAR models for real-world wearable sensing environments.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper investigates the integration of Kolmogorov-Arnold Networks (KANs) into IMU-based Human Activity Recognition (HAR) models. It identifies limitations of pure KANs on noisy real-world data and proposes a hybrid KAN-MLP architecture using KAN-based input embedding, MLP layers for feature mixing, and a specialized LarctanKAN module for classification. Experiments across eight public HAR datasets report an average 5.33% relative macro F1 improvement over pure-MLP baselines, with the hybrid strategy also boosting other state-of-the-art HAR architectures.

Significance. If the performance gains hold under subject-independent evaluation protocols, the work offers a practical hybrid strategy that combines KAN precision with MLP robustness for noisy IMU signals. The consistent improvements across multiple datasets and successful integration into existing SOTA models represent a useful empirical contribution to wearable sensing, particularly if accompanied by reproducible code or full ablation tables.

major comments (2)
  1. [§4.2] §4.2 (Experimental Setup): The evaluation protocol is not explicitly described as using subject-independent splits such as leave-one-subject-out or equivalent. IMU HAR performance is known to be sensitive to subject identity and sensor bias; without confirmation of no leakage, the reported 5.33% macro F1 gain and claims of robustness to real-world environments risk being partly attributable to memorization of individual patterns rather than genuine generalization.
  2. [Table 2] Table 2 (or equivalent results table): The 5.33% average relative improvement is presented without per-dataset standard deviations, statistical significance tests (e.g., paired t-tests or Wilcoxon), or full ablation details on KAN placement variants. This leaves open the possibility of post-hoc selection effects and weakens the central claim that the specific hybrid placement is optimal.
minor comments (2)
  1. [Abstract] Abstract: The reported 5.33% figure should include error bars or confidence intervals and a brief note on the number of runs or statistical testing to improve transparency.
  2. [§3.3] §3.3 (LarctanKAN module): The definition and activation function of the specialized LarctanKAN could be clarified with an explicit equation or pseudocode to aid reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback on our manuscript investigating hybrid KAN-MLP architectures for IMU-based human activity recognition. We address each major comment point by point below and outline the revisions we will make.

read point-by-point responses

  1. Referee: [§4.2] §4.2 (Experimental Setup): The evaluation protocol is not explicitly described as using subject-independent splits such as leave-one-subject-out or equivalent. IMU HAR performance is known to be sensitive to subject identity and sensor bias; without confirmation of no leakage, the reported 5.33% macro F1 gain and claims of robustness to real-world environments risk being partly attributable to memorization of individual patterns rather than genuine generalization.

    Authors: We appreciate the referee's emphasis on rigorous evaluation protocols in IMU-based HAR, where subject-specific biases can indeed affect generalization. The public datasets used in our experiments follow their standard predefined train/test splits, which are designed to be subject-independent to avoid leakage. However, we agree that explicit clarification is essential. In the revised manuscript, we will update §4.2 to explicitly describe the evaluation protocol, confirm the use of subject-independent splits across all eight datasets, and discuss the absence of subject leakage. This will better support our claims of robustness in real-world settings. revision: yes

  2. Referee: Table 2 (or equivalent results table): The 5.33% average relative improvement is presented without per-dataset standard deviations, statistical significance tests (e.g., paired t-tests or Wilcoxon), or full ablation details on KAN placement variants. This leaves open the possibility of post-hoc selection effects and weakens the central claim that the specific hybrid placement is optimal.

    Authors: We agree that additional statistical details and comprehensive ablations would strengthen the empirical claims. In the revised version, we will expand Table 2 (and related results) to include per-dataset standard deviations computed over multiple runs with different random seeds, report results of paired statistical significance tests (e.g., paired t-tests) on the macro F1 improvements, and provide fuller ablation tables detailing all KAN placement variants explored. These additions will reduce concerns about selection effects and more robustly justify the optimality of the proposed hybrid strategy. revision: yes

Circularity Check

0 steps flagged

No circularity: purely empirical comparison on public datasets

full rationale

The paper conducts an empirical investigation of KAN placements within HAR networks, proposing a hybrid KAN-MLP architecture and reporting a 5.33% average macro F1 improvement across eight public datasets. All claims rest on direct experimental results against baselines rather than any derivation, equation, or first-principles result. No self-definitional steps, fitted inputs renamed as predictions, or load-bearing self-citations appear in the provided text or abstract. The work is self-contained against external benchmarks and does not invoke uniqueness theorems or ansatzes from prior author work.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The work rests on standard neural-network assumptions about noise robustness and empirical validation rather than new mathematical axioms or invented physical entities.

