Researchers introduced Finite-Library Input-Warped Bayesian Optimization (FLIWBO), a method that adapts input geometry in Gaussian-process Bayesian optimization while retaining convergence guarantees,…
A new study from arXiv (arXiv:2608.14594v1) finds that trajectory-level diagnostics from Projected Gradient Descent (PGD) attacks, such as loss evolution and gradient alignment, do not independently m…
A researcher's independent control experiment on the PIN v5 dataset found that the feature-level jurisdiction signal survives a held-out unseen class identity in Fashion-MNIST, with direct jurisdictio…
Researchers have introduced SechKAN, a Kolmogorov-Arnold Network architecture using hyperbolic secant functions, which achieves superior performance over MLPs and other KAN variants on benchmark datas…
A developer has built a Fashion-MNIST neural network trainer in pure SQL using xarray-sql, demonstrating that matrix multiplication and backpropagation can be expressed as joins and aggregations. The …
A developer implemented a neural network in SQL using the xarray_sql library, achieving a speedup of approximately 1.8x on Fashion-MNIST by skipping zero-valued pixels during layer-0 contraction. The …
Researchers propose S-GAI, a spectral geometry-aware initialization framework for sigmoidal MLPs that encodes dataset geometry into network weights. Using SVD to extract class-wise spectral features, …
Researchers introduced Pre-Warm, a zero-training-cost method for data-conditioned initialization of the first convolutional layer in CNNs. The technique clusters patches from a single training batch t…