Get Started
Home
Topics
Search
Library
Research questionHow can Gaussian kernel bandwidths adapt to local manifold dimension without losing geometric discrimination?A bandwidth that is too small can make each sample appear to define a separate direction, while one that is too large can merge distinct geometric directions. The useful scale therefore depends on the local complexity of the data manifold.
AI
Machine Learning
Research Paper
Statistical Machine Learning
Technology
Latest papersRecent research connected to this question, newest first.Geometry-Aware Graph Construction via Adaptive Spectral Bandwidth ControlThe problem concerns Gaussian-kernel graph methods applied to self-supervised embeddings. Evidence in the source comes from leave-one-out classification and label propagation on CIFAR-100 embeddings produced by six encoders, with comparisons against fixed-bandwidth and other adaptive approaches.research paper · Sep 3, 2026
Related questions
What are the fundamental sample-size limits for estimating diffusion-based local intrinsic dimension at finite smoothing scales?How can agnostic learning handle low-intrinsic-dimensional concepts under arbitrary distributions with Gaussian-robust benchmarks?How can GPUs efficiently compute unnormalized Gaussian kernel sums with signed weights without custom kernels?How many linear samples are needed to approximate Lipschitz operators under Gaussian measures, and how fast can errors decay?