free parameters (1)
  • LarctanKAN design choices
    The specialized final classification module introduces design decisions whose exact parameterization is fitted during training on the target datasets.
axioms (1)
  • domain assumption KANs perform well on clean low-dimensional data but degrade on noisy real-world sensor data
    Invoked in the abstract to motivate why pure KAN replacements underperform and why a hybrid is needed.

pith-pipeline@v0.9.0 · 5815 in / 1348 out tokens · 48325 ms · 2026-05-20T10:32:12.380741+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

60 extracted references · 60 canonical work pages · 1 internal anchor

  1. [1]

    Sidra Abbas, Shtwai Alsubai, Muhammad Ibrar Ul Haque, Gabriel Avelino Sampedro, Ahmad Almadhor, Abdullah Al Hejaili, and Iryna Ivanochko

    work page

  2. [2]

    Active machine learning for heterogeneity activity recognition through smartwatch sensors.IEEE Access12 (2024), 22595–22607

    work page 2024

  3. [3]

    Reem Abdel-Salam, Rana Mostafa, and Mayada Hadhood. 2021. Human activity recognition using wearable sensors: review, challenges, evaluation benchmark. InInternational workshop on deep learning for human activity recognition. Springer, 1–15

    work page 2021

  4. [4]

    Mohammed Ghaith Altarabichi. 2024. Dropkan: Regularizing kans by masking post-activations.arXiv preprint arXiv:2407.13044(2024)

    work page arXiv 2024

  5. [5]

    Vladimir I. Arnold. 1963. On Functions of Three Variables.Doklady Akademii Nauk SSSR148 (1963), 9–12

    work page 1963

  6. [6]

    Vladimir I Arnold. 2009. On functions of three variables.Collected Works: Representations of Functions, Celestial Mechanics and KAM Theory, 1957–1965(2009), 5–8

    work page 2009

  7. [7]

    Marc Bachlin, Meir Plotnik, Daniel Roggen, Inbal Maidan, Jeffrey M Hausdorff, Nir Giladi, and Gerhard Troster. 2009. Wearable assistant for Parkinson’s disease patients with the freezing of gait symptom.IEEE Transactions on Information Technology in Biomedicine14, 2 (2009), 436–446

    work page 2009

  8. [8]

    Oresti Banos, Rafael Garcia, Juan A Holgado-Terriza, Miguel Damas, Hector Pomares, Ignacio Rojas, Alejandro Saez, and Claudia Villalonga. 2014. mHealthDroid: A novel framework for agile development of mobile health applications.Ambient Assisted Living and Daily Activities8868, 14 (2014), 91–98

    work page 2014

  9. [9]

    Sizhen Bian, Mengxi Liu, Bo Zhou, and Paul Lukowicz. 2022. The state-of-the-art sensing techniques in human activity recognition: A survey. Sensors22, 12 (2022), 4596

    work page 2022

  10. [10]

    Alexander Dylan Bodner, Antonio Santiago Tepsich, Jack Natan Spolski, and Santiago Pourteau. 2024. Convolutional kolmogorov-arnold networks. arXiv preprint arXiv:2406.13155(2024)

    work page arXiv 2024

  11. [11]

    Zavareh Bozorgasl and Hao Chen. [n. d.]. Wav-kan: Wavelet kolmogorov-arnold networks, 2024.arXiv preprint arXiv:2405.12832([n. d.])

    work page arXiv 2024

  12. [12]

    Yueyang Cang, Li Shi, et al. 2024. Can KAN Work? Exploring the Potential of Kolmogorov-Arnold Networks in Computer Vision.arXiv preprint arXiv:2411.06727(2024)

    work page arXiv 2024

  13. [13]

    Blealtan Cao. 2024. An Efficient Implementation of Kolmogorov-Arnold Network. https://github.com/Blealtan/efficient-kan. Accessed: 2025-04-10

    work page 2024

  14. [14]

    Ricardo Chavarriaga, Hesam Sagha, Alberto Calatroni, Sundara Tejaswi Digumarti, Gerhard Tröster, José del R Millán, and Daniel Roggen. 2013. The Opportunity challenge: A benchmark database for on-body sensor-based activity recognition.Pattern Recognition Letters34, 15 (2013), 2033–2042

    work page 2013

  15. [15]

