DeepOKAN: Deep operator network based on Kolmogorov Arnold networks for mechanics problems

Computer Methods in Applied Mechanics and Engineering, Год журнала: 2025, Номер 436, С. 117699 - 117699

Опубликована: Янв. 14, 2025

Язык: Английский

---

From PINNs to PIKANs: recent advances in physics-informed machine learning

Machine learning for computational science and engineering, Год журнала: 2025, Номер 1(1)

Опубликована: Март 11, 2025

Язык: Английский

---

Multispectral non-line-of-sight imaging via deep fusion photography

Science China Information Sciences, Год журнала: 2025, Номер 68(4)

Опубликована: Март 4, 2025

Язык: Английский

---

Comparison of Kolmogorov–Arnold Network and Multi-Layer Perceptron models for modelling and optimisation analysis of energy systems

Energy and AI, Год журнала: 2025, Номер unknown, С. 100473 - 100473

Опубликована: Янв. 1, 2025

Язык: Английский

---

Kolmogorov–Arnold PointNet: Deep learning for prediction of fluid fields on irregular geometries

Computer Methods in Applied Mechanics and Engineering, Год журнала: 2025, Номер 439, С. 117888 - 117888

Опубликована: Март 11, 2025

Язык: Английский

---

SineKAN: Kolmogorov-Arnold Networks using sinusoidal activation functions

Frontiers in Artificial Intelligence, Год журнала: 2025, Номер 7

Опубликована: Янв. 15, 2025

Recent work has established an alternative to traditional multi-layer perceptron neural networks in the form of Kolmogorov-Arnold Networks (KAN). The general KAN framework uses learnable activation functions on edges computational graph followed by summation nodes. edge original implementation are basis spline (B-Spline). Here, we present a model which grids B-Spline replaced re-weighted sine (SineKAN). We evaluate numerical performance our benchmark vision task. show that can perform better than or comparable models and based periodic cosine representing Fourier Series. Further, SineKAN accuracy could scale comparably dense (DNNs). Compared two baseline models, achieves substantial speed increase at all hidden layer sizes, batch depths. Current advantage DNNs due hardware software optimizations discussed along with theoretical scaling. Additionally, properties compared other implementations current limitations also discussed.

Язык: Английский

---

jaxKAN: A unified JAX framework for Kolmogorov-Arnold Networks

The Journal of Open Source Software, Год журнала: 2025, Номер 10(108), С. 7830 - 7830

Опубликована: Апрель 7, 2025

Язык: Английский

---

Explainable Fault Classification and Severity Diagnosis in Rotating Machinery Using Kolmogorov–Arnold Networks

Entropy, Год журнала: 2025, Номер 27(4), С. 403 - 403

Опубликована: Апрель 9, 2025

Rolling element bearings are critical components of rotating machinery, with their performance directly influencing the efficiency and reliability industrial systems. At same time, bearing faults a leading cause machinery failures, often resulting in costly downtime, reduced productivity, and, extreme cases, catastrophic damage. This study presents methodology that utilizes Kolmogorov–Arnold Networks—a recent deep learning alternative to Multilayer Perceptrons. The proposed method automatically selects most relevant features from sensor data searches for optimal hyper-parameters within single unified approach. By using shallow network architectures fewer features, models lightweight, easily interpretable, practical real-time applications. Validated on two widely recognized datasets fault diagnosis, framework achieved perfect F1-Scores detection high severity classification tasks, including 100% cases. Notably, it demonstrated adaptability by handling diverse types, such as imbalance misalignment, dataset. availability symbolic representations provided model interpretability, while feature attribution offered insights into types or signals each studied task. These results highlight framework’s potential applications, monitoring, scientific research requiring efficient explainable models.

Язык: Английский

---

Inferring in vivo murine cerebrospinal fluid flow using artificial intelligence velocimetry with moving boundaries and uncertainty quantification

Interface Focus, Год журнала: 2024, Номер 14(6)

Опубликована: Дек. 6, 2024

Cerebrospinal fluid (CSF) flow is crucial for clearing metabolic waste from the brain, a process whose dysregulation linked to neurodegenerative diseases like Alzheimer’s. Traditional approaches particle tracking velocimetry (PTV) are limited by their reliance on single-plane two-dimensional measurements, which fail capture complex dynamics of CSF fully. To overcome these limitations, we employ artificial intelligence (AIV) reconstruct three-dimensional velocities, infer pressure and wall shear stress quantify rates. Given experimental nature data inherent variability in biological systems, robust uncertainty quantification (UQ) essential. Towards this end, have modified baseline AIV architecture address aleatoric caused noisy data, enhancing our measurement refinement capabilities. We also implement UQ model epistemic uncertainties arising governing equations network representation. test multiple laws, representation models initializations. Our approach not only advances accuracy but can be adapted other applications that use physics-informed machine learning fields providing versatile tool inverse problems.

Язык: Английский

---

Automatic Discovery of Optimal Meta-Solvers Via Multi-Objective Optimization

Опубликована: Янв. 1, 2025

Язык: Английский

---