Kolmogorov-Arnold Networks are Radial Basis Function Networks

cs.LG 2026-07 conditional novelty 7.0 of 10

In small language models, KAN feed-forward blocks are auditable and pruneable, but on standardized benchmarks and scale tests they show no consistent accuracy, quality, or latency advantage over MLP baselines.

cs.CE 2025-07 conditional novelty 7.0 of 10

A boundary-only holomorphic KAN framework solves Laplace, Helmholtz, and elasticity problems on simply and multiply connected 2D domains, outperforming standard PINNs in the tested cases.

gr-qc 2026-07 accept novelty 6.0 of 10

A piecewise MLP surrogate emulates NRSur7dq4 over its full domain at NR-faithful accuracy with ~1 ms GPU latency and a fully differentiable JAX likelihood pipeline.

cs.LG 2025-06 conditional novelty 6.0 of 10

MetaKANs generates each KAN activation function from a shared prompt-conditioned meta-learner, cutting trainable parameters toward MLP level while retaining comparable or better accuracy on tested benchmarks.

cs.LG 2025-04 conditional novelty 6.0 of 10

ChemKANs, a physics-structured KAN-ODE network, infer chemical kinetic models from noisy data and accelerate hydrogen-air combustion chemistry with 344 parameters at 2x speedup.

cs.LG 2025-01 conditional novelty 6.0 of 10

PRKAN lowers KAN parameter counts to near-MLP levels via attention, convolution/pooling, dimension summation, and feature-vector projections, reaching MLP-like accuracy on MNIST and Fashion-MNIST.

cs.LG 2024-12 conditional novelty 6.0 of 10

KKANs, a two-block KART-based architecture with MLP inner functions and basis-function outer functions, universally approximate continuous functions and empirically outperform MLP and cKAN baselines in regression, PIN...

stat.ML 2025-08 reject novelty 5.0 of 10

The paper states universal approximation theorems for sine-based Kolmogorov-Arnold networks, but the proof relies on an invalid lemma and an unproven matrix invertibility.

physics.optics 2025-07 conditional novelty 5.0 of 10

A physics-informed PointNet that encodes spatially varying permittivity predicts near-field and far-field scattering from metasurfaces with inclined nanopillars, reaching 1.7% MAPE for low-contrast 2D cases.

physics.comp-ph 2025-07 conditional novelty 5.0 of 10

MR-PIKAN, a multi-resolution training schedule that alternates coarse and fine collocation grids, cuts training time while keeping accuracy on multi-scale forward and inverse PDE problems.

cs.LG 2025-06 conditional novelty 5.0 of 10

The first NTK analysis of cPIKANs finds their kernel spectra stay stable during training, correlating with large accuracy gains over PINNs, especially when time is split into subdomains.

cs.CL 2025-05 conditional novelty 5.0 of 10

Placing a KAN layer between two linear layers improves spoken language understanding accuracy over linear-only baselines on several speech-intent datasets.

cs.CV 2025-04 conditional novelty 5.0 of 10

The proposed PICO/LeAFNet pipeline compresses point clouds by fitting two small neural networks with learnable activations, reporting 4.92 dB average BD-PSNR gain over MPEG codecs on 8iVFB.

cs.LG 2025-02 conditional novelty 5.0 of 10

MatrixKAN replaces KAN's recursive B-spline evaluation with precomputed matrix multiplications, making training time nearly independent of spline degree.

cs.LG 2025-02 conditional novelty 5.0 of 10

A frequency-decomposing KAN architecture achieves state-of-the-art or near-state-of-the-art long-term forecasting on five of six datasets with 12-38K parameters.

cs.LG 2025-01 conditional novelty 5.0 of 10

Swapping attentive GNN score mappings for a single-layer Kolmogorov-Arnold Network improves benchmark performance and, on a specially constructed input matrix, provably achieves zero maximum ranking error.

q-bio.SC 2025-01 conditional novelty 5.0 of 10

AEMS-Net, a U-Net variant with KAN convolutions, attention, and brightness adaptation, reconstructs mitochondrial and microtubule images from one fluorescence image and reports large gains over vanilla U-Net.

cs.NE 2024-11 conditional novelty 5.0 of 10

Comparing fixed and learnable activations in PINNs across five PDEs shows learnable bases help in small networks, destabilize large ones, and low spectral bias does not guarantee accuracy.

cs.LG 2026-07 reject novelty 4.0 of 10

PG-KINN pairs a KAN trial space with a Petrov–Galerkin test space for forward and inverse PDEs, but the inverse benchmark data is inconsistent with the governing equation and the accuracy claims are not supported by t...

cs.LG 2026-07 conditional novelty 4.0 of 10

STKAN inserts Taylor-polynomial KAN token mixers into spatial and temporal mixing blocks and achieves small but consistent gains over strong baselines on three traffic-flow benchmarks and a tie on a fourth.

cs.LG 2026-06 conditional novelty 4.0 of 10

SechKAN combines sech basis functions with a 1D linear projection to build a KAN-style model whose parameter count matches MLPs and which is competitive or better than several KAN variants on tested benchmarks.

cs.LG 2025-06 conditional novelty 4.0 of 10

Augmenting Kolmogorov-Arnold Networks with prediction exits at each layer improves accuracy and often yields more parsimonious models, and a differentiable learning-to-exit algorithm automates the choice of exit weights.

eess.IV 2025-05 conditional novelty 4.0 of 10

A Taylor-expansion KAN variant outperforms B-spline, orthogonal-polynomial, wavelet, and Fourier KAN variants and MLPs on five authentic BIQA databases, with PCA and shallow layers reducing training cost.

cs.LG 2025-04 conditional novelty 4.0 of 10

DAE-KAN, a dual-network KAN-PINN, is reported to solve index-1 through index-3 differential-algebraic equations with one to two orders of magnitude lower error than MLP-based PINNs on two benchmark problems.

physics.geo-ph 2025-02 conditional novelty 4.0 of 10

EFKAN, an FNO-branch with KAN-trunk neural operator, gives mixed but generally lower average L1 error than EFNO on synthetic 2D magnetotelluric forward modeling and is much faster than FDM.

math.NA 2025-01 conditional novelty 4.0 of 10

Rescaling PDE spatial variables to [-1,1] before training Chebyshev-based physics-informed Kolmogorov-Arnold networks improves accuracy and convergence on wide oscillatory domains.

cs.LG 2024-12 reject novelty 4.0 of 10

PowerMLP is a ReLU-power MLP that trains about 40x faster than KAN in the reported benchmarks and often beats it, but the main proof that KANs are contained in PowerMLPs at the same depth is flawed.