    Kaixuan Chen, Dalin Zhang, Lina Yao, Bin Guo, Zhiwen Yu, and Yunhao Liu. 2021. Deep learning for sensor-based human activity recognition: Overview, challenges, and opportunities.ACM Computing Surveys (CSUR)54, 4 (2021), 1–40

    work page 2021

  16. [16]

    Zhijie Chen and Xinglin Zhang. 2024. Larctan-skan: Simple and efficient single-parameterized kolmogorov-arnold networks using learnable trigonometric function.arXiv preprint arXiv:2410.19360(2024)

    work page arXiv 2024

  17. [17]

    Zhijie Chen and Xinglin Zhang. 2024. Lss-skan: Efficient kolmogorov-arnold networks based on single-parameterized function.arXiv preprint arXiv:2410.14951(2024)

    work page arXiv 2024

  18. [18]

    Iveta Dirgová Luptáková, Martin Kubovčík, and Jiří Pospíchal. 2022. Wearable sensor-based human activity recognition with transformer model. Sensors22, 5 (2022), 1911

    work page 2022

  19. [19]

    Ivan Drokin. 2024. Kolmogorov-arnold convolutions: Design principles and empirical studies.arXiv preprint arXiv:2407.01092(2024)

    work page arXiv 2024

  20. [20]

    Yu Enokibori. 2024. rTsfNet: a DNN model with Multi-head 3D Rotation and Time Series Feature Extraction for IMU-based Human Activity Recognition.Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies8, 4 (2024), 1–26

    work page 2024

  21. [21]

    Changjun Fan and Fei Gao. 2021. Enhanced human activity recognition using wearable sensors via a hybrid feature selection method.Sensors21, 19 (2021), 6434

    work page 2021

  22. [22]

    Remi Genet and Hugo Inzirillo. 2024. Tkan: Temporal kolmogorov-arnold networks.arXiv preprint arXiv:2405.07344(2024)

    work page arXiv 2024

  23. [23]

    Ahmed Dawod Mohammed Ibrahum, Zhengyu Shang, and Jang-Eui Hong. 2024. How Resilient Are Kolmogorov–Arnold Networks in Classification Tasks? A Robustness Investigation.Applied Sciences14, 22 (2024), 10173

    work page 2024

  24. [24]

    Petr Ivashkov, Po-Wei Huang, Kelvin Koor, Lirandë Pira, and Patrick Rebentrost. 2026. QKAN: quantum Kolmogorov-Arnold networks with applications in machine learning and multivariate state preparation.npj Quantum Information12, 1 (11 Mar 2026), 73. doi:10.1038/s41534-026-01202-5

    work page doi:10.1038/s41534-026-01202-5 2026

  25. [25]

    Ali Jamali, Swalpa Kumar Roy, Danfeng Hong, Bing Lu, and Pedram Ghamisi. 2024. How to learn more? Exploring Kolmogorov–Arnold networks for hyperspectral image classification.Remote Sensing16, 21 (2024), 4015

    work page 2024

  26. [26]

    Zanobya N Khan and Jamil Ahmad. 2021. Attention induced multi-head convolutional neural network for human activity recognition.Applied soft computing110 (2021), 107671

    work page 2021

  27. [27]

    Benjamin C Koenig, Suyong Kim, and Sili Deng. 2024. KAN-ODEs: Kolmogorov–Arnold network ordinary differential equations for learning dynamical systems and hidden physics.Computer Methods in Applied Mechanics and Engineering432 (2024), 117397

    work page 2024

  28. [28]

    Andrei Nikolaevich Kolmogorov. 1957. On the representations of continuous functions of many variables by superposition of continuous functions of one variable and addition. InDokl. Akad. Nauk USSR, Vol. 114. 953–956. Manuscript submitted to ACM KAN-MLP-Mixer 23

    work page 1957

  29. [29]

    Tran Xuan Hieu Le, Thi Diem Tran, Hoai Luan Pham, Vu Trung Duong Le, Tuan Hai Vu, Van Tinh Nguyen, Yasuhiko Nakashima, et al . 2024. Exploring the limitations of kolmogorov-arnold networks in classification: Insights to software training and hardware implementation. In2024 Twelfth International Symposium on Computing and Networking Workshops (CANDARW). IE...

    work page 2024

  30. [30]

    Ziyao Li. 2024. Kolmogorov-Arnold Networks are Radial Basis Function Networks. (2024). arXiv:2405.06721 [cs.LG]

    work page arXiv 2024

  31. [31]

