Get Started
Home
Topics
Search
Library
Research questionHow can finite compact datasets be transformed to enable linear separation with narrow neural networks?Compact class sets can have geometries that prevent linear separation in their original coordinates. A central challenge is to modify or augment that geometry without requiring a wide network.
AI
Machine Learning
Neural and Evolutionary Computing
Research Paper
Latest papersRecent research connected to this question, newest first.Relocation of compact sets in $\mathbb{R}^n$ by diffeomorphisms and linear separability of datasets in $\mathbb{R}^n$The source gives theoretical results using self-diffeomorphisms in the original space and differentiable embeddings into one higher dimension, with applications to networks using Leaky-ReLU, ELU, or SELU activations. The stated width guarantees depend on a mild condition in one case and on mutual disjointness in another; the source does not specify a training procedure.research paper · Sep 4, 2026
Related questions
How can SPD-manifold representations capture intrinsic relational geometry while preserving valid matrix constraints?How can fixed-dimensional subspaces of one-dimensional convolutional filters be parametrized injectively as smooth closed projective varieties?How can neural signed distance fields yield exact zero-level-set meshes without losing high-frequency geometry?How can agnostic learning handle low-intrinsic-dimensional concepts under arbitrary distributions with Gaussian-robust benchmarks?