Multilevel Training for Kolmogorov Arnold Networks

This paper introduces a multilevel training framework for Kolmogorov-Arnold Networks (KANs) that leverages their structural equivalence to multichannel MLPs and the properties of spline basis functions to create a properly nested hierarchy of models, resulting in orders-of-magnitude improvements in training accuracy and speed, particularly for physics-informed neural networks.

Ben S. Southworth, Jonas A. Actor, Graham Harper + 1 more2026-03-06🔢 math

BBP Phase Transition for a Doubly Sparse Deformed Model

This paper establishes a BBP phase transition for a doubly sparse deformed Wigner ensemble, proving that sparse signal vectors with strength greater than one generate outlier eigenvalues and correlate with top eigenvectors in the supercritical sparsity regime, thereby generalizing previous results to models where both the noise matrix and signal vectors are sparse without requiring specific relationships between their sparsity levels.

Ioana Dumitriu, JD Flynn, Zhichao Wang2026-03-06🔢 math

Stability conditions on noncommutative crepant resolutions of 3-dimensional isolated singularities

This paper constructs a mutation cone and a corresponding wall-and-chamber structure for maximal modifying modules over 3-dimensional Gorenstein isolated singularities, proving that the tilting-noetherian property holds if and only if all such modules are mutation-connected, and establishing a regular covering map from a specific subspace of Bridgeland stability conditions to the complexified mutation cone to describe the associated autoequivalence group.

Wahei Hara, Yuki Hirano2026-03-06🔢 math

Construction of higher Chow cycles on cyclic coverings of P1×P1\mathbb{P}^1 \times \mathbb{P}^1, Part II

This paper constructs higher Chow cycles of type (2,1)(2,1) on a family of degree NN abelian covers of P1\mathbb{P}^1 branched over n+2n+2 points and proves that for a very general member, these cycles generate a subgroup of the indecomposable part of rank at least nϕ(N)n\cdot \phi(N) by computing their images under the transcendental regulator map.

Yusuke Nemoto, Ken Sato2026-03-06🔢 math

How Does the ReLU Activation Affect the Implicit Bias of Gradient Descent on High-dimensional Neural Network Regression?

This paper demonstrates that for high-dimensional random data, gradient descent on shallow ReLU networks exhibits an implicit bias that approximates the minimum L2L_2-norm solution with high probability, bridging the gap between worst-case non-existence and exact orthogonality results through a novel primal-dual analysis.

Kuo-Wei Lai, Guanghui Wang, Molei Tao + 1 more2026-03-06🔢 math