    Hanxiao Liu, Zihang Dai, David So, and Quoc V Le. 2021. Pay attention to mlps.Advances in neural information processing systems34 (2021), 9204–9215

    work page 2021

  32. [32]

    Mengxi Liu, Daniel Geißler, Dominique Nshimyimana, Sizhen Bian, Bo Zhou, and Paul Lukowicz. 2024. Initial investigation of kolmogorov-arnold networks (kans) as feature extractors for imu based human activity recognition. InCompanion of the 2024 on ACM International Joint Conference on Pervasive and Ubiquitous Computing. 500–506

    work page 2024

  33. [33]

    Ziming Liu, Pingchuan Ma, Yixuan Wang, Wojciech Matusik, and Max Tegmark. 2024. Kan 2.0: Kolmogorov-arnold networks meet science.arXiv preprint arXiv:2408.10205(2024)

    work page arXiv 2024

  34. [34]

    Ziming Liu, Yixuan Wang, Sachin Vaidya, Fabian Ruehle, James Halverson, Marin Soljačić, Thomas Y Hou, and Max Tegmark. 2024. Kan: Kolmogorov-arnold networks.arXiv preprint arXiv:2404.19756(2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  35. [35]

    Mohammad Malekzadeh, Richard Clegg, Andrea Cavallaro, and Hamed Haddadi. 2021. Dana: Dimension-adaptive neural architecture for multivariate sensor data.Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies5, 3 (2021), 1–27

    work page 2021

  36. [36]

    Takeru Miyoshi, Makoto Koshino, and Hidetaka Nambo. 2025. Applying MLP-Mixer and gMLP to Human Activity Recognition.Sensors25, 2 (2025), 311

    work page 2025

  37. [37]

    Kamsiriochukwu Ojiako and Katayoun Farrahi. 2023. MLPs Are All You Need for Human Activity Recognition.Applied Sciences13, 20 (2023), 11154

    work page 2023

  38. [38]

    Ordóñez and Daniel Roggen

    Francisco J. Ordóñez and Daniel Roggen. 2016. Deep Convolutional and LSTM Recurrent Neural Networks for Multimodal Wearable Activity Recognition.Sensors16, 1 (2016), 115

    work page 2016

  39. [39]

    Allan Pinkus. 1999. Approximation theory of the MLP model in neural networks.Acta numerica8 (1999), 143–195

    work page 1999

  40. [40]

    Eleonora Poeta, Flavio Giobergia, Eliana Pastor, Tania Cerquitelli, and Elena Baralis. 2024. A benchmarking study of kolmogorov-arnold networks on tabular data. In2024 IEEE 18th International Conference on Application of Information and Communication Technologies (AICT). IEEE, 1–6

    work page 2024

  41. [41]

    Attila Reiss and Didier Stricker. 2012. Introducing a new benchmarked dataset for activity monitoring. In2012 16th international symposium on wearable computers. IEEE, 108–109

    work page 2012

  42. [42]

    Anguita, Alessandro Ghio, L

    Jorge L. Reyes-Ortiz, Davide Anguita, Alessandro Ghio, Luca Oneto, and Xavier Parra. 2013.Human Activity Recognition Using Smartphones. doi:10.24432/C54S4K

    work page doi:10.24432/c54s4k 2013

  43. [43]

    Yoli Shavit and Itzik Klein. 2021. Boosting inertial-based human activity recognition with transformers.IEEE Access9 (2021), 53540–53547

    work page 2021

  44. [44]

    Haoran Shen, Chen Zeng, Jiahui Wang, and Qiao Wang. 2025. Reduced effectiveness of kolmogorov-arnold networks on functions with noise. In ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 1–5

    work page 2025

  45. [45]

    Shriyank Somvanshi, Syed Aaqib Javed, Md Monzurul Islam, Diwas Pandit, and Subasish Das. 2024. A survey on kolmogorov-arnold network.arXiv preprint arXiv:2411.06078(2024)

    work page arXiv 2024

  46. [46]

    Yueyuan Sui, Minghui Zhao, Junxi Xia, Xiaofan Jiang, and Stephen Xia. 2024. Tramba: A hybrid transformer and mamba architecture for practical audio and bone conduction speech super resolution and enhancement on mobile and wearable platforms.Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies8, 4 (2024), 1–29

    work page 2024

  47. [47]

    Ilya O Tolstikhin, Neil Houlsby, Alexander Kolesnikov, Lucas Beyer, Xiaohua Zhai, Thomas Unterthiner, Jessica Yung, Andreas Steiner, Daniel Keysers, Jakob Uszkoreit, et al. 2021. Mlp-mixer: An all-mlp architecture for vision.Advances in neural information processing systems34 (2021), 24261–24272

    work page 2021

  48. [48]

    Juan Diego Toscano, Vivek Oommen, Alan John Varghese, Zongren Zou, Nazanin Ahmadi Daryakenari, Chenxi Wu, and George Em Karniadakis

    work page

  49. [49]

    From pinns to pikans: Recent advances in physics-informed machine learning.Machine Learning for Computational Science and Engineering1, 1 (2025), 1–43

    work page 2025

  50. [50]

    Yu-Hsuan Tseng and Chih-Yu Wen. 2023. Hybrid Learning Models for IMU-Based HAR with Feature Analysis and Data Correction.Sensors23, 18 (2023), 7802

    work page 2023

  51. [51]

    Yannick Werner, Akash Malemath, Mengxi Liu, Vitor Fortes Rey, Nikolaos Palaiodimopoulos, Paul Lukowicz, and Maximilian Kiefer-Emmanouilidis

    work page

  52. [52]

    doi:10.1038/s41598-025-22705-9

    QuKAN: A Quantum Circuit Born Machine Approach to Quantum Kolmogorov Arnold Networks.Scientific Reports15, 1 (09 Oct 2025), 35239. doi:10.1038/s41598-025-22705-9

    work page doi:10.1038/s41598-025-22705-9 2025

  53. [53]

    Jinfeng Xu, Zheyu Chen, Jinze Li, Shuo Yang, Wei Wang, Xiping Hu, and Edith C-H Ngai. 2024. FourierKAN-GCF: Fourier Kolmogorov-Arnold Network–An Effective and Efficient Feature Transformation for Graph Collaborative Filtering.arXiv preprint arXiv:2406.01034(2024)

    work page arXiv 2024

  54. [54]

    Xingyi Yang and Xinchao Wang. 2024. Kolmogorov-arnold transformer.arXiv preprint arXiv:2409.10594(2024)

    work page arXiv 2024

  55. [55]

    Yafeng Yin, Lei Xie, Zhiwei Jiang, Fu Xiao, Jiannong Cao, and Sanglu Lu. 2024. A systematic review of human activity recognition based on mobile devices: overview, progress and trends.IEEE Communications Surveys & Tutorials26, 2 (2024), 890–929

    work page 2024

  56. [56]

    Piero Zappi, Clemens Lombriser, Thomas Stiefmeier, Elisabetta Farella, Daniel Roggen, Luca Benini, and Gerhard Tröster. 2008. Activity recognition from on-body sensors: accuracy-power trade-off by dynamic sensor selection. InWireless Sensor Networks: 5th European Conference, EWSN 2008, Bologna, Italy, January 30-February 1, 2008. Proceedings. Springer, 17–33

    work page 2008

  57. [57]

    Licheng Zhang, Xihong Wu, and Dingsheng Luo. 2015. Recognizing human activities from raw accelerometer data using deep neural networks. In 2015 IEEE 14th International conference on machine learning and applications (ICMLA). IEEE, 865–870. Manuscript submitted to ACM 24 Liu et al

    work page 2015

  58. [58]

    Ye Zhang, Longguang Wang, Huiling Chen, Aosheng Tian, Shilin Zhou, and Yulan Guo. 2022. IF-ConvTransformer: A framework for human activity recognition using IMU fusion and ConvTransformer.Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies6, 2 (2022), 1–26

    work page 2022

  59. [59]

    Yexu Zhou, Tobias King, Haibin Zhao, Yiran Huang, Till Riedel, and Michael Beigl. 2024. MLP-HAR: Boosting Performance and Efficiency of HAR Models on Edge Devices with Purely Fully Connected Layers. InProceedings of the 2024 ACM International Symposium on Wearable Computers. 133–139

    work page 2024

  60. [60]

    Yexu Zhou, Haibin Zhao, Yiran Huang, Till Riedel, Michael Hefenbrock, and Michael Beigl. 2022. Tinyhar: A lightweight deep learning model designed for human activity recognition. InProceedings of the 2022 ACM International Symposium on Wearable Computers. 89–93. Received 20 February 2007; revised 12 March 2009; accepted 5 June 2009 Manuscript submitted to ACM

    work page